Moods, Bots, and Bodies: University Students’ Emotional and Physiological responses to Human vs. GenAI Chatbots

Moods, Bots, and Bodies: University Students’ Emotional and Physiological responses to Human vs. GenAI Chatbots | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Moods, Bots, and Bodies: University Students’ Emotional and Physiological responses to Human vs. GenAI Chatbots Jia'en Yee, Fei Victor Lim, Jerrold Quek This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7785914/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract As generative AI (Gen AI) chatbots become more common as learning partners, questions remain about students’ emotional and physiological responses to them. This study used a multimodal design to compare university students’ experiences during a 25‑minute brainstorming session with either a human teacher or Gen AI chatbot. Thirty participants wore EmbracePlus sensors to record heart rate, electrodermal activity (EDA), and skin temperature while completing the task, and completed mood questionnaires before and after brainstorming. Analyses compared mood change scores (controlling for age and gender) and examined physiological data for both temporal patterns and total activation (area-under-the-curve; AUC). While both groups reported improved mood, students brainstorming with a human teacher showed greater gains in positive mood, whereas the chatbot group reported increased stress and discouragement, and exhibited higher cumulative cardiovascular activation. Although physiological change trajectories did not differ by condition, specific AUC measures were associated with mood: higher pulse AUC was linked to negative moods, and higher skin temperature AUC to positive moods. These findings suggest that while human facilitation produces stronger emotional benefits, GenAI chatbots can sustain comparable physiological engagement and serve as valuable complementary tools. Physiological signals also reveal distinctive patterns between bodily states and learning experiences, underscoring the value of integrating multimodal data into research on AI‑mediated education. Educational Psychology Artificial Intelligence and Machine Learning Psychology customised generative AI chatbots multimodal learning analytics emotional responses physiological responses embodied learning Figures Figure 1 Figure 2 Figure 3 1. Introduction As the education landscape rapidly evolves with the integration of generative AI (Gen AI), opportunities and challenges abound, particularly in its potential to transform traditional teaching methods. There is now a sharp rise in AI-based education applications; for example, educational games, adaptive learning platforms, AI chatbots serving as virtual teaching assistants, personalised tutors, and interactive learning companions across diverse education levels all over the world. AI systems can personalise learning content and experience to match one’s pace, needs and abilities regardless of time or geography, and provide feedback (e.g., Labadze, Grigolia, & Machaidze, 2023 ), making it increasingly accessible and personalised. Emerging research underscores the potential of Gen AI tools in language and literacy education. Chandel and Lim ( 2025 ), in a systematic review of empirical studies, examine how GenAI applications such as AI-powered writing assistants, and conversational chatbots are being integrated into classroom practice to support literacy development across reading, writing, and multimodal meaning-making. The review highlights the affordances of GenAI in facilitating ideation and planning, with studies by Lee et al. ( 2024 ), Mahapatra ( 2024 ), and Xiao and Zhi ( 2023 ) demonstrating the benefits of such tools in supporting learners during the brainstorming phase of writing. However, there is limited literature on students’ receptivity to using Gen AI for learning or help-seeking compared to human educators, especially when explored through experimental methods rather than interviews. While AI offers a multitude of advantages, aspects unique to Gen AI can potentially impede learning. For example, in a study consisting of students learning a second language through virtual human interactions, students’ levels of frustration were modulated by factors such as “not being understood or heard as expected” (Ericsson et al., 2023 ). This is a common trend when engaging with AI where nuanced human dynamics are missing. Other features of GenAI, such as perceived warmth and emotional responsiveness, have been shown to affect individuals’ attitudes towards having AI as a teammate (Harris-Watson et al., 2023 ). Specifically, perceived warmth and competence positively predict individuals’ receptivity to AI in three aspects, including their willingness to adapt routines, integrate AI's expertise and skills and regard AI as a valued teammate. Even though individuals may appreciate the efficiency or flexibility of AI systems, the absence of “human” qualities like empathy, adaptability, encouragement, personal connection and trustworthiness in AI bots can affect their satisfaction levels and how willing they are to engage and rely on them for knowledge building (e.g., Cevher & Yıldırım, 2023 ; Demeure et al., 2022). Given the established link between emotional arousal, attention and learning effectiveness (Loderer et al., 2020 ), along with the increasing integration of AI in education, it is crucial to deepen our understanding of how students perceive and interact with Gen AI compared to human educators. Such insights can inform the development of blended learning environments that optimise the strengths of both AI and human educators, ensuring that AI serves as a pedagogical aide for the teacher. This proof-of-concept study investigates how students’ emotional and physiological responses differ when brainstorming with a Gen AI chatbot versus a human teacher. We focus on self-rated mood and physiological arousal as indicators of interest and engagement, aiming to identify the potential reasons for receptivity or resistance towards the use of Gen AI for learning, and how each can be leveraged to support learning. The study addresses three questions: 1) How do students’ emotional responses differ between chatbot- and human-facilitated brainstorming? 2) How do students’ physiological responses differ between chatbot- and human-facilitated brainstorming? 3) How are students’ physiological signals related to their self-reported moods, regardless of condition? 2. Literature Review 2.1. Brainstorming and facilitation for writing Brainstorming is widely recognised as an effective pre-writing strategy that enhances idea generation, organisation, and refinement, leading to higher-quality essays with greater originality and depth (Crossley, Muldner, & McNamara, 2016 ). In externalising their thoughts during brainstorming, writers engage in metacognitive reflection, structuring ideas more clearly, drawing on prior knowledge, considering various perspectives, and approaching writing more fluently (Kellog, 1990; O'Mealia, 2011 ). Across age groups, language backgrounds, and abilities, structured planning before essay writing has been shown to improve clarity and overall quality (Amoush, 2015 ; Rao, 2007 ; Nugraha & Indihadi, 2019 ). Brainstorming is especially valuable for students who struggle with idea expression, confidence, lexical retrieval, or adherence to academic writing conventions (Abedianpour & Omidvari, 2018 ; Hashempour, Rostampour & Behjat, 2015 ). For example, for English as a foreign language (EFL) learners, brainstorming provides cognitive and linguistic scaffolding as well as emotional support, helping to reduce writing anxiety, build confidence, and strengthen their ability to generate, organise and articulate ideas more effectively (Unin, 2016 ; Shirvani and Porkar, 2021 ). The cognitive and emotional benefits of brainstorming can be further amplified in collaborative settings, where social interaction, peer feedback and co-construction of meaning help students engage more actively with their ideas revise their thinking in real time. Ideation with others encourages critical thinking, creativity, and the exchange of alternative perspectives, helping students to build richer and more multidimensional ideas (Al-Khatib, 2012 ; Ghabanchi & Behrooznia, 2014 ). However, human-facilitated brainstorming may not always be accessible. Increasingly, Gen AI is filling this gap, offering personalised, on-demand brainstorming chatbots that can simulate aspects of effective human-led support. These systems can be designed to possess in-depth domain knowledge, ask probing questions, and respond adaptively to student input. Critically, they also offer a “judgment-free” space – an important feature for students who may hesitate to seek help due to fear of negative evaluation (Downing et al., 2020 ; Ryan et al., 1998 ). By lowering social barriers, such chatbots reduce anxiety, encourage intellectual risk-taking, and promote deeper inquiry and self-regulated learning (Zimmerman, 2002 ). 2.2. Chatbots in educational support Recent studies illustrate both the potential and complexity of chatbot-facilitated brainstorming. In a quasi-experimental study, Zhang et al. ( 2025 ) found that students using an AI chatbot (Spark Desk) during argumentative writing showed significant gains in critical thinking skills and intrinsic motivation, including higher enjoyment, greater perceived value of the task and reduced stress, as compared to peers engaged in peer interaction. Similarly, Guo and Li ( 2024 ) showed that students who built their own chatbots for idea generation, outlining and language correction, developed clearer goals, greater writing confidence, and more positive attitudes toward writing. These chatbots also handled diverse requests, including ‘assistance, customisation, and translation’, which would have been overwhelming for a single human teacher to address for multiple students. Still, chatbot-led ideation has drawbacks. AI-generated brainstorming can be less diverse, with high overlap across responses, compared to human brainstorming (e.g., studies report 94% of ideas overlapping). Educators also express reservations: Karanjakwut and Charunsri ( 2025 ) found that while Thai university students using AI chatbot tools outperformed a control group on several assignments, lecturers still preferred student-generated ideas over chatbot suggestions. Meta-analytic evidence provides a more balanced view – Wang and Fan’s ( 2025 ) review of 51 studies concluded that ChatGPT use has a large positive impact on learning performance, and a moderately positive effect on higher-order thinking and perceived learning. Together, these studies suggest that, when thoughtfully integrated, chatbots can serve as flexible, personalised brainstorming partners, and that their influence on students’ engagement and confidence may differ from that of human teachers. Yet most existing work has focused narrowly on linguistic outcomes, without examining students’ emotional responses, let alone make comparisons to experiences with human teachers (Jeon & Lee, 2024 ). This study addresses that gap by examining learners’ emotional and physiological responses, to explore how human and AI tools shape the brainstorming experience. 2.3. Emotion and learning “Emotion is the foundation of learning” (Zull 2006 , p. 7). Traditionally viewed as a cognitive process, learning is now widely recognised as deeply influenced by motivation and emotional processes, including what is learned and retained (Lajoie, 2014 ; Plass & Kaplan, 2016 ; Seli et al., 2016 ). Emotions direct attention, motivation and affect memory (Mayer, 2020 ), and are integral to the causal chain leading to learning outcomes, particularly in digital environments. Learning results not only from instructional design and content, but also from learners’ emotional responses to the process. Emotions, short-lived, intense reactions to specific events or stimuli, can either facilitate or hinder learning, depending on their nature and timing (Duffy et al., 2020 ). Understanding how learners feel is therefore critical for designing more effective learning experiences (Boekaerts, 2010 ). Empirical studies support the influence of specific emotional states on cognitive performance. Emotional content is remembered more clearly than neutral content, as emotional experiences engage both the amygdala and hippocampus during memory formation, enhancing the consolidation of emotionally significant information and leading to stronger long-term recall (Richardson et al., 2004 ; Tyng et al., 2017 ). Positive emotions like enjoyment, pride and hope are strong drivers of learning motivation, academic performance (Pekrun et al., 2002), self-regulation, cognitive flexibility and problem solving (Li et al., 2020 ). However, not all emotions affect learning in the same way. For example, stress can either support or hinder learning depending on its intensity and duration (Vogel & Schwabe, 2016 ). Mild, acute stress may enhance attention and memory, while chronic or excessive stress tends to impair cognitive performance. Importantly, not only positive emotions contribute to effective learning. Certain negative states like confusion, can improve learning outcomes by increasing focus on the learning content (D’Mello et al., 2014 ). Positive emotions like curiosity promote exploration and prepare the brain for learning and material retention in both children and adults (Oudeyer et al., 2016 ). Together, these studies demonstrate that both positive and negative emotional experiences can facilitate learning under the right conditions, highlighting the importance of how emotional states influence attention, engagement and memory in educational contexts. 2.4. AI support for emotional engagement in learning Given that learning is not merely a cognitive process, but also about how learners feel, react and regulate during the process, as educational research increasingly recognises the role of emotion in shaping attention, motivation and memory, there is growing interest in how AI-based systems can support not just what students learn, but how they experience learning emotionally. Recent advances in emotion detection technologies have enabled AI tools to predict and identify emotions that impede learning, such as boredom and frustration, by analysing cues in their chatlogs or interaction patterns. Adaptive systems can then respond to students’ emotions and behaviour dynamically to increase motivation and productivity (Arguel et al., 2019 ; D’Mello & Graesser, 2012; Mehigan & Pitt, 2019 ). For example, Sumithra and colleagues ( 2022 ) demonstrated how by using emotion-detection algorithms, AI systems can adapt lesson content and pace to match students’ emotional intensity and attention levels, while ensuring timely completion. In this sense, AI chatbots can mirror a human tutor’s responsiveness to individual students, while remaining consistent, unbiased and available on demand. Emerging models can even simulate social-emotional intelligence (Gorga & Schneider, 2009 ), thus providing more responsive, emotionally supportive digital learning environments that begin to rival the empathy, adaptability and relational support traditionally offered by human teachers. As Gen AI tools become more emotionally intelligent, integrating physiological data offers a promising next step – providing real time, complementary insight into learners’ emotional states. 2.5. Embodiment and the physiological measurement of emotion Emotions are rooted in the body as much as the mind, involving complex interactions between physiological and neural processes. Theories of interoception and embodied cognition suggest that emotions are shaped not only by subjective interpretations, but also by the perception and regulation of internal bodily states (Barrett & Simmons, 2015 ). Relatedly, emotions reflect the body’s efforts to maintain homeostasis – its internal equilibrium. This balance is disrupted in times of stress or discomfort, and individuals at these moments, tend to experience negative emotional states. Contrastingly, restoring this balance is linked to positive feelings such as calmness (Craig, 2002 ). In educational settings, positive emotional states like interest, curiosity and focus support deeper cognitive processing and sustained attention (Tyng et al., 2017 ). While these experiences are often reported subjectively, they also have physiological signatures, such as changes in skin temperature, heart rate, and electrodermal activity (EDA) (Malmberg et al., 2019 ), which reflect shifts in arousal, attention, and emotional reactivity. However, while self-report questionnaires remain useful for assessing learning and emotion, they are subjected to biases (e.g., social desirability, limited recall, lack of awareness), and do not capture moment-to-moment or subconscious changes in emotional state (Pekrun, 2020 ). To address these challenges, researchers increasingly adopt multimodal techniques that combine self-reports with objective indicators of emotional arousal and engagement (Ketonen et al., 2022), including classroom observations, eye-tracking, learning analytics, and physiological signals (e.g., McNeal et al. 2020 ). EDA, commonly used in studies assessing student engagement (Horvers et al., 2021 ), tracks changes in skin conductance driven by the sympathetic nervous system and is sensitive to shifts in emotional intensity and attention (Boucsein, 2011 ). Heart rate provide insight into emotional valence and cognitive effort; higher heart rate for instance, is associated with increased cognitive workload (Darnell & Krieg, 2019 ; Heine et al., 2017 ). Skin temperature typically drops during sympathetic activation (e.g., anxiety) due to vasoconstriction, and rises during relaxed or regulated states (Gouzi et al., 2011). Because physiological data and self-report tap into different aspects of experience, they do not always align. Self-reports tend to offer summarised or retrospective accounts of how learners felt, whereas physiological data capture the immediate and embodied dimensions of emotional responses that may be fleeting, subtle or subconscious (e.g., Ketonen et al., 2019). Capturing the embodied dimensions of emotional dynamics allows researchers to move beyond what learners say they feel, to observe how their bodies respond to specific instructional events (Harley, 2016 ). This is especially useful when comparing interactions with human instructors versus Gen AI systems, where differences in emotional and cognitive engagement may not be fully verbalised but can still be detected physiologically. 3. Methods Ethics approval from Nanyang Technological University's institutional review board (IRB) has been obtained prior to the commencement of the study (IRB-2024-1031). This study employed a multimodal design to investigate learners’ experiences with AI- versus human-facilitated brainstorming during an essay planning task. Participants first completed a pre-task mood questionnaire, then engaged in a 25-minute brainstorming session with either a human or chatbot facilitator. Immediately afterward, they completed a post-task mood questionnaire, and wrote an essay outline. Throughout the brainstorming phase, continuous physiological signals were recorded using a wearable sensor, enabling assessment of not only how learners report feeling, but also how their bodies respond, thus offering alternative perspective into the embodied dimensions of human-AI interaction. The essay outline writing activity served two primary purposes. Firstly, it provided a concrete context and goal, helping to anchor participants’ sense of purpose and engagement during the brainstorming process. Secondly, it offered an objective behavioural outcome measure of brainstorming effectiveness of each facilitation condition. However, because this paper focuses on emotional and physiological responses, analyses of writing performance fall outside the current scope and will be reported separately. We recruited 30 undergraduates and postgraduates, as they are likely to engage in learning activities with both AI and the human facilitator. Participants were randomly assigned to one of two conditions: 1) AI Chatbot Condition (n = 15; 11 female; M age = 28.6, SD = 6.92) : Participants brainstormed ideas with the assistance of a customised messaging chatbot; 2) Human Teacher Condition (n = 15; 11 female; M age = 29.2, SD = 5.15): Participants engaged in the same brainstorming task with the help of a human teacher via the chat function on Zoom, without the audio and video feature turned on. The identities of the participants in the human teacher condition were not revealed to the human teacher and vice versa. The brainstorming task centred on the essay question, ‘Should Singapore embrace Singlish as a key part of its national identity, or does it undermine its global image?’ Participants were instructed to develop an outline for an argumentative essay in response to this prompt after the brainstorming session. The essay outlines were then graded according to a rubric developed. The Gen AI chatbot used in the study was developed using SchoolAI platform (app.schoolai.com), which enables the customisation of chatbots using tailored prompts and supplementary materials. For this study, the chatbot’s knowledge base was augmented with selected readings on Singapore English and Singlish. Unlike typical chatbots that serve to provide direct and complete answers, the design strategy adopted for our chatbot followed a Socratic questioning approach – posing reflective, open-ended questions to encourage critical thinking and avoiding giving participants direct answers. The specific system prompts used can be found in Appendix A. The human teacher engaged in the study is a lecturer in a reputable university in Singapore with teaching experience in Academic Writing, Professional Communication, Linguistics and English Language across various tertiary institutions. Their academic specialisation includes English in Singapore, providing a strong match to the chatbot’s domain expertise. Before brainstorming, participants completed a baseline mood questionnaire and a 5-minute control condition viewing a neutral video ( https://www.youtube.com/watch?v=zpHOIlQbj8Y ) to elicit a stable, low-arousal state. These baseline readings served as emotional and physiological baselines, accounting for individual differences in resting states and enabling more accurate condition comparisons. Participants were explicitly informed in advance whether they would be interacting with a chatbot or a human teacher to mimic real-world learning scenarios, where individuals are typically aware of the nature of their learning facilitators. By doing so, we aimed to enhance the ecological validity of the study as it reflects more accurately the variety of help-seeking behaviours and communication experiences participants may encounter outside the study setting, while avoiding introducing confusion or expectancy effects during the task. In particular, since it is highly plausible that participants in the chatbot condition might infer the identity of conversational partner over time, we sought to prevent any potential disruptions in emotional or physiological responses caused by such realisations. Following the brainstorming phase, participants completed a follow-up questionnaire containing the same mood items found in the pre-brainstorming questionnaire as well as additional 15 questions on perceptions of the brainstorming interaction. These self-reported measures not only allowed for direct comparison to their baseline mood, but also served to contextualise and supplement the physiological data, providing a more comprehensive view and nuanced analysis of the impact of the two experimental conditions (AI vs human). The mood questionnaire contained 20 items (10 positive, 10 negative; e.g., excited, stressed, frustrated), the intensity of each was rated on a 5-point Likert scale from “None” to “Very intense”. These ratings were used to assess mood change over time and compare emotional outcomes across the conditions (human vs chatbot). The second questionnaire’s additional 15 perception questions covered aspects such as effectiveness and ease of communication. See Appendix B for the full list of items. Physiological responses were collected using the EmbracePlus wearable sensor, which records continuous, real-time, non-invasive biomarkers relevant to emotional and cognitive engagement. Specifically, we measured heart rate, EDA, and skin temperature. These measures were selected based on prior research linking them as proxies for affective arousal, attentional focus, and engagement in learning contexts (Fang et al., 2018 ; Soltis et al., 2020 ). 3.1. Data analyses All analyses were conducted in R (version 4.3.1; R Core Team, 2025) using descriptive statistics, MANCOVA, regression, and mixed-effects models to examine physiological and emotional responses during the brainstorming task. Analyses addressed: (1) group differences between the Chatbot and Human teacher conditions, and (2) associations between physiological activation and mood outcomes. Mood Analyses Mean mood ratings were first compared descriptively at baseline and brainstorming phases. However, analysing raw mood ratings at each phase separately risks misinterpretation. For instance, a higher brainstorming phase mood rating in the human condition does not necessarily mean the chatbot group showed no improvement; it simply reflects an endpoint comparison without accounting for each participant’s baseline. To more accurately capture emotional shifts, mood change scores were computed for each item by subtracting the baseline rating from the brainstorming rating (Brainstorming – Baseline). Positive change scores indicate increased mood intensity during brainstorming (Brainstorm > Baseline), negative scores reflect decreases (Brainstorming < Baseline). This approach 1) implicitly controls for baseline differences without adding 20 baseline mood covariates alongside Gender and Age; and 2) preserves statistical power, which is crucial given the relatively small sample size and the risk of overfitting in multivariate models. Change scores were entered into a MANCOVA to assess overall condition effects, controlling for gender and age. Significant or marginal multivariate effects were followed up with univariate ANCOVAs and regressions to identify specific moods differing by condition (Chatbot vs. Human). Physiological Analyses : Physiological data (heart rate, EDA, temperature) were analysed separately for the baseline (5 min; 5 data points) and brainstorming (25 min; 25 data points) phases. For each phase, two types of metrics were used: raw means and Area Under the Curve (AUC) values. AUCs were calculated using the trapezoidal method to capture sustained physiological engagement across each phase (5-min baseline, 25-min brainstorming). AUC reflects total physiological activation over time and is useful for summarising prolonged states such as engagement, arousal and stress. One participant contributed only 4 baseline data points, and thus their baseline AUC could not be computed. This participant was excluded from analyses involving baseline AUC. All physiological measures (raw means and AUCs) were analysed using the same statistical approach. For the baseline phase, MANCOVAs compared groups (chatbot vs human) with age and gender as covariates. Significant multivariate effects were followed by univariate ANCOVAs. For the brainstorming phase, MANCOVAs tested group differences while controlling for all corresponding baseline measures, age and gender. This approach allowed the examination of group differences in the overall physiological profile during brainstorming while adjusting for baseline levels. Significant multivariate effects were followed by univariate ANCOVAs, and covariate-adjusted means with 95% confidence intervals were extracted to determine the direction of effects. To analyse minute-by-minute trajectories of physiological change during brainstorming, linear mixed-effects models (using the lmer function in R) were fitted. This approach was chosen because it accommodates repeated measures data, allowing us to model both between-subject differences and within-subjects changes over time. The models included fixed effects for Time, Condition and their interaction, as well as random intercepts by participant to account for individual variability (e.g., each participant may have inherently different physiological levels at baseline). The random effects structure was constructed following the recommendations of Bates et al. ( 2015 ), prioritising an accurate reflection of the experimental design while also maintaining model parsimony. This growth curve modelling approach allowed us to assess whether the trajectories of physiological responses differed by condition across the 25-minute brainstorming session. Integration of Mood and Physiological Data To explore the relationship between physiological responses and mood changes, independent of experimental condition, multiple regression analyses were conducted in which mood change scores (both composite positive/negative mood scores and individual mood items) were predicted from physiological AUC values. This approach assessed the predictive value of physiological responses on subjective mood experiences, regardless of condition. 4. Findings 4.1. Mood Descriptive statistics At baseline, participants in the chatbot condition reported higher levels of positive emotions, while those in the human condition reported higher levels of negative emotions. However, this pattern reversed post-brainstorming: chatbot participants experienced more intense negative moods, whereas the human condition saw greater increases in positive emotions. Descriptive statistics and mood change scores (brainstorm – baseline) are presented in Appendix C Table C1 . Both groups showed increases in positive moods 7 of 10 moods, with larger gains in the human teacher group. Three positive moods saw chatbot-only decreases – Curious, Motivated and Excited. The chatbot group also showed increases in negative moods, particularly Stress, Discouragement, Boredom and Distraction. Inferential statistics for baseline mood and mood change scores At baseline, a MANCOVA controlling for gender and age revealed no significant multivariate effect of Condition (Pillai’s Trace = 0.60, F (20, 6) = 0.45, p = .92), indicating that both groups began the brainstorming session with comparable mood states. For mood change scores (Brainstorming mood intensity – Baseline mood intensity), a MANCOVA controlling for gender and age revealed marginal divergent emotional shifts between conditions ( Pillai’s Trace = 0.91, F (20, 6) = 3.20, p = 0.077). Univariate ANCOVAs indicated significant Condition effects for Motivated, Inspired, Empowered, Engaged, Stressed, Discouraged, Bored, and Distracted, and marginally significant condition effects for Focused, Excited, Connected, Determined, and Frustrated. Follow-up regressions showed that effects were driven by greater increases in positive moods (Motivated, Inspired, Empowered, and Engaged) for participants in the human teacher group, with marginal gains in Excited and Connected. Conversely, the chatbot group showed significantly greater increases in negative moods – Stressed, Discouraged, Bored, and Distracted, with marginally greater Frustrated. Focused and Determined did not show significant condition effects in the regression models. Table 1 presents inferential statistics for mood changes scores. Table 1 Significant and marginal condition effects on mood change scores from ANCOVAs and follow-up regressions Mood ANCOVA F (1, 25 ), p β (SE) t(25), p Direction Motivated Change 10.76, .003 0.84 (0.26) 3.25, .003 ** Human > Chatbot Inspired Change 8.01, .009 0.93 (0.33) 2.84, .009 ** Human > Chatbot Empowered Change 10.86, .003 0.86 (0.26) 3.23, .003 ** Human > Chatbot Engaged Change 11.70, .002 1.00 (0.30) 3.37, .002 ** Human > Chatbot Excited Change 3.74, .06 0.75 (0.49) 1.88, .07 † Human > Chatbot Connected Change 3.59, .07 0.71 (0.38) 1.88, .07 † Human > Chatbot Stressed Change 12.99, .001 -1.18 (0.33) -3.62, .001 ** Human < Chatbot Discouraged Change 11.11, .003 -1.00 (0.29) -3.41, .002 ** Human < Chatbot Bored Change 6.65, .02 -0.81 (0.32) -2.54, .02 * Human < Chatbot Distracted Change 9.23, .006 -1.27 (0.41) -3.06, .005 ** Human < Chatbot Frustrated Change 3.24, .08 -0.64 (.34) -1.91, .07 † Human < Chatbot Determined Change 3.22, .08 0.47 (0.28) 1.66, .11 - - Focused Change 3.02, .09 0.51 (0.30) -0.33, .75 - - Note. p values are two-tailed. * denotes p < .05, ** denotes p < .01, † denotes 05 ≤ p < .10. 4.2. Perception of the Brainstorming Session A MANCOVA was conducted to examine whether perceptions relating to the value and cognitive impact of the brainstorming session differed between the chatbot and human teacher conditions, while controlling for gender and age. The overall multivariate effect of Condition was statistically significant, Pillai’s Trace = 0.82, F (15, 10) = 3.05, p = .040, indicating that participants' perceptions of the brainstorming session varied depending on the interaction partner. Specifically, participants who brainstormed with a human teacher reported more positive perceptions of the session across several items. They were more likely to describe their brainstorming partner as knowledgeable, felt more comfortable interacting, more likely to indicate learning something new, and also found it easier to communicate ideas. Marginally, they reported greater enjoyment. In contrast, those in the chatbot condition were marginally more likely to report feeling judged and significantly more likely to find the experience mentally exhausting. See Table 2 for descriptive and inferential statistics. Table 2 Descriptive and inferential statistics on participants’ perceptions of the brainstorming session Item M ( SD) (Chatbot) M (SD) (Human) t (25) p 1. The brainstorming session helped me generate useful ideas for my essay outline. 4.07 (0.80) 4.27 (0.88) 0.90 .375 2. I found the brainstorming partner to be knowledgeable. 3.20 (1.37) 4.53 (0.74) 3.45 .002 ** 3. I feel better prepared to write my essay after the session. 3.87 (0.92) 4.20 (1.08) 0.85 0.40 4. It was easy to communicate my ideas during the session. 3.40 (1.18) 4.20 (0.68) 2.22 .036 * 5. I felt judged during the brainstorming session. 2.27 (1.03) 1.60 (0.91) -1.76 .091 6. I felt engaged throughout the brainstorming session. 3.80 (0.86) 4.27 (1.03) 1.29 .211 7. I enjoyed the brainstorming process. 3.60 (1.12) 4.27 (1.10) 2.06 .050 8. The session maintained my interest in the topic. 3.64 (1.0) 4.53 (0.52) 2.93 .007** 9. I found the brainstorming process mentally exhausting. 2.53 (1.19) 1.53 (0.64) -2.63 .014 * 10. The session challenged me to think critically about the topic. 3.27 (1.44) 4.07 (0.70) 1.91 .067 11. I was encouraged to explore different perspectives. 3.93 (1.16) 4.33 (0.82) 1.20 .243 12. I learned something new during the brainstorming session. 3.27 (1.39) 4.47 (0.52) 3.39 .002 ** 13. I felt comfortable interacting with the brainstorming partner. 3.73 (1.22) 4.60 (0.83) 2.93 .007 ** 14. I think brainstorming with a teacher/chatbot* would be more effective. 3.73 (1.03) 3.13 (1.41) -1.58 .128 15. I communicate the same way when using an AI chatbot as when interacting with a human 2.20 (1.08) 1.80 (1.26) -1.61 .120 Despite these differences, both groups rated the session comparably in key outcome areas: they felt equally engaged, challenged to think critically, encouraged to consider different perspectives, and believed the session helped generate useful ideas and better prepared them for essay writing. Notably, both groups remained neutral on whether they believed a chatbot or human would be more effective beforehand. 4.3. Physiological Responses Raw (unadjusted) summary measures (raw means and AUCs) for each physiological variable by condition and phase are reported in Appendix C (Tables C2 – C3 ). Baseline phase: mean values and AUC. A MANCOVA on baseline mean values (controlling for age and gender) revealed a significant overall effect of Condition, Pillai’s Trace = 0.347, F (3, 23) = 4.08, p = .02. Pulse was significantly higher in the chatbot than human condition, F (1, 25) = 11.22, p = .003 ( M = 85.2, 95% CI[79.9, 90.5] vs. M = 73.8, 95% CI[68.3, 79.2]); EDA and Temperature did not differ between conditions (both ps > .20). A MANCOVA on baseline AUCs also showed a significant effect of Condition, Pillai’s Trace = 0.500, F (3, 22) = 7.32, p = .001, with higher pulse AUC in the chatbot condition, F (1, 24) = 10.28, p = .004, M = 341, 95% CI[318, 364] vs. M = 295, 95% CI [273, 318]), and marginally predicted Temperature ( F (1, 24) = 4.18, p = .05), but not EDA ( p > .50). Temperature AUC was marginally higher in the human condition ( M = 125, 95% CI [122, 127]) than chatbot condition ( M = 122, 95% CI [120, 124]). Brainstorming phase: mean values and AUC. A MANCOVA on mean values during brainstorming (controlling for all three baseline measures, gender and age), yielded a significant multivariate effect of Condition, Pillai’s Trace = 0.827, F (3, 22) = 31.86, p < .001. Follow-up univariate tests revealed that Condition significantly predicted Pulse, F (1, 22) = 86.27, p .22). Covariate-adjusted means for Pulse showed that heart rate was slightly higher in the chatbot condition ( M = 80.4, 95% CI [78.1, 82.7]) than in the human condition ( M = 79.2, 95% CI [76.9, 81.5]) (see Fig. 1 ). The corresponding AUC analysis showed a significant overall effect of Condition, Pillai’s Trace = 0.864, F (3, 19) = 40.36, p < .001, with higher pulse AUC in the chatbot condition, F (1, 21) = 130.74, p < .001 ( M = 1917, 95% CI[1871, 1964] vs M = 1867, 95% CI[1819, 1916]), and marginally higher cumulative peripheral temperature (AUC) in the human condition, F (1, 21) = 3.77, p = .07 ( M = 763, 95% CI[754, 773] vs. ( M = 757, 95% CI[748, 767]). There was no significant condition effect on EDA AUC ( p > .47). These are illustrated in Figs. 2 and 3 . Time series analysis. Linear mixed-effects models of minute-by-minute physiological data from the 25-minute brainstorming phase, with baseline values as covariates and random intercepts for participants, revealed no significant differences between the chatbot and human conditions in physiological change trajectories (all ps > .44) Full model outputs for each physiological measure are provided in Appendix D. 4.4. Predicting mood change through brainstorming from physiology (AUCs) To investigate whether brainstorming mood could be explained by participants’ physiological responses during the task, a series of regression models were fitted using AUC values (Pulse, Temperature, EDA) to predict mood change. Analyses were conducted at two levels: (1) aggregated positive/negative mood change scores, and (2) individual mood change outcomes. For the composite negative mood change score, the regression showed that Pulse AUC positively predicted increased negative mood intensities ( t (26) = 2.47, p = .02). Examining individual negative mood items, pulse AUC significantly predicted increased stress ( t (26) = 3.03, p = .006), boredom ( t (26) = 2.25, p = .03), and was marginally associated with discouragement ( t (26) = 1.96, p = .06). Pulse AUC also showed a marginal negative association with positive moods ( t (26) = -1.98, p = 0.06), with lower pulse AUC associated with feeling more empowered ( t (26) = -3.77, p < .001), engaged ( t (26) = -3.18, p = .004), connected ( t (26) = -3.12, p = .004), and determined ( t (26) = -2.42, p = .02). Temperature AUC emerged as a predictor of mood outcomes, showing a marginally positive relationship with composite positive mood change ( t (26) = 2.02, p = .05). Specifically, higher temperature AUC significantly predicted increases in positive moods such as empowered ( t (26) = 3.54, p = .002), connected ( t (26) = 2.93, p = .007), and inspired ( t (26) = 2.13, p = .04). Additionally, excitement was also marginally predicted by Temperature AUC ( t (26) = 1.80, p = .08). Conversely, lower temperature AUC was associated with stronger negative emotional responses, including feeling more frustrated ( t (26) = -2.20, p = .04), and marginally more stressed ( t (26) = -1.78, p = .087). For the composite positive and negative mood change scores, EDA AUC was not a significant predictor. However, at the individual level, it was positively associated with increased stress ( t (26) = 2.48, p = .02), and was marginally associated with feeling less connected ( t (26) = -1.81, p = .08), less confused ( t (26) = -2.02, p = .05), and feeling less challenged ( t (26) = -1.72, p = .098). 5. Discussion 5.1. Mood At baseline, the mood ratings between the two conditions (Chatbot vs. Human) were not significantly different. While both groups experienced an increase in positive mood intensities through the brainstorming session, the participants in the human teacher condition showed greater increase in positive feelings, such as ‘Motivated’, ‘Inspired’, ‘Empowered’, and ‘Engaged’ than those who interacted with the chatbot. Conversely, those in the chatbot condition experienced increased intensities in several negative moods while those in the human condition experienced a decrease. These results suggest that while both groups started on equal emotional footing, the quality of interaction shaped the emotional impact of the brainstorming session. Although the human-led condition elicited stronger emotional improvements overall, the intensity change for many mood states – both ‘positive’ and ‘negative’, such as Focused, Curious, Confident, Determined, Confused, Challenged, Overwhelmed, Anxious, and Annoyed – did not differ significantly between groups. This highlights that the chatbot’s ability to support a wide range of engagement-related emotional states should not be discounted. Moreover, not all negative emotions are detrimental; moods like frustration or feeling challenged can be facilitative of learning, reflecting productive cognitive effort. This experience, also known as cognitive disequilibrium, arises when individuals engage with unfamiliar material that signals knowledge gaps or conflicts with their existing knowledge (Piaget, 2005/1950). To resolve this tension and move forward, individuals are driven to assimilate new information into existing schemas or accommodate by adjusting their schemas. Thus, certain ‘negative’ emotions may indicate meaningful cognitive engagement and can even be beneficial for deeper learning. 5.2. Perceptions of brainstorming session Participants’ perceptions of the brainstorming session were generally positive, ranging from 3.20 to 4.60, with those who interacted with the human teacher rating their experience more positively. For several aspects, these group differences were significant. This included finding the human teacher more knowledgeable, feeling more comfortable interacting with the human teacher, being more likely to indicate having learnt something new, and also finding it easier to communicate ideas. Conversely, those in the chatbot condition found the experience more mentally taxing and were marginally more likely to feel judged. This may be attributed to the way we designed the chatbot to intentionally avoided providing direct answers. Instead, it was designed to use a Socratic approach and engage participants with a series of question prompts aimed at helping them to think critically and consider different perspectives to ultimately come up with original ideas. While this approach aligns with educational goals of fostering deeper thinking, it may have deviated from participants’ usual experience with Gen AI tools such as ChatGPT, tools that often provide more immediate and ‘fuller answers’. As a result, participants may have perceived the chatbot interaction as more cognitively demanding, leading to greater fatigue or frustration. Despite the differences noted between conditions, both groups rated the brainstorming session comparably on key outcome areas. Participants across both conditions felt engaged throughout, challenged to think critically, encouraged to consider different perspectives, and believed the session helped them to generate useful ideas and better prepared them for essay writing. These findings highlight the value of human interaction in fostering a sense of comfort, perceived expertise, and cognitive ease, while also suggesting that Gen AI chatbots – despite limitations in emotional nuance – can still support core learning aspects like engagement, intellectual stimulation and critical thinking. Notably, both groups remained neutral when asked, after the brainstorming session, whether they believed a chatbot or human would be more effective. This suggests that despite differences in user experience, both interaction types were perceived as similarly effective in achieving the goal of the task – generating ideas for an essay outline. 5.3. Physiological responses A significant difference in pulse was found, with participants in the chatbot condition showing higher cardiovascular activation than those in the human condition (for both mean values and AUC). Temperature AUC was marginally lower in the chatbot condition. This physiological profile – elevated heart rate alongside slightly reduced peripheral temperature may reflect greater sympathetic arousal, potentially involving some vasoconstriction which is commonly associated with heightened alertness or stress (e.g., Hayashi et al., 2008). When considered alongside the mood results, which showed greater increases in negative emotions in the chatbot condition, this suggests participants may have experienced the task as more cognitively demanding. As noted earlier, the chatbot’s Socratic questioning style, while aimed at fostering deeper thinking, may have been more effortful than expected. However, time series analysis using mixed-effects modelling revealed no significant differences between conditions in physiological change trajectories over the 25-minute brainstorming session, once baseline AUC, gender and age were controlled. Despite differences in mood and perceived cognitive impact, both interaction types elicited similar patterns of physiological change across time. These findings suggest that GenAI chatbots can match human facilitators in fostering physiological engagement and serve as helpful pedagogical aides for the teacher. 5.4. Prediction of mood from physiology Pulse AUC was positively associated with increased negative moods (stress, discouragement, boredom), and negatively with positive moods (empowerment, engagement, connection, determination). This is consistent with evidence that heart rate tracks emotional valence and cognitive effort (e.g., Darnell & Kreig, 2019). Mechanistically, increased negative emotions likely involves increased sympathetic activation and/or reduced parasympathetic control, which produce increased and more sustained pulse rate and amplitude over time (higher AUC) (Kreibig, 2010 ). In contrast, positive moods like empowerment, engagement, connection and determination may index ‘regulated activation’ where the body is engaged for the task, but not in an over-aroused stressed state, and the parasympathetic system is helping to keep the heart rate in check, thus producing lower or more transient elevated cardiovascular activation (lower AUC). According to the biopsychosocial model of challenge and threat (Blascovich, 2013 ; Seery, 2011 ), when individuals perceive demands as threats (e.g., discouragement, stress), their cardiovascular responses are less efficient and more ‘costly’ over time, whereas perceiving demands as challenges (e.g., empowerment, determination) produce a more efficient profile, which fits with the patterns we identified. Temperature AUC was marginally associated with increases in positive mood intensity (empowered, connected, inspired, and excitement), and negatively to negative mood change including frustration and marginally increased stress. This pattern is aligned with research that positive emotional engagement is accompanied by peripheral vasodilation and higher skin temperature, potentially mediated by parasympathetic activity (Stefano et al., 2008 ). Conversely, negative moods like stress and frustration are associated with sympathetic-driven vasoconstriction, which decreases peripheral blood flow and lowers skin temperature (e.g., Hayashi et al., 2008). This inverse relationship with Pulse AUC suggests a physiological trade-off between cardiovascular arousal and warmth, where heightened sympathetic activation (high pulse AUC) may coincide with peripheral cooling, while higher skin temperature may reflect a more regulated, emotionally positive state supported by parasympathetic vasodilation (Aristizabal-Tique et al., 2023 ). EDA AUC was positively related to increased stress, consistent with its role in tracking sympathetic nervous system activation (Akbulut, 2022 ). Its negative association with feelings of connection may reflect the social consequences of heightened arousal. Stress may skew social appraisal, reducing perceived social support and fostering a sense of disconnection or increased social distance (Wang et al., 2024 ). Under stress, individuals may also be more task-oriented, investing less in relational engagement or in building rapport with the brainstorming partner. The physiological indices that differed between conditions (pulse and temperature AUCs) were also systematically linked to mood variation across participants, independent of facilitator type. Although causality cannot be inferred, this convergence suggests that these physiological indicators capture emotional shifts and reflect core bodily processes underpinning participants’ emotional experiences during the task, whether interacting with a human or chatbot. This alignment between mood reports and physiological activity validates the self-reported data, showing that the emotional states were reflected in real-time bodily activity – indicating that these experiences were not merely subjective impressions, but embodied and measurable. Importantly, the predictive value of physiological markers, particularly Pulse and Temperature AUC, illustrates that beyond group-level differences, the pattern and emotional meaning of physiological activation differ across individuals. Higher pulse AUC predicted more negative mood states, while higher temperature AUC predicted increases in positive emotions. These results indicate that what participants felt during the session is not just explained by group averages, but also by how their bodies responded within each condition. This reinforces the importance of looking beyond group-level comparisons to examine within-individual variation and its emotional significance. Such insights provide meaningful implications for the development of emotionally intelligent GenAI tools. For instance, designers can use physiological research to better understand how different interaction styles or system prompts affect emotional and cognitive responses. Insights from physiological patterns, such as the association between higher heartrate and emotional strain, or between higher skin temperature and positive engagement can inform the development of chatbots that better support learners’ sustained engagement more effectively. While physiological data provided insights into participant’s emotional and cognitive states, it is important to note that in future studies of this nature, physiological responses alone is unlikely to reliably distinguish between the chatbot and human conditions. Physiological metrics, though valuable should be interpreted alongside subjective and behavioural measures to gain a more holistic understanding of user experiences. Relying solely on physiological data may obscure subtle differences that may only emerge through self-report or qualitative data, while relying solely on subjective data may overlook meaningful physiological patterns. 6. Conclusion While chatbots are not necessarily superior to human teachers, they are also not demonstrably worse in many respects. Both groups in our study experienced emotional gains, reflected in improved mood, and exhibited largely similar physiological responses, with only modest differences in pulse and temperature AUCs. Perceived effectiveness as expressed through their questionnaire responses was also high across both conditions. Importantly, the chatbot environment still supported key aspects of cognitive and motivational engagement. Pedagogically, the findings reaffirm the enduring value of the human teacher in engaging and motivating the students. Participants consistently rated the human teacher more favourably across multiple dimensions – comfort, enjoyment, learning gains, and ease of communication – emphasising the irreplaceable role of human empathy, adaptability, and non-verbal reassurance in teaching. Notwithstanding, the study also makes a strong case for the educational viability of GenAI chatbots, particularly as cognitive partners that can stimulate reflection and critical thinking. Despite limitations in emotional warmth, the chatbot condition still supported high levels of engagement and intellectual challenge, as evidenced by both self-reports and physiological data. This suggests that, when designed thoughtfully, chatbots can serve as scalable pedagogical aides that can support the teacher’s design of the learning experience. Its value is especially apparent under circumstances where it is advantageous to have round-the-clock availability, personalised and objective tutoring. They can help mitigate the practical and time constraints that often limit human teaching capacity, despite being unable to fully replicate the nuances of ideal human interaction. Our study contributes to the understanding of the relationship between learning and emotion in the use of GenAI for education. Our study is premised on the recognition that learning is not only a cognitive activity but also a deeply emotional and physiological experience. The observed mood shifts, particularly the greater positive emotional gains in the human teacher condition, highlight the importance of relational dynamics and perceived emotional presence in shaping students’ receptivity and engagement. Our study employs a multimodal, mixed-methods approach that integrates self-reported mood and continuous physiological tracking. This design enhances the reliability of emotional data by triangulating subjective experiences with physiological indicators such as electrodermal activity (EDA), pulse, and skin temperature. In doing so, it mitigates common limitations of self-report measures – such as social desirability bias and retrospective inaccuracies. The integration of physiological data also offers empirical evidence that emotional states are not only subjectively experienced but physiologically embodied. The predictive power of physiological markers – particularly skin temperature and heartrate – strengthens the case for multimodal learning theories on embodied teaching (Lim, 2020 ) and learning (Barsalou, 2008 ). The use of AUC and growth curve modelling of physiological data offers methodological contribution in capturing sustained emotional and cognitive engagement over time, rather than relying on static point measurements. These techniques illuminate not just whether physiological differences exist, but how they unfold across learning episodes. The nuanced analyses, which control for baseline individual variability and contextual factors like age and gender, ensure rigour in interpreting affective responses attributed to interaction. Our study also contributes towards a replicable methodological template for future studies aiming to assess affective receptivity and engagement in human–AI interaction, particularly in educational contexts where physiological data can complement more traditional outcome measures. While the study has found that the GenAI chatbot was able to increase students’ positive moods, thereby motivating them in their learning, a limitation in this study relates to the design of the chatbot we used. Our chatbot was programmed to use a Socratic questioning approach to promote critical thinking; however, some participants reported feeling mentally exhausted or frustrated as it was atypical to the usual chatbots they had used which would offer them answers directly. Future iterations of the chatbot could incorporate adaptive mechanisms to detect circular or stalled conversations and provide more direct information or examples when necessary, which may help reduce students’ feelings of frustration. Feedback from the participants also suggests that the chatbot would benefit from emotional augmentation features, such as detecting and responding to signs of user frustration in real time (Arguel et al., 2019 ), adjusting its tone to match that of participants, explicitly acknowledging their effort or confusion, in order to generate greater perceived feelings of empathy and support. Ultimately, it is essential to recognise that AI chatbots and human teachers are fundamentally different (e.g., communication styles, adaptability, sensitivity to nuance). Finding the right fit between learner needs and the type of facilitator will likely be complex, and there is unlikely to be a single, universal solution. An important value that GenAI affords is the myriad of ways that the chatbots can be designed for different educational purposes and audiences. For example, some contexts may benefit from highly structured, informative chatbots, while others might require more conversational agents that are capable of providing emotional support and nuanced dialogue. Future work could explore how different learners respond to varying levels of cognitive and emotional support from GenAI chatbots, to inform the development of personalised chatbot designs that can effectively support a variety of pedagogical innovations. References Abedianpour, S., & Omidvari, A. (2018). Brainstorming strategy and writing performance: Effects and attitudes. Journal of Language Teaching and Research , 9 (5), 1084-1095. Akbulut, F. P. (2022). Evaluating the effects of the autonomic nervous system and sympathetic activity on emotional states. Istanbul Ticaret Üniversitesi Fen Bilimleri Dergisi , 21 (41), 156-169. Al-Khatib, B. A. (2012). The effect of using brainstorming strategy in developing creative problem solving skills among female students in Princess Alia University College. American International Journal of Contemporary Research , 2 (10), 29-38. Amoush, K. H. (2015). The impact of employing brainstorming strategy on improving writing performance of English major students at Balqa Applied University in Jordan. Journal of Education and Practice , 6 (35), 88-92. Arguel, A., Lockyer, L., Kennedy, G., Lodge, J. M., & Pachman, M. (2019). Seeking optimal confusion: a review on epistemic emotion management in interactive digital learning environments. Interactive Learning Environments , 27 (2), 200-210. Aristizabal-Tique, V. H., Henao-Pérez, M., López-Medina, D. C., Zambrano-Cruz, R., & Díaz-Londoño, G. (2023). Facial thermal and blood perfusion patterns of human emotions: Proof-of-Concept. Journal of thermal biology , 112 , 103464. Barrett, L. F., & Simmons, W. K. (2015). Interoceptive predictions in the brain. Nature reviews neuroscience , 16 (7), 419-429. Barsalou, L. W. (2008). Grounded cognition. Annual Review of Psychology, 59, 617–645. https://doi.org/10.1146/annurev.psych.59.103006.093639 Bates, D., Kliegl, R., Vasishth, S., & Baayen, H. (2015). Parsimonious mixed models. arXiv preprint arXiv:1506.04967 . Blascovich, J. (2013). Challenge and threat appraisal. In Handbook of approach and avoidance motivation (pp. 431-445). Psychology Press. Boekaerts, M. (2010). The crucial role of motivation and emotion in classroom learning. The nature of learning: Using research to inspire practice , 91-111 Boucsein, W. (2011). Applications of electrodermal recording. In Electrodermal activity (pp. 259-523). Boston, MA: Springer US. Cevher, A. Y., & Yildirim, S. (2023). Design and Use of Chatbots for Educational Purposes: A Study on Student Opinions. Ileri Teknolojilerde Çalismalar Dergisi , 1 (2), 76-83. Chandel, P., & Lim, F. V. (2025). Generative AI and literacy development in the language classroom: A systematic review of literature. Ubiquitous Learning , 18 (2), 31. Craig, A. D. (2002). How do you feel? Interoception: the sense of the physiological condition of the body. Nature Reviews Neuroscience, 3(8), 655–666. https://doi.org/10.1038/nrn894 Crossley, S. A., Muldner, K., & McNamara, D. S. (2016). Idea generation in student writing: Computational assessments and links to successful writing. Written Communication , 33 (3), 328-354. D'Mello, S. K., & Graesser, A. (2012). Language and discourse are powerful signals of student emotions during tutoring. IEEE Transactions on Learning Technologies , 5 (4), 304-317. D’Mello, S., Lehman, B., Pekrun, R., & Graesser, A. (2014). Confusion can be beneficial for learning. Learning and Instruction , 29 , 153-170. Demeure, V., Niewiadomski, R., & Pelachaud, C. (2011). How is believability of a virtual agent related to warmth, competence, personification, and embodiment? Presence. Teleoperators & Virtual Environments, 20(5), 431–448. https://doi.org/10.1162/ PRES_a_00065 Darnell, D. K., & Krieg, P. A. (2019). Student engagement, assessed using heart rate, shows no reset following active learning sessions in lectures. PloS one , 14 (12), e0225709. Downing, V. R., Cooper, K. M., Cala, J. M., Gin, L. E., & Brownell, S. E. (2020). Fear of negative evaluation and student anxiety in community college active-learning science courses. CBE—life sciences education , 19 (2), ar20. Duffy, M. C., Lajoie, S. P., Pekrun, R., & Lachapelle, K. (2020). Emotions in medical education: Examining the validity of the Medical Emotion Scale (MES) across authentic medical learning environments. Learning and Instruction , 70 , 101150. Ericsson, E., Sofkova Hashemi, S., & Lundin, J. (2023). Fun and frustrating: Students’ perspectives on practising speaking English with virtual humans. Cogent Education , 10 (1), 2170088. Fang, C., Liu, Z., Li, Q., & Liu, S. (2018, December). Sensor Data-Driven Emotion Perception in Physical Learning Spaces-A Review and Prospect. In 2018 Seventh International Conference of Educational Innovation through Technology (EITT) (pp. 1-5). IEEE. Ghabanchi, Z., & Behrooznia, S. (2014). The impact of brainstorming on reading comprehension and critical thinking ability of EFL learners. Procedia-Social and Behavioral Sciences , 98 , 513-521. Gorga, D., & Schneider, D. K. (2009). Computer-based learning environments with emotional agents. In Handbook of research on synthetic emotions and sociable robotics: new applications in affective computing and artificial intelligence (pp. 413-441). IGI Global Scientific Publishing. Gouizi, K., Bereksi Reguig, F., & Maaoui, C. (2011). Emotion recognition from physiological signals. Journal of medical engineering & technology , 35 (6-7), 300-307. Guo, K., & Li, D. (2024). Understanding EFL students’ use of self-made AI chatbots as personalized writing assistance tools: A mixed methods study. System , 124 , 103362. Harley, J. M. (2016). Measuring emotions: A survey of cutting edge methodologies used in computer-based learning environment research. Emotions, technology, design, and learning , 89-114. Harris-Watson, A. M., Larson, L. E., Lauharatanahirun, N., DeChurch, L. A., & Contractor, N. S. (2023). Social perception in Human-AI teams: Warmth and competence predict receptivity to AI teammates. Computers in Human Behavior , 145 , 107765. Hashempour, Z., Rostampour, M., & Behjat, F. (2015). The effect of brainstorming as a pre-writing strategy on EFL advanced learners’ writing ability. Journal of Applied Linguistics and Language Research , 2 (1), 86-99. Hayashi, N., Someya, N., Endo, M. Y., Miura, A., & Fukuba, Y. (2006). Vasoconstriction and blood flow responses in visceral arteries to mental task in humans. Experimental physiology , 91 (1), 215-220. Heine, T., Lenis, G., Reichensperger, P., Beran, T., Doessel, O., & Deml, B. (2017). Electrocardiographic features for the measurement of drivers' mental workload. Applied ergonomics , 61 , 31-43. Horvers, A., Tombeng, N., Bosse, T., Lazonder, A. W., & Molenaar, I. (2021). Detecting emotions through electrodermal activity in learning contexts: A systematic review. Sensors , 21 (23), 7869. Jeon, J., & Lee, S. (2024). Can learners benefit from chatbots instead of humans? A systematic review of human-chatbot comparison research in language education. Education and Information Technologies , 29 (17), 23329-23360. Karanjakwut, C., & Charunsri, K. (2025). Transforming AI Chatbots for a Brainstorming Teaching Technique of Process Writing. Malaysian Online Journal of Educational Technology , 13 (1), 1-18. Kellogg, R. T. (1990). Effectiveness of prewriting strategies as a function of task demands. The American Journal of Psychology , 327-342. Ketonen, E. E., Salonen, V., Lonka, K., & Salmela-Aro, K. (2023). Can you feel the excitement? Physiological correlates of students' self-reported emotions. British Journal of Educational Psychology , 93 , 113-129. Kreibig, S. D. (2010). Autonomic nervous system activity in emotion: A review. Biological psychology , 84 (3), 394-421. Labadze, L., Grigolia, M., & Machaidze, L. (2023). Role of AI chatbots in education: systematic literature review. International journal of Educational Technology in Higher education , 20 (1), 56. Lajoie, S. P. (2014). Multimedia learning of cognitive processes. In R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning (2nd ed., pp. 623–646). Cambridge University Press. https://doi.org/10.1017/CBO9781139547369.031. Lee, U., Jeong, Y., Koh, J., Byun, G., Lee, Y., Hwang, Y., ... & Lim, C. (2024). Can ChatGPT be a debate partner? Developing ChatGPT-based application “DEBO” for debate education, findings and limitations. Educational Technology & Society , 27 (2), 321-346. Li, L., Gow, A. D. I., & Zhou, J. (2020). The role of positive emotions in education: A neuroscience perspective. Mind, Brain, and Education , 14 (3), 220-234. Lim, F. V. (2020). Designing learning with embodied teaching: Perspectives from multimodality . Routledge. Loderer, K., Pekrun, R., & Lester, J. C. (2020). Beyond cold technology: A systematic review and meta-analysis on emotions in technology-based learning environments. Learning and instruction , 70 , 101162. Mahapatra, S. (2024). Impact of ChatGPT on ESL students’ academic writing skills: A mixed methods intervention study. Smart Learning Environments , 11 (1), 9. Malmberg, J., Haataja, E., Seppänen, T., & Järvelä, S. (2019). Are we together or not? The temporal interplay of monitoring, physiological arousal and physiological synchrony during a collaborative exam. International Journal of Computer-Supported Collaborative Learning , 14 (4), 467-490. Mayer, R. E. (2020). Searching for the role of emotions in e-learning. Learning and Instruction , 70 , 101213. McNeal, K. S., Zhong, M., Soltis, N. A., Doukopoulos, L., Johnson, E. T., Courtney, S., ... & Porch, M. (2020). Biosensors show promise as a measure of student engagement in a large introductory biology course. CBE—Life Sciences Education , 19 (4), ar50. Mehigan, T., & Pitt, I. (2019). Engaging learners through emotion in artificially intelligent environments. In EDULEARN19 Proceedings (pp. 5661-5668). IATED. Nugraha, D., & Indihadi, D. (2019, October). Brainstorming in scientific writing in elementary school. In Journal of Physics: Conference Series (Vol. 1318, No. 1, p. 012026). IOP Publishing. O'Mealia, S. (2011). How Can Prewriting Strategies Benefit Students?. Education Masters. Paper 14. https://fisherpub.sjf.edu/education_ETD_masters/14 Oudeyer, P. Y., Gottlieb, J., & Lopes, M. (2016). Intrinsic motivation, curiosity, and learning: Theory and applications in educational technologies. Progress in brain research , 229 , 257-284. Piaget, J. (2005/1950). The psychology of intelligence . Routledge. Plass, J. L., & Kaplan, U. (2016). Emotional design in digital media for learning. In Emotions, technology, design, and learning (pp. 131-161). Academic Press. Pekrun, R. (2020). Self-report is indispensable to assess students’ learning. Frontline learning research , 8 (3), 185-193. Rao, Z. (2007). Training in brainstorming and developing writing skills. ELT journal , 61 (2), 100-106. Seery, M. D. (2011). Challenge or threat? Cardiovascular indexes of resilience and vulnerability to potential stress in humans. Neuroscience & Biobehavioral Reviews , 35 (7), 1603-1610. Seli, P., Wammes, J. D., Risko, E. F., & Smilek, D. (2016). On the relation between motivation and retention in educational contexts: The role of intentional and unintentional mind wandering. Psychonomic bulletin & review , 23 (4), 1280-1287. Shirvani, M., & Porkar, R. (2021). How do EFL students perceive brainstorming in L2 writing classes. Theory and Practice in Language Studies , 11 (12), 1602-1609. Soltis, N. A., McNeal, K. S., Atkins, R. M., & Maudlin, L. C. (2020). A novel approach to measuring student engagement while using an augmented reality sandbox. Journal of Geography in Higher Education , 44 (4), 512-531. Stefano, G. B., Stefano, J. M., & Esch, T. (2008). Anticipatory stress response: a significant commonality in stress, relaxation, pleasure and love responses. Medical Science Monitor , 14 (2), RA17. Sumithra, M., Buvaneswar, B., Jessica Judith, S., & Punitha, R. (2022). Innovation for better education system using artificial intelligence. Journal of Cognitive Human-Computer Interaction , 2 (1), 19-28. Richardson, M. P., Strange, B. A., & Dolan, R. J. (2004). Encoding of emotional memories depends on amygdala and hippocampus and their interactions. Nature neuroscience , 7 (3), 278-285. Ryan, A. M., Gheen, M. H., & Midgley, C. (1998). Why do some students avoid asking for help? An examination of the interplay among students' academic efficacy, teachers' social–emotional role, and the classroom goal structure. Journal of educational psychology , 90 (3), 528. Tyng, C. M., Amin, H. U., Saad, M. N., & Malik, A. S. (2017). The influences of emotion on learning and memory. Frontiers in psychology , 8 , 235933. Unin, N. (2016). Brainstorming as a Way to Approach Student-centered Learning in the ESL Classroom. Procedia-Social and Behavioral Sciences , 224 , 605-612. Vogel, S., & Schwabe, L. (2016). Learning and memory under stress: implications for the classroom. npj Science of Learning , 1 (1), 1-10. Wang, J., & Fan, W. (2025). The effect of ChatGPT on students’ learning performance, learning perception, and higher-order thinking: insights from a meta-analysis. Humanities and Social Sciences Communications , 12 (1), 1-21. Wang, L., Yu, J., Diao, X., Zhang, Y., Miao, Y., & He, W. (2024). The chain mediating effects of resilience and perceived social support in the relationship between perceived stress and depression in patients with COVID-19. Frontiers in Psychology , 15 , 1400267. Xiao, Y., & Zhi, Y. (2023). An exploratory study of EFL learners’ use of ChatGPT for language learning tasks: Experience and perceptions. Languages , 8 (3), 212. Zhang, Q., Siraj, S. B., & Abdul Razak, R. B. (2025). Effects of AI chatbots on EFL students’ critical thinking skills and intrinsic motivation in argumentative writing. Innovation in Language Learning and Teaching , 1-29. Zimmerman, B. J. (2002). Becoming a self-regulated learner: An overview. Theory into practice , 41 (2), 64-70. Zull, J. E. (2006). Key aspects of how the brain learns. In S. Johnson & K. Taylor (Eds.), The neuroscience of adult learning (pp. 3–9). Additional Declarations The authors declare no competing interests. Supplementary Files Appendix.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7785914","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":524960282,"identity":"1d5ca2de-786e-4ffd-9e24-9836b4ecd797","order_by":0,"name":"Jia'en Yee","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA10lEQVRIiWNgGAWjYBACxmYYi70BSBQwMEgQr4XnAAPDAQMitCCARAKRWpjbmR8+LmA4LMcv+fiY9AcDGznJBuaHj27gdRibsfEMhsPGkrPT0iQOGKQZSzMARXLw+8VMmofhcOKG2zlmQC2HE+cx8LBJ49fC/v03UEv9/ptniNbCY8YM1JJgIMED0TKbCC3F0jwG6YYzzqQlW5wB+kWymYBfDPuPb/zMU2Etz99++OCNigobOYnjzQ8f49XSACINmpGEmPEoBwF5CFVHQNkoGAWjYBSMaAAAYXBCtxEUkf4AAAAASUVORK5CYII=","orcid":"https://orcid.org/0000-0002-9734-7312","institution":"National Institute of Education, Nanyang Technological University, Singapore 637616","correspondingAuthor":true,"prefix":"","firstName":"Jia'en","middleName":"","lastName":"Yee","suffix":""},{"id":524960363,"identity":"0c2cb133-f324-42e3-ad4a-d59b4e4a4eb1","order_by":1,"name":"Fei Victor Lim","email":"","orcid":"https://orcid.org/0000-0003-3046-1011","institution":"National Institute of Education, Nanyang Technological University, Singapore 637616","correspondingAuthor":false,"prefix":"","firstName":"Fei","middleName":"Victor","lastName":"Lim","suffix":""},{"id":524960364,"identity":"1822c680-bdc0-4884-a75a-bc21ceb7f116","order_by":2,"name":"Jerrold Quek","email":"","orcid":"","institution":"Nanyang Technological University, Singapore 639798","correspondingAuthor":false,"prefix":"","firstName":"Jerrold","middleName":"","lastName":"Quek","suffix":""}],"badges":[],"createdAt":"2025-10-05 16:47:58","currentVersionCode":1,"declarations":{"humanSubjects":true,"vertebrateSubjects":false,"conflictsOfInterestStatement":false,"humanSubjectEthicalGuidelines":true,"humanSubjectConsent":true,"humanSubjectClinicalTrial":false,"humanSubjectCaseReport":false,"vertebrateSubjectEthicalGuidelines":false},"doi":"10.21203/rs.3.rs-7785914/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7785914/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":93024460,"identity":"148e9842-e9a5-4d5c-9370-ad45bb4bac2a","added_by":"auto","created_at":"2025-10-08 09:16:57","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":165300,"visible":true,"origin":"","legend":"","description":"","filename":"ManuscriptMoodsBotsandBodiesPreprint.docx","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/6f09c578163472159911b36c.docx"},{"id":93024459,"identity":"70e8a405-6dc7-4f8e-ba08-24192f1d8204","added_by":"auto","created_at":"2025-10-08 09:16:57","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":342,"visible":true,"origin":"","legend":"","description":"","filename":"rs7785914.json","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/d4fee57de549585519faed0e.json"},{"id":93024463,"identity":"2b26b2c4-ebed-4016-be60-56eca68f568a","added_by":"auto","created_at":"2025-10-08 09:16:57","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":218035,"visible":true,"origin":"","legend":"","description":"","filename":"rs77859140enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/5dcca622716d3a36c0f81220.xml"},{"id":93025875,"identity":"bba9594b-bd8b-4507-9ea2-3062e863a832","added_by":"auto","created_at":"2025-10-08 09:24:57","extension":"png","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":13591,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/132e579a39be70d7654da030.png"},{"id":93024462,"identity":"5c1d374a-32e7-46f1-b8c5-183fc7148567","added_by":"auto","created_at":"2025-10-08 09:16:57","extension":"jpeg","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":119431,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/48c562099d89518685c960c8.jpeg"},{"id":93025878,"identity":"c1ffd457-2ec6-43fd-83d9-098cbda44efe","added_by":"auto","created_at":"2025-10-08 09:24:57","extension":"png","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5843,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/7833bd47acba3ac95a8308f2.png"},{"id":93025879,"identity":"ab9923f5-e162-403c-bf78-5575e5718458","added_by":"auto","created_at":"2025-10-08 09:24:57","extension":"png","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":5083,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/69440657d010934af46c7cc2.png"},{"id":93024466,"identity":"c6c5091a-f7a9-4728-af5e-fb6badf73248","added_by":"auto","created_at":"2025-10-08 09:16:57","extension":"png","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":38280,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/6590f3e827cf052711a2122d.png"},{"id":93024464,"identity":"7e43c415-5b37-4a4d-b880-e8afc715d3c0","added_by":"auto","created_at":"2025-10-08 09:16:57","extension":"png","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6551,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/71789c5017d240c865071b56.png"},{"id":93025880,"identity":"ad96d235-52bf-46a4-bdb6-78f80b839308","added_by":"auto","created_at":"2025-10-08 09:24:58","extension":"xml","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":214246,"visible":true,"origin":"","legend":"","description":"","filename":"rs77859140structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/923d1853944024b304fed9b1.xml"},{"id":93024467,"identity":"ff847e39-fa92-404b-a92f-52048790295d","added_by":"auto","created_at":"2025-10-08 09:16:57","extension":"html","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":229035,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/5664843c85ca370a0ed78a75.html"},{"id":93024455,"identity":"5fbe928b-0d19-4f84-8a9f-cb982a5e3d4f","added_by":"auto","created_at":"2025-10-08 09:16:57","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":75555,"visible":true,"origin":"","legend":"\u003cp\u003ePulse by Condition (raw points and adjusted means) during brainstorming\u003c/p\u003e","description":"","filename":"Figure1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/93f2b708c8a7e8b877ac326f.jpg"},{"id":93025877,"identity":"90c15051-1112-4a84-bb84-383a8da45188","added_by":"auto","created_at":"2025-10-08 09:24:57","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":83927,"visible":true,"origin":"","legend":"\u003cp\u003ePulse AUC by Condition during Brainstorming (raw points and adjusted means)\u003c/p\u003e","description":"","filename":"Figure2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/057ea8693ecbe16613d4f3fc.jpg"},{"id":93026615,"identity":"9dd12879-dc57-4ac8-b376-312070dd44ac","added_by":"auto","created_at":"2025-10-08 09:32:57","extension":"jpg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":67008,"visible":true,"origin":"","legend":"\u003cp\u003eTemperature AUC by Condition during Brainstorming (raw points and adjusted means)\u003c/p\u003e","description":"","filename":"Figure3.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/e303e2aa49191aa34a4f676c.jpg"},{"id":93027643,"identity":"72fa13fb-f7bb-40db-a502-3f5b1b4ec711","added_by":"auto","created_at":"2025-10-08 09:40:58","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1256134,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/966f5c1c-1ea4-4bf0-8ab4-11e42596cff5.pdf"},{"id":93024454,"identity":"fbc34e2f-41ff-4987-884e-e7ad9e2659f1","added_by":"auto","created_at":"2025-10-08 09:16:57","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":34495,"visible":true,"origin":"","legend":"","description":"","filename":"Appendix.docx","url":"https://assets-eu.researchsquare.com/files/rs-7785914/v1/e11ad93181ad806cb192d106.docx"}],"financialInterests":"The authors declare no competing interests.","formattedTitle":"\u003cp\u003eMoods, Bots, and Bodies: University Students’ Emotional and Physiological responses to Human vs. GenAI Chatbots\u003c/p\u003e","fulltext":[{"header":"1. Introduction","content":"\u003cp\u003eAs the education landscape rapidly evolves with the integration of generative AI (Gen AI), opportunities and challenges abound, particularly in its potential to transform traditional teaching methods. There is now a sharp rise in AI-based education applications; for example, educational games, adaptive learning platforms, AI chatbots serving as virtual teaching assistants, personalised tutors, and interactive learning companions across diverse education levels all over the world. AI systems can personalise learning content and experience to match one\u0026rsquo;s pace, needs and abilities regardless of time or geography, and provide feedback (e.g., Labadze, Grigolia, \u0026amp; Machaidze, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2023\u003c/span\u003e), making it increasingly accessible and personalised.\u003c/p\u003e\u003cp\u003eEmerging research underscores the potential of Gen AI tools in language and literacy education. Chandel and Lim (\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), in a systematic review of empirical studies, examine how GenAI applications such as AI-powered writing assistants, and conversational chatbots are being integrated into classroom practice to support literacy development across reading, writing, and multimodal meaning-making. The review highlights the affordances of GenAI in facilitating ideation and planning, with studies by Lee et al. (\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), Mahapatra (\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), and Xiao and Zhi (\u003cspan citationid=\"CR71\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) demonstrating the benefits of such tools in supporting learners during the brainstorming phase of writing. However, there is limited literature on students\u0026rsquo; receptivity to using Gen AI for learning or help-seeking compared to human educators, especially when explored through experimental methods rather than interviews.\u003c/p\u003e\u003cp\u003eWhile AI offers a multitude of advantages, aspects unique to Gen AI can potentially impede learning. For example, in a study consisting of students learning a second language through virtual human interactions, students\u0026rsquo; levels of frustration were modulated by factors such as \u0026ldquo;not being understood or heard as expected\u0026rdquo; (Ericsson et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). This is a common trend when engaging with AI where nuanced human dynamics are missing. Other features of GenAI, such as perceived warmth and emotional responsiveness, have been shown to affect individuals\u0026rsquo; attitudes towards having AI as a teammate (Harris-Watson et al., \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Specifically, perceived warmth and competence positively predict individuals\u0026rsquo; receptivity to AI in three aspects, including their willingness to adapt routines, integrate AI's expertise and skills and regard AI as a valued teammate. Even though individuals may appreciate the efficiency or flexibility of AI systems, the absence of \u0026ldquo;human\u0026rdquo; qualities like empathy, adaptability, encouragement, personal connection and trustworthiness in AI bots can affect their satisfaction levels and how willing they are to engage and rely on them for knowledge building (e.g., Cevher \u0026amp; Yıldırım, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Demeure et al., 2022).\u003c/p\u003e\u003cp\u003eGiven the established link between emotional arousal, attention and learning effectiveness (Loderer et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), along with the increasing integration of AI in education, it is crucial to deepen our understanding of how students perceive and interact with Gen AI compared to human educators. Such insights can inform the development of blended learning environments that optimise the strengths of both AI and human educators, ensuring that AI serves as a pedagogical aide for the teacher.\u003c/p\u003e\u003cp\u003eThis proof-of-concept study investigates how students\u0026rsquo; emotional and physiological responses differ when brainstorming with a Gen AI chatbot versus a human teacher. We focus on self-rated mood and physiological arousal as indicators of interest and engagement, aiming to identify the potential reasons for receptivity or resistance towards the use of Gen AI for learning, and how each can be leveraged to support learning. The study addresses three questions: 1) How do students\u0026rsquo; emotional responses differ between chatbot- and human-facilitated brainstorming? 2) How do students\u0026rsquo; physiological responses differ between chatbot- and human-facilitated brainstorming? 3) How are students\u0026rsquo; physiological signals related to their self-reported moods, regardless of condition?\u003c/p\u003e"},{"header":"2. Literature Review","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003e2.1. Brainstorming and facilitation for writing\u003c/h2\u003e\u003cp\u003eBrainstorming is widely recognised as an effective pre-writing strategy that enhances idea generation, organisation, and refinement, leading to higher-quality essays with greater originality and depth (Crossley, Muldner, \u0026amp; McNamara, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). In externalising their thoughts during brainstorming, writers engage in metacognitive reflection, structuring ideas more clearly, drawing on prior knowledge, considering various perspectives, and approaching writing more fluently (Kellog, 1990; O'Mealia, \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). Across age groups, language backgrounds, and abilities, structured planning before essay writing has been shown to improve clarity and overall quality (Amoush, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Rao, \u003cspan citationid=\"CR57\" class=\"CitationRef\"\u003e2007\u003c/span\u003e; Nugraha \u0026amp; Indihadi, \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2019\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eBrainstorming is especially valuable for students who struggle with idea expression, confidence, lexical retrieval, or adherence to academic writing conventions (Abedianpour \u0026amp; Omidvari, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Hashempour, Rostampour \u0026amp; Behjat, \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). For example, for English as a foreign language (EFL) learners, brainstorming provides cognitive and linguistic scaffolding as well as emotional support, helping to reduce writing anxiety, build confidence, and strengthen their ability to generate, organise and articulate ideas more effectively (Unin, \u003cspan citationid=\"CR67\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Shirvani and Porkar, \u003cspan citationid=\"CR60\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe cognitive and emotional benefits of brainstorming can be further amplified in collaborative settings, where social interaction, peer feedback and co-construction of meaning help students engage more actively with their ideas revise their thinking in real time. Ideation with others encourages critical thinking, creativity, and the exchange of alternative perspectives, helping students to build richer and more multidimensional ideas (Al-Khatib, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Ghabanchi \u0026amp; Behrooznia, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). However, human-facilitated brainstorming may not always be accessible. Increasingly, Gen AI is filling this gap, offering personalised, on-demand brainstorming chatbots that can simulate aspects of effective human-led support. These systems can be designed to possess in-depth domain knowledge, ask probing questions, and respond adaptively to student input. Critically, they also offer a \u0026ldquo;judgment-free\u0026rdquo; space \u0026ndash; an important feature for students who may hesitate to seek help due to fear of negative evaluation (Downing et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Ryan et al., \u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e1998\u003c/span\u003e). By lowering social barriers, such chatbots reduce anxiety, encourage intellectual risk-taking, and promote deeper inquiry and self-regulated learning (Zimmerman, \u003cspan citationid=\"CR73\" class=\"CitationRef\"\u003e2002\u003c/span\u003e).\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e\u003ch2\u003e2.2. Chatbots in educational support\u003c/h2\u003e\u003cp\u003eRecent studies illustrate both the potential and complexity of chatbot-facilitated brainstorming. In a quasi-experimental study, Zhang et al. (\u003cspan citationid=\"CR72\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) found that students using an AI chatbot (Spark Desk) during argumentative writing showed significant gains in critical thinking skills and intrinsic motivation, including higher enjoyment, greater perceived value of the task and reduced stress, as compared to peers engaged in peer interaction. Similarly, Guo and Li (\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) showed that students who built their own chatbots for idea generation, outlining and language correction, developed clearer goals, greater writing confidence, and more positive attitudes toward writing. These chatbots also handled diverse requests, including \u0026lsquo;assistance, customisation, and translation\u0026rsquo;, which would have been overwhelming for a single human teacher to address for multiple students.\u003c/p\u003e\u003cp\u003eStill, chatbot-led ideation has drawbacks. AI-generated brainstorming can be less diverse, with high overlap across responses, compared to human brainstorming (e.g., studies report 94% of ideas overlapping). Educators also express reservations: Karanjakwut and Charunsri (\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) found that while Thai university students using AI chatbot tools outperformed a control group on several assignments, lecturers still preferred student-generated ideas over chatbot suggestions. Meta-analytic evidence provides a more balanced view \u0026ndash; Wang and Fan\u0026rsquo;s (\u003cspan citationid=\"CR69\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) review of 51 studies concluded that ChatGPT use has a large positive impact on learning performance, and a moderately positive effect on higher-order thinking and perceived learning.\u003c/p\u003e\u003cp\u003eTogether, these studies suggest that, when thoughtfully integrated, chatbots can serve as flexible, personalised brainstorming partners, and that their influence on students\u0026rsquo; engagement and confidence may differ from that of human teachers. Yet most existing work has focused narrowly on linguistic outcomes, without examining students\u0026rsquo; emotional responses, let alone make comparisons to experiences with human teachers (Jeon \u0026amp; Lee, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). This study addresses that gap by examining learners\u0026rsquo; emotional and physiological responses, to explore how human and AI tools shape the brainstorming experience.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec5\" class=\"Section2\"\u003e\u003ch2\u003e2.3. Emotion and learning\u003c/h2\u003e\u003cp\u003e\u0026ldquo;Emotion is the foundation of learning\u0026rdquo; (Zull \u003cspan citationid=\"CR74\" class=\"CitationRef\"\u003e2006\u003c/span\u003e, p. 7). Traditionally viewed as a cognitive process, learning is now widely recognised as deeply influenced by motivation and emotional processes, including what is learned and retained (Lajoie, \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2014\u003c/span\u003e; Plass \u0026amp; Kaplan, \u003cspan citationid=\"CR55\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Seli et al., \u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Emotions direct attention, motivation and affect memory (Mayer, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), and are integral to the causal chain leading to learning outcomes, particularly in digital environments. Learning results not only from instructional design and content, but also from learners\u0026rsquo; emotional responses to the process. Emotions, short-lived, intense reactions to specific events or stimuli, can either facilitate or hinder learning, depending on their nature and timing (Duffy et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). Understanding how learners feel is therefore critical for designing more effective learning experiences (Boekaerts, \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2010\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eEmpirical studies support the influence of specific emotional states on cognitive performance. Emotional content is remembered more clearly than neutral content, as emotional experiences engage both the amygdala and hippocampus during memory formation, enhancing the consolidation of emotionally significant information and leading to stronger long-term recall (Richardson et al., \u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e2004\u003c/span\u003e; Tyng et al., \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Positive emotions like enjoyment, pride and hope are strong drivers of learning motivation, academic performance (Pekrun et al., 2002), self-regulation, cognitive flexibility and problem solving (Li et al., \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). However, not all emotions affect learning in the same way. For example, stress can either support or hinder learning depending on its intensity and duration (Vogel \u0026amp; Schwabe, \u003cspan citationid=\"CR68\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Mild, acute stress may enhance attention and memory, while chronic or excessive stress tends to impair cognitive performance.\u003c/p\u003e\u003cp\u003eImportantly, not only positive emotions contribute to effective learning. Certain negative states like confusion, can improve learning outcomes by increasing focus on the learning content (D\u0026rsquo;Mello et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2014\u003c/span\u003e). Positive emotions like curiosity promote exploration and prepare the brain for learning and material retention in both children and adults (Oudeyer et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). Together, these studies demonstrate that both positive and negative emotional experiences can facilitate learning under the right conditions, highlighting the importance of how emotional states influence attention, engagement and memory in educational contexts.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec6\" class=\"Section2\"\u003e\u003ch2\u003e2.4. AI support for emotional engagement in learning\u003c/h2\u003e\u003cp\u003eGiven that learning is not merely a cognitive process, but also about how learners feel, react and regulate during the process, as educational research increasingly recognises the role of emotion in shaping attention, motivation and memory, there is growing interest in how AI-based systems can support not just what students learn, but how they experience learning emotionally.\u003c/p\u003e\u003cp\u003eRecent advances in emotion detection technologies have enabled AI tools to predict and identify emotions that impede learning, such as boredom and frustration, by analysing cues in their chatlogs or interaction patterns. Adaptive systems can then respond to students\u0026rsquo; emotions and behaviour dynamically to increase motivation and productivity (Arguel et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; D\u0026rsquo;Mello \u0026amp; Graesser, 2012; Mehigan \u0026amp; Pitt, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). For example, Sumithra and colleagues (\u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e2022\u003c/span\u003e) demonstrated how by using emotion-detection algorithms, AI systems can adapt lesson content and pace to match students\u0026rsquo; emotional intensity and attention levels, while ensuring timely completion. In this sense, AI chatbots can mirror a human tutor\u0026rsquo;s responsiveness to individual students, while remaining consistent, unbiased and available on demand. Emerging models can even simulate social-emotional intelligence (Gorga \u0026amp; Schneider, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2009\u003c/span\u003e), thus providing more responsive, emotionally supportive digital learning environments that begin to rival the empathy, adaptability and relational support traditionally offered by human teachers. As Gen AI tools become more emotionally intelligent, integrating physiological data offers a promising next step \u0026ndash; providing real time, complementary insight into learners\u0026rsquo; emotional states.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec7\" class=\"Section2\"\u003e\u003ch2\u003e2.5. Embodiment and the physiological measurement of emotion\u003c/h2\u003e\u003cp\u003eEmotions are rooted in the body as much as the mind, involving complex interactions between physiological and neural processes. Theories of interoception and embodied cognition suggest that emotions are shaped not only by subjective interpretations, but also by the perception and regulation of internal bodily states (Barrett \u0026amp; Simmons, \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). Relatedly, emotions reflect the body\u0026rsquo;s efforts to maintain homeostasis \u0026ndash; its internal equilibrium. This balance is disrupted in times of stress or discomfort, and individuals at these moments, tend to experience negative emotional states. Contrastingly, restoring this balance is linked to positive feelings such as calmness (Craig, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2002\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eIn educational settings, positive emotional states like interest, curiosity and focus support deeper cognitive processing and sustained attention (Tyng et al., \u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). While these experiences are often reported subjectively, they also have physiological signatures, such as changes in skin temperature, heart rate, and electrodermal activity (EDA) (Malmberg et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), which reflect shifts in arousal, attention, and emotional reactivity.\u003c/p\u003e\u003cp\u003eHowever, while self-report questionnaires remain useful for assessing learning and emotion, they are subjected to biases (e.g., social desirability, limited recall, lack of awareness), and do not capture moment-to-moment or subconscious changes in emotional state (Pekrun, \u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). To address these challenges, researchers increasingly adopt multimodal techniques that combine self-reports with objective indicators of emotional arousal and engagement (Ketonen et al., 2022), including classroom observations, eye-tracking, learning analytics, and physiological signals (e.g., McNeal et al. \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). EDA, commonly used in studies assessing student engagement (Horvers et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), tracks changes in skin conductance driven by the sympathetic nervous system and is sensitive to shifts in emotional intensity and attention (Boucsein, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2011\u003c/span\u003e). Heart rate provide insight into emotional valence and cognitive effort; higher heart rate for instance, is associated with increased cognitive workload (Darnell \u0026amp; Krieg, \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2019\u003c/span\u003e; Heine et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Skin temperature typically drops during sympathetic activation (e.g., anxiety) due to vasoconstriction, and rises during relaxed or regulated states (Gouzi et al., 2011). Because physiological data and self-report tap into different aspects of experience, they do not always align. Self-reports tend to offer summarised or retrospective accounts of how learners felt, whereas physiological data capture the immediate and embodied dimensions of emotional responses that may be fleeting, subtle or subconscious (e.g., Ketonen et al., 2019).\u003c/p\u003e\u003cp\u003eCapturing the embodied dimensions of emotional dynamics allows researchers to move beyond what learners say they feel, to observe how their bodies respond to specific instructional events (Harley, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). This is especially useful when comparing interactions with human instructors versus Gen AI systems, where differences in emotional and cognitive engagement may not be fully verbalised but can still be detected physiologically.\u003c/p\u003e\u003c/div\u003e"},{"header":"3. Methods","content":"\u003cp\u003eEthics approval from Nanyang Technological University's institutional review board (IRB) has been obtained prior to the commencement of the study (IRB-2024-1031). This study employed a multimodal design to investigate learners\u0026rsquo; experiences with AI- versus human-facilitated brainstorming during an essay planning task. Participants first completed a pre-task mood questionnaire, then engaged in a 25-minute brainstorming session with either a human or chatbot facilitator. Immediately afterward, they completed a post-task mood questionnaire, and wrote an essay outline. Throughout the brainstorming phase, continuous physiological signals were recorded using a wearable sensor, enabling assessment of not only how learners \u003cem\u003ereport\u003c/em\u003e feeling, but also how their bodies respond, thus offering alternative perspective into the embodied dimensions of human-AI interaction.\u003c/p\u003e\u003cp\u003eThe essay outline writing activity served two primary purposes. Firstly, it provided a concrete context and goal, helping to anchor participants\u0026rsquo; sense of purpose and engagement during the brainstorming process. Secondly, it offered an objective behavioural outcome measure of brainstorming effectiveness of each facilitation condition. However, because this paper focuses on emotional and physiological responses, analyses of writing performance fall outside the current scope and will be reported separately.\u003c/p\u003e\u003cp\u003eWe recruited 30 undergraduates and postgraduates, as they are likely to engage in learning activities with both AI and the human facilitator. Participants were randomly assigned to one of two conditions: 1) AI Chatbot Condition (n\u0026thinsp;=\u0026thinsp;15; 11 female; \u003cem\u003eM\u003c/em\u003e age\u0026thinsp;=\u0026thinsp;28.6, \u003cem\u003eSD\u003c/em\u003e\u0026thinsp;=\u0026thinsp;6.92) : Participants brainstormed ideas with the assistance of a customised messaging chatbot; 2) Human Teacher Condition (n\u0026thinsp;=\u0026thinsp;15; 11 female; \u003cem\u003eM\u003c/em\u003e age\u0026thinsp;=\u0026thinsp;29.2, \u003cem\u003eSD\u003c/em\u003e\u0026thinsp;=\u0026thinsp;5.15): Participants engaged in the same brainstorming task with the help of a human teacher via the chat function on Zoom, without the audio and video feature turned on. The identities of the participants in the human teacher condition were not revealed to the human teacher and vice versa.\u003c/p\u003e\u003cp\u003eThe brainstorming task centred on the essay question, \u0026lsquo;Should Singapore embrace Singlish as a key part of its national identity, or does it undermine its global image?\u0026rsquo; Participants were instructed to develop an outline for an argumentative essay in response to this prompt after the brainstorming session. The essay outlines were then graded according to a rubric developed.\u003c/p\u003e\u003cp\u003eThe Gen AI chatbot used in the study was developed using SchoolAI platform (app.schoolai.com), which enables the customisation of chatbots using tailored prompts and supplementary materials. For this study, the chatbot\u0026rsquo;s knowledge base was augmented with selected readings on Singapore English and Singlish. Unlike typical chatbots that serve to provide direct and complete answers, the design strategy adopted for our chatbot followed a Socratic questioning approach \u0026ndash; posing reflective, open-ended questions to encourage critical thinking and avoiding giving participants direct answers. The specific system prompts used can be found in Appendix A.\u003c/p\u003e\u003cp\u003eThe human teacher engaged in the study is a lecturer in a reputable university in Singapore with teaching experience in Academic Writing, Professional Communication, Linguistics and English Language across various tertiary institutions. Their academic specialisation includes English in Singapore, providing a strong match to the chatbot\u0026rsquo;s domain expertise.\u003c/p\u003e\u003cp\u003eBefore brainstorming, participants completed a baseline mood questionnaire and a 5-minute control condition viewing a neutral video (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://www.youtube.com/watch?v=zpHOIlQbj8Y\u003c/span\u003e\u003cspan address=\"https://www.youtube.com/watch?v=zpHOIlQbj8Y\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) to elicit a stable, low-arousal state. These baseline readings served as emotional and physiological baselines, accounting for individual differences in resting states and enabling more accurate condition comparisons. Participants were explicitly informed in advance whether they would be interacting with a chatbot or a human teacher to mimic real-world learning scenarios, where individuals are typically aware of the nature of their learning facilitators. By doing so, we aimed to enhance the ecological validity of the study as it reflects more accurately the variety of help-seeking behaviours and communication experiences participants may encounter outside the study setting, while avoiding introducing confusion or expectancy effects during the task. In particular, since it is highly plausible that participants in the chatbot condition might infer the identity of conversational partner over time, we sought to prevent any potential disruptions in emotional or physiological responses caused by such realisations.\u003c/p\u003e\u003cp\u003eFollowing the brainstorming phase, participants completed a follow-up questionnaire containing the same mood items found in the pre-brainstorming questionnaire as well as additional 15 questions on perceptions of the brainstorming interaction. These self-reported measures not only allowed for direct comparison to their baseline mood, but also served to contextualise and supplement the physiological data, providing a more comprehensive view and nuanced analysis of the impact of the two experimental conditions (AI vs human).\u003c/p\u003e\u003cp\u003eThe mood questionnaire contained 20 items (10 positive, 10 negative; e.g., excited, stressed, frustrated), the intensity of each was rated on a 5-point Likert scale from \u0026ldquo;None\u0026rdquo; to \u0026ldquo;Very intense\u0026rdquo;. These ratings were used to assess mood change over time and compare emotional outcomes across the conditions (human vs chatbot). The second questionnaire\u0026rsquo;s additional 15 perception questions covered aspects such as effectiveness and ease of communication. See Appendix B for the full list of items.\u003c/p\u003e\u003cp\u003ePhysiological responses were collected using the EmbracePlus wearable sensor, which records continuous, real-time, non-invasive biomarkers relevant to emotional and cognitive engagement. Specifically, we measured heart rate, EDA, and skin temperature. These measures were selected based on prior research linking them as proxies for affective arousal, attentional focus, and engagement in learning contexts (Fang et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2018\u003c/span\u003e; Soltis et al., \u003cspan citationid=\"CR61\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e\u003ch2\u003e3.1. Data analyses\u003c/h2\u003e\u003cp\u003eAll analyses were conducted in R (version 4.3.1; R Core Team, 2025) using descriptive statistics, MANCOVA, regression, and mixed-effects models to examine physiological and emotional responses during the brainstorming task. Analyses addressed: (1) group differences between the Chatbot and Human teacher conditions, and (2) associations between physiological activation and mood outcomes.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eMood Analyses\u003c/strong\u003e\u003cp\u003eMean mood ratings were first compared descriptively at baseline and brainstorming phases. However, analysing raw mood ratings at each phase separately risks misinterpretation. For instance, a higher brainstorming phase mood rating in the human condition does not necessarily mean the chatbot group showed no improvement; it simply reflects an endpoint comparison without accounting for each participant\u0026rsquo;s baseline.\u003c/p\u003e\u003c/p\u003e\u003cp\u003eTo more accurately capture emotional shifts, mood change scores were computed for each item by subtracting the baseline rating from the brainstorming rating (Brainstorming \u0026ndash; Baseline). Positive change scores indicate increased mood intensity during brainstorming (Brainstorm\u0026thinsp;\u0026gt;\u0026thinsp;Baseline), negative scores reflect decreases (Brainstorming\u0026thinsp;\u0026lt;\u0026thinsp;Baseline). This approach 1) implicitly controls for baseline differences without adding 20 baseline mood covariates alongside Gender and Age; and 2) preserves statistical power, which is crucial given the relatively small sample size and the risk of overfitting in multivariate models.\u003c/p\u003e\u003cp\u003eChange scores were entered into a MANCOVA to assess overall condition effects, controlling for gender and age. Significant or marginal multivariate effects were followed up with univariate ANCOVAs and regressions to identify specific moods differing by condition (Chatbot vs. Human).\u003c/p\u003e\u003cp\u003e\u003cb\u003ePhysiological Analyses\u003c/b\u003e: Physiological data (heart rate, EDA, temperature) were analysed separately for the baseline (5 min; 5 data points) and brainstorming (25 min; 25 data points) phases. For each phase, two types of metrics were used: raw means and Area Under the Curve (AUC) values. AUCs were calculated using the trapezoidal method to capture sustained physiological engagement across each phase (5-min baseline, 25-min brainstorming). AUC reflects total physiological activation over time and is useful for summarising prolonged states such as engagement, arousal and stress. One participant contributed only 4 baseline data points, and thus their baseline AUC could not be computed. This participant was excluded from analyses involving baseline AUC.\u003c/p\u003e\u003cp\u003eAll physiological measures (raw means and AUCs) were analysed using the same statistical approach. For the baseline phase, MANCOVAs compared groups (chatbot vs human) with age and gender as covariates. Significant multivariate effects were followed by univariate ANCOVAs. For the brainstorming phase, MANCOVAs tested group differences while controlling for all corresponding baseline measures, age and gender. This approach allowed the examination of group differences in the overall physiological profile during brainstorming while adjusting for baseline levels. Significant multivariate effects were followed by univariate ANCOVAs, and covariate-adjusted means with 95% confidence intervals were extracted to determine the direction of effects.\u003c/p\u003e\u003cp\u003eTo analyse minute-by-minute trajectories of physiological change during brainstorming, linear mixed-effects models (using the lmer function in R) were fitted. This approach was chosen because it accommodates repeated measures data, allowing us to model both between-subject differences and within-subjects changes over time. The models included fixed effects for Time, Condition and their interaction, as well as random intercepts by participant to account for individual variability (e.g., each participant may have inherently different physiological levels at baseline). The random effects structure was constructed following the recommendations of Bates et al. (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2015\u003c/span\u003e), prioritising an accurate reflection of the experimental design while also maintaining model parsimony. This growth curve modelling approach allowed us to assess whether the trajectories of physiological responses differed by condition across the 25-minute brainstorming session.\u003c/p\u003e\u003cp\u003e\u003cstrong\u003eIntegration of Mood and Physiological Data\u003c/strong\u003e\u003cp\u003eTo explore the relationship between physiological responses and mood changes, independent of experimental condition, multiple regression analyses were conducted in which mood change scores (both composite positive/negative mood scores and individual mood items) were predicted from physiological AUC values. This approach assessed the predictive value of physiological responses on subjective mood experiences, regardless of condition.\u003c/p\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"4. Findings","content":"\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\n \u003ch2\u003e4.1. Mood\u003c/h2\u003e\n \u003cp\u003e\u003cem\u003eDescriptive statistics\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003eAt baseline, participants in the chatbot condition reported higher levels of positive emotions, while those in the human condition reported higher levels of negative emotions. However, this pattern reversed post-brainstorming: chatbot participants experienced more intense negative moods, whereas the human condition saw greater increases in positive emotions.\u003c/p\u003e\n \u003cp\u003eDescriptive statistics and mood change scores (brainstorm \u0026ndash; baseline) are presented in Appendix C Table \u003cspan class=\"InternalRef\"\u003eC1\u003c/span\u003e. Both groups showed increases in positive moods 7 of 10 moods, with larger gains in the human teacher group. Three positive moods saw chatbot-only decreases \u0026ndash; Curious, Motivated and Excited. The chatbot group also showed increases in negative moods, particularly Stress, Discouragement, Boredom and Distraction.\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eInferential statistics for baseline mood and mood change scores\u003c/em\u003e\u003c/p\u003e\n \u003cp\u003eAt baseline, a MANCOVA controlling for gender and age revealed no significant multivariate effect of Condition (Pillai\u0026rsquo;s Trace\u0026thinsp;=\u0026thinsp;0.60, \u003cem\u003eF\u003c/em\u003e(20, 6)\u0026thinsp;=\u0026thinsp;0.45, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.92), indicating that both groups began the brainstorming session with comparable mood states. For mood change scores (Brainstorming mood intensity \u0026ndash; Baseline mood intensity), a MANCOVA controlling for gender and age revealed marginal divergent emotional shifts between conditions (\u003cem\u003ePillai\u0026rsquo;s Trace\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.91, \u003cem\u003eF\u003c/em\u003e(20, 6)\u0026thinsp;=\u0026thinsp;3.20, p\u0026thinsp;=\u0026thinsp;0.077). Univariate ANCOVAs indicated significant Condition effects for Motivated, Inspired, Empowered, Engaged, Stressed, Discouraged, Bored, and Distracted, and marginally significant condition effects for Focused, Excited, Connected, Determined, and Frustrated.\u003c/p\u003e\n \u003cp\u003eFollow-up regressions showed that effects were driven by greater increases in positive moods (Motivated, Inspired, Empowered, and Engaged) for participants in the human teacher group, with marginal gains in Excited and Connected. Conversely, the chatbot group showed significantly greater increases in negative moods \u0026ndash; Stressed, Discouraged, Bored, and Distracted, with marginally greater Frustrated. Focused and Determined did not show significant condition effects in the regression models. Table \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e presents inferential statistics for mood changes scores.\u003c/p\u003e\n \u003ctable id=\"Tab1\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eSignificant and marginal condition effects on mood change scores from ANCOVAs and follow-up regressions\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eMood\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eANCOVA \u003cem\u003eF\u003c/em\u003e(1, 25 ), \u003cem\u003ep\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u0026beta; (SE)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003et(25), \u003cem\u003ep\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\u0026nbsp;\u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eDirection\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eMotivated Change\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e10.76, .003\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.84 (0.26)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.25, .003\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHuman\u0026thinsp;\u0026gt;\u0026thinsp;Chatbot\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eInspired Change\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e8.01, .009\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.93 (0.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.84, .009\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHuman\u0026thinsp;\u0026gt;\u0026thinsp;Chatbot\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEmpowered Change\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e10.86, .003\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.86 (0.26)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.23, .003\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHuman\u0026thinsp;\u0026gt;\u0026thinsp;Chatbot\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eEngaged Change\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e11.70, .002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.00 (0.30)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.37, .002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHuman\u0026thinsp;\u0026gt;\u0026thinsp;Chatbot\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eExcited Change\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.74, .06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.75 (0.49)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.88, .07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026dagger;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHuman\u0026thinsp;\u0026gt;\u0026thinsp;Chatbot\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eConnected Change\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.59, .07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.71 (0.38)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.88, .07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026dagger;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHuman\u0026thinsp;\u0026gt;\u0026thinsp;Chatbot\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eStressed Change\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e12.99, .001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-1.18 (0.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-3.62, .001\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHuman\u0026thinsp;\u0026lt;\u0026thinsp;Chatbot\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDiscouraged Change\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e11.11, .003\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-1.00 (0.29)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-3.41, .002\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHuman\u0026thinsp;\u0026lt;\u0026thinsp;Chatbot\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eBored Change\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e6.65, .02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.81 (0.32)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-2.54, .02\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e*\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHuman\u0026thinsp;\u0026lt;\u0026thinsp;Chatbot\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDistracted Change\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e9.23, .006\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-1.27 (0.41)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-3.06, .005\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e**\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHuman\u0026thinsp;\u0026lt;\u0026thinsp;Chatbot\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFrustrated Change\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.24, .08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.64 (.34)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-1.91, .07\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u0026dagger;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eHuman\u0026thinsp;\u0026lt;\u0026thinsp;Chatbot\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eDetermined Change\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.22, .08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.47 (0.28)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.66, .11\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eFocused Change\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.02, .09\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.51 (0.30)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-0.33, .75\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003ctfoot\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"6\"\u003e\u003cstrong\u003eNote.\u003c/strong\u003e \u003cem\u003ep\u003c/em\u003e values are two-tailed. * denotes \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.05, ** denotes \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.01, \u0026dagger; denotes 05\u0026thinsp;\u0026le;\u0026thinsp;\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.10.\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tfoot\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\n \u003ch2\u003e4.2. Perception of the Brainstorming Session\u003c/h2\u003e\n \u003cp\u003eA MANCOVA was conducted to examine whether perceptions relating to the value and cognitive impact of the brainstorming session differed between the chatbot and human teacher conditions, while controlling for gender and age. The overall multivariate effect of Condition was statistically significant, \u003cem\u003ePillai\u0026rsquo;s Trace\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.82, \u003cem\u003eF\u003c/em\u003e(15, 10)\u0026thinsp;=\u0026thinsp;3.05, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.040, indicating that participants\u0026apos; perceptions of the brainstorming session varied depending on the interaction partner. Specifically, participants who brainstormed with a human teacher reported more positive perceptions of the session across several items. They were more likely to describe their brainstorming partner as knowledgeable, felt more comfortable interacting, more likely to indicate learning something new, and also found it easier to communicate ideas. Marginally, they reported greater enjoyment. In contrast, those in the chatbot condition were marginally more likely to report feeling judged and significantly more likely to find the experience mentally exhausting. See Table \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e for descriptive and inferential statistics.\u003c/p\u003e\n \u003ctable id=\"Tab3\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eDescriptive and inferential statistics on participants\u0026rsquo; perceptions of the brainstorming session\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eItem\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eM\u0026nbsp;(\u003c/em\u003eSD)\u003c/p\u003e\n \u003cp\u003e(Chatbot)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003eM\u003c/em\u003e (SD)\u003c/p\u003e\n \u003cp\u003e(Human)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003et\u003c/em\u003e(25)\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003e\u003cem\u003ep\u003c/em\u003e\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e1. The brainstorming session helped me generate useful ideas for my essay outline.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.07\u0026nbsp;(0.80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.27\u0026nbsp;(0.88)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.375\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e2. I found the brainstorming partner to be knowledgeable.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.20\u0026nbsp;(1.37)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.53\u0026nbsp;(0.74)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.45\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.002 **\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e3. I feel better prepared to write my essay after the session.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.87\u0026nbsp;(0.92)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.20\u0026nbsp;(1.08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e0.40\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e4. It was easy to communicate my ideas during the session.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.40\u0026nbsp;(1.18)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.20\u0026nbsp;(0.68)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.036 *\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e5. I felt judged during the brainstorming session.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.27\u0026nbsp;(1.03)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.60\u0026nbsp;(0.91)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-1.76\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.091\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e6. I felt engaged throughout the brainstorming session.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.80\u0026nbsp;(0.86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.27\u0026nbsp;(1.03)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.29\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.211\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e7. I enjoyed the brainstorming process.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.60\u0026nbsp;(1.12)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.27\u0026nbsp;(1.10)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.06\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.050\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e8. The session maintained my interest in the topic.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.64\u0026nbsp;(1.0)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.53\u0026nbsp;(0.52)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.007**\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e9. I found the brainstorming process mentally exhausting.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.53\u0026nbsp;(1.19)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.53\u0026nbsp;(0.64)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-2.63\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.014 *\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e10. The session challenged me to think critically about the topic.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.27\u0026nbsp;(1.44)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.07\u0026nbsp;(0.70)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.067\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e11. I was encouraged to explore different perspectives.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.93\u0026nbsp;(1.16)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.33\u0026nbsp;(0.82)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.243\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e12. I learned something new during the brainstorming session.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.27\u0026nbsp;(1.39)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.47\u0026nbsp;(0.52)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.39\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.002 **\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e13. I felt comfortable interacting with the brainstorming partner.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.73\u0026nbsp;(1.22)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e4.60\u0026nbsp;(0.83)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.007 **\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e14. I think brainstorming with a teacher/chatbot* would be more effective.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.73\u0026nbsp;(1.03)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e3.13\u0026nbsp;(1.41)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-1.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.128\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e15. I communicate the same way when using an AI chatbot as when interacting with a human\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e2.20\u0026nbsp;(1.08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e1.80\u0026nbsp;(1.26)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"char\"\u003e\n \u003cp\u003e-1.61\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e.120\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003c/p\u003e\n \u003cdiv class=\"gridtable\"\u003e\n \u003cdiv align=\"char\" class=\"colspec\"\u003e\u003cbr\u003e\u003c/div\u003e\n \u003c/div\u003e\n \u003cp\u003eDespite these differences, both groups rated the session comparably in key outcome areas: they felt equally engaged, challenged to think critically, encouraged to consider different perspectives, and believed the session helped generate useful ideas and better prepared them for essay writing. Notably, both groups remained neutral on whether they believed a chatbot or human would be more effective beforehand.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec13\" class=\"Section2\"\u003e\n \u003ch2\u003e4.3. Physiological Responses\u003c/h2\u003e\n \u003cp\u003eRaw (unadjusted) summary measures (raw means and AUCs) for each physiological variable by condition and phase are reported in Appendix C (Tables \u003cspan class=\"InternalRef\"\u003eC2\u003c/span\u003e \u0026ndash; \u003cspan class=\"InternalRef\"\u003eC3\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eBaseline phase: mean values and AUC.\u003c/strong\u003e A MANCOVA on baseline mean values (controlling for age and gender) revealed a significant overall effect of Condition, \u003cem\u003ePillai\u0026rsquo;s Trace\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.347, \u003cem\u003eF\u003c/em\u003e(3, 23)\u0026thinsp;=\u0026thinsp;4.08, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.02. Pulse was significantly higher in the chatbot than human condition, \u003cem\u003eF\u003c/em\u003e(1, 25)\u0026thinsp;=\u0026thinsp;11.22, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.003 (\u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;85.2, 95% CI[79.9, 90.5] vs. \u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;73.8, 95% CI[68.3, 79.2]); EDA and Temperature did not differ between conditions (both \u003cem\u003eps\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;.20).\u003c/p\u003e\n \u003cp\u003eA MANCOVA on baseline AUCs also showed a significant effect of Condition, \u003cem\u003ePillai\u0026rsquo;s Trace\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.500, \u003cem\u003eF\u003c/em\u003e(3, 22)\u0026thinsp;=\u0026thinsp;7.32, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.001, with higher pulse AUC in the chatbot condition, \u003cem\u003eF\u003c/em\u003e(1, 24)\u0026thinsp;=\u0026thinsp;10.28, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.004, \u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;341, 95% CI[318, 364] vs. \u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;295, 95% \u003cem\u003eCI\u003c/em\u003e[273, 318]), and marginally predicted Temperature (\u003cem\u003eF\u003c/em\u003e(1, 24)\u0026thinsp;=\u0026thinsp;4.18, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.05), but not EDA (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;.50). Temperature AUC was marginally higher in the human condition (\u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;125, 95% \u003cem\u003eCI\u003c/em\u003e[122, 127]) than chatbot condition (\u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;122, 95% \u003cem\u003eCI\u003c/em\u003e[120, 124]).\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eBrainstorming phase: mean values and AUC.\u003c/strong\u003e A MANCOVA on mean values during brainstorming (controlling for all three baseline measures, gender and age), yielded a significant multivariate effect of Condition, \u003cem\u003ePillai\u0026rsquo;s Trace\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.827, \u003cem\u003eF\u003c/em\u003e(3, 22)\u0026thinsp;=\u0026thinsp;31.86, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001. Follow-up univariate tests revealed that Condition significantly predicted Pulse, \u003cem\u003eF\u003c/em\u003e(1, 22)\u0026thinsp;=\u0026thinsp;86.27, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001; but not EDA or Temperature (both \u003cem\u003eps\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;.22). Covariate-adjusted means for Pulse showed that heart rate was slightly higher in the chatbot condition (\u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;80.4, 95% CI [78.1, 82.7]) than in the human condition (\u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;79.2, 95% CI [76.9, 81.5]) (see Fig. \u003cspan class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\n \u003cp\u003eThe corresponding AUC analysis showed a significant overall effect of Condition, \u003cem\u003ePillai\u0026rsquo;s Trace\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.864, \u003cem\u003eF\u003c/em\u003e(3, 19)\u0026thinsp;=\u0026thinsp;40.36, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001, with higher pulse AUC in the chatbot condition, \u003cem\u003eF\u003c/em\u003e(1, 21)\u0026thinsp;=\u0026thinsp;130.74, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001 (\u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1917, 95% CI[1871, 1964] vs \u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;1867, 95% CI[1819, 1916]), and marginally higher cumulative peripheral temperature (AUC) in the human condition, \u003cem\u003eF\u003c/em\u003e(1, 21)\u0026thinsp;=\u0026thinsp;3.77, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.07 (\u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;763, 95% CI[754, 773] vs. (\u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;757, 95% CI[748, 767]). There was no significant condition effect on EDA AUC (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;.47). These are illustrated in Figs. \u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e and \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eTime series analysis.\u003c/strong\u003e Linear mixed-effects models of minute-by-minute physiological data from the 25-minute brainstorming phase, with baseline values as covariates and random intercepts for participants, revealed no significant differences between the chatbot and human conditions in physiological change trajectories (all ps\u0026thinsp;\u0026gt;\u0026thinsp;.44) Full model outputs for each physiological measure are provided in Appendix D.\u003c/p\u003e\n\u003c/div\u003e\n\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\n \u003ch2\u003e4.4. Predicting mood change through brainstorming from physiology (AUCs)\u003c/h2\u003e\n \u003cp\u003eTo investigate whether brainstorming mood could be explained by participants\u0026rsquo; physiological responses during the task, a series of regression models were fitted using AUC values (Pulse, Temperature, EDA) to predict mood change. Analyses were conducted at two levels: (1) aggregated positive/negative mood change scores, and (2) individual mood change outcomes.\u003c/p\u003e\n \u003cp\u003eFor the composite negative mood change score, the regression showed that \u003cem\u003ePulse AUC\u003c/em\u003e positively predicted increased negative mood intensities (\u003cem\u003et\u003c/em\u003e(26)\u0026thinsp;=\u0026thinsp;2.47, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.02). Examining individual negative mood items, pulse AUC significantly predicted increased stress (\u003cem\u003et\u003c/em\u003e(26)\u0026thinsp;=\u0026thinsp;3.03, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.006), boredom (\u003cem\u003et\u003c/em\u003e(26)\u0026thinsp;=\u0026thinsp;2.25, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.03), and was marginally associated with discouragement (\u003cem\u003et\u003c/em\u003e(26)\u0026thinsp;=\u0026thinsp;1.96, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.06). Pulse AUC also showed a marginal negative association with positive moods (\u003cem\u003et\u003c/em\u003e(26) = -1.98, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;0.06), with lower pulse AUC associated with feeling more empowered (\u003cem\u003et\u003c/em\u003e(26) = -3.77, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;.001), engaged (\u003cem\u003et\u003c/em\u003e(26) = -3.18, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.004), connected (\u003cem\u003et\u003c/em\u003e(26) = -3.12, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.004), and determined (\u003cem\u003et\u003c/em\u003e(26) = -2.42, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.02).\u003c/p\u003e\n \u003cp\u003e\u003cem\u003eTemperature AUC\u003c/em\u003e emerged as a predictor of mood outcomes, showing a marginally positive relationship with composite positive mood change (\u003cem\u003et\u003c/em\u003e(26)\u0026thinsp;=\u0026thinsp;2.02, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.05). Specifically, higher temperature AUC significantly predicted increases in positive moods such as empowered (\u003cem\u003et\u003c/em\u003e(26)\u0026thinsp;=\u0026thinsp;3.54, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.002), connected (\u003cem\u003et\u003c/em\u003e(26)\u0026thinsp;=\u0026thinsp;2.93, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.007), and inspired (\u003cem\u003et\u003c/em\u003e(26)\u0026thinsp;=\u0026thinsp;2.13, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.04). Additionally, excitement was also marginally predicted by Temperature AUC (\u003cem\u003et\u003c/em\u003e(26)\u0026thinsp;=\u0026thinsp;1.80, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.08). Conversely, lower temperature AUC was associated with stronger negative emotional responses, including feeling more frustrated (\u003cem\u003et\u003c/em\u003e(26) = -2.20, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.04), and marginally more stressed (\u003cem\u003et\u003c/em\u003e(26) = -1.78, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.087).\u003c/p\u003e\n \u003cp\u003eFor the composite positive and negative mood change scores, \u003cem\u003eEDA AUC\u003c/em\u003e was not a significant predictor. However, at the individual level, it was positively associated with increased stress (\u003cem\u003et\u003c/em\u003e(26)\u0026thinsp;=\u0026thinsp;2.48, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.02), and was marginally associated with feeling less connected (\u003cem\u003et\u003c/em\u003e(26) = -1.81, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.08), less confused (\u003cem\u003et\u003c/em\u003e(26) = -2.02, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.05), and feeling less challenged (\u003cem\u003et\u003c/em\u003e(26) = -1.72, \u003cem\u003ep\u003c/em\u003e\u0026thinsp;=\u0026thinsp;.098).\u003c/p\u003e\n\u003c/div\u003e"},{"header":"5. Discussion","content":"\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003e5.1. Mood\u003c/h2\u003e\u003cp\u003eAt baseline, the mood ratings between the two conditions (Chatbot vs. Human) were not significantly different. While both groups experienced an increase in positive mood intensities through the brainstorming session, the participants in the human teacher condition showed greater increase in positive feelings, such as \u0026lsquo;Motivated\u0026rsquo;, \u0026lsquo;Inspired\u0026rsquo;, \u0026lsquo;Empowered\u0026rsquo;, and \u0026lsquo;Engaged\u0026rsquo; than those who interacted with the chatbot. Conversely, those in the chatbot condition experienced increased intensities in several negative moods while those in the human condition experienced a decrease. These results suggest that while both groups started on equal emotional footing, the quality of interaction shaped the emotional impact of the brainstorming session.\u003c/p\u003e\u003cp\u003eAlthough the human-led condition elicited stronger emotional improvements overall, the intensity change for many mood states \u0026ndash; both \u0026lsquo;positive\u0026rsquo; and \u0026lsquo;negative\u0026rsquo;, such as Focused, Curious, Confident, Determined, Confused, Challenged, Overwhelmed, Anxious, and Annoyed \u0026ndash; did not differ significantly between groups. This highlights that the chatbot\u0026rsquo;s ability to support a wide range of engagement-related emotional states should not be discounted. Moreover, not all negative emotions are detrimental; moods like frustration or feeling challenged can be facilitative of learning, reflecting productive cognitive effort. This experience, also known as cognitive disequilibrium, arises when individuals engage with unfamiliar material that signals knowledge gaps or conflicts with their existing knowledge (Piaget, 2005/1950). To resolve this tension and move forward, individuals are driven to assimilate new information into existing schemas or accommodate by adjusting their schemas. Thus, certain \u0026lsquo;negative\u0026rsquo; emotions may indicate meaningful cognitive engagement and can even be beneficial for deeper learning.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003e5.2. Perceptions of brainstorming session\u003c/h2\u003e\u003cp\u003eParticipants\u0026rsquo; perceptions of the brainstorming session were generally positive, ranging from 3.20 to 4.60, with those who interacted with the human teacher rating their experience more positively. For several aspects, these group differences were significant. This included finding the human teacher more knowledgeable, feeling more comfortable interacting with the human teacher, being more likely to indicate having learnt something new, and also finding it easier to communicate ideas. Conversely, those in the chatbot condition found the experience more mentally taxing and were marginally more likely to feel judged. This may be attributed to the way we designed the chatbot to intentionally avoided providing direct answers. Instead, it was designed to use a Socratic approach and engage participants with a series of question prompts aimed at helping them to think critically and consider different perspectives to ultimately come up with original ideas. While this approach aligns with educational goals of fostering deeper thinking, it may have deviated from participants\u0026rsquo; usual experience with Gen AI tools such as ChatGPT, tools that often provide more immediate and \u0026lsquo;fuller answers\u0026rsquo;. As a result, participants may have perceived the chatbot interaction as more cognitively demanding, leading to greater fatigue or frustration.\u003c/p\u003e\u003cp\u003eDespite the differences noted between conditions, both groups rated the brainstorming session comparably on key outcome areas. Participants across both conditions felt engaged throughout, challenged to think critically, encouraged to consider different perspectives, and believed the session helped them to generate useful ideas and better prepared them for essay writing. These findings highlight the value of human interaction in fostering a sense of comfort, perceived expertise, and cognitive ease, while also suggesting that Gen AI chatbots \u0026ndash; despite limitations in emotional nuance \u0026ndash; can still support core learning aspects like engagement, intellectual stimulation and critical thinking. Notably, both groups remained neutral when asked, after the brainstorming session, whether they believed a chatbot or human would be more effective. This suggests that despite differences in user experience, both interaction types were perceived as similarly effective in achieving the goal of the task \u0026ndash; generating ideas for an essay outline.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\u003ch2\u003e5.3. Physiological responses\u003c/h2\u003e\u003cp\u003e A significant difference in pulse was found, with participants in the chatbot condition showing higher cardiovascular activation than those in the human condition (for both mean values and AUC). Temperature AUC was marginally lower in the chatbot condition. This physiological profile \u0026ndash; elevated heart rate alongside slightly reduced peripheral temperature may reflect greater sympathetic arousal, potentially involving some vasoconstriction which is commonly associated with heightened alertness or stress (e.g., Hayashi et al., 2008). When considered alongside the mood results, which showed greater increases in negative emotions in the chatbot condition, this suggests participants may have experienced the task as more cognitively demanding. As noted earlier, the chatbot\u0026rsquo;s Socratic questioning style, while aimed at fostering deeper thinking, may have been more effortful than expected.\u003c/p\u003e\u003cp\u003eHowever, time series analysis using mixed-effects modelling revealed no significant differences between conditions in physiological change trajectories over the 25-minute brainstorming session, once baseline AUC, gender and age were controlled. Despite differences in mood and perceived cognitive impact, both interaction types elicited similar patterns of physiological change across time. These findings suggest that GenAI chatbots can match human facilitators in fostering physiological engagement and serve as helpful pedagogical aides for the teacher.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec19\" class=\"Section2\"\u003e\u003ch2\u003e5.4. Prediction of mood from physiology\u003c/h2\u003e\u003cp\u003e\u003cem\u003ePulse AUC\u003c/em\u003e was positively associated with increased negative moods (stress, discouragement, boredom), and negatively with positive moods (empowerment, engagement, connection, determination). This is consistent with evidence that heart rate tracks emotional valence and cognitive effort (e.g., Darnell \u0026amp; Kreig, 2019). Mechanistically, increased negative emotions likely involves increased sympathetic activation and/or reduced parasympathetic control, which produce increased and more sustained pulse rate and amplitude over time (higher AUC) (Kreibig, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2010\u003c/span\u003e). In contrast, positive moods like empowerment, engagement, connection and determination may index \u0026lsquo;regulated activation\u0026rsquo; where the body is engaged for the task, but not in an over-aroused stressed state, and the parasympathetic system is helping to keep the heart rate in check, thus producing lower or more transient elevated cardiovascular activation (lower AUC). According to the biopsychosocial model of challenge and threat (Blascovich, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Seery, \u003cspan citationid=\"CR58\" class=\"CitationRef\"\u003e2011\u003c/span\u003e), when individuals perceive demands as threats (e.g., discouragement, stress), their cardiovascular responses are less efficient and more \u0026lsquo;costly\u0026rsquo; over time, whereas perceiving demands as challenges (e.g., empowerment, determination) produce a more efficient profile, which fits with the patterns we identified.\u003c/p\u003e\u003cp\u003e\u003cem\u003eTemperature AUC\u003c/em\u003e was marginally associated with increases in positive mood intensity (empowered, connected, inspired, and excitement), and negatively to negative mood change including frustration and marginally increased stress. This pattern is aligned with research that positive emotional engagement is accompanied by peripheral vasodilation and higher skin temperature, potentially mediated by parasympathetic activity (Stefano et al., \u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e2008\u003c/span\u003e). Conversely, negative moods like stress and frustration are associated with sympathetic-driven vasoconstriction, which decreases peripheral blood flow and lowers skin temperature (e.g., Hayashi et al., 2008). This inverse relationship with Pulse AUC suggests a physiological trade-off between cardiovascular arousal and warmth, where heightened sympathetic activation (high pulse AUC) may coincide with peripheral cooling, while higher skin temperature may reflect a more regulated, emotionally positive state supported by parasympathetic vasodilation (Aristizabal-Tique et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e\u003cp\u003e\u003cem\u003eEDA AUC\u003c/em\u003e was positively related to increased stress, consistent with its role in tracking sympathetic nervous system activation (Akbulut, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Its negative association with feelings of connection may reflect the social consequences of heightened arousal. Stress may skew social appraisal, reducing perceived social support and fostering a sense of disconnection or increased social distance (Wang et al., \u003cspan citationid=\"CR70\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Under stress, individuals may also be more task-oriented, investing less in relational engagement or in building rapport with the brainstorming partner.\u003c/p\u003e\u003cp\u003e The physiological indices that differed between conditions (pulse and temperature AUCs) were also systematically linked to mood variation across participants, independent of facilitator type. Although causality cannot be inferred, this convergence suggests that these physiological indicators capture emotional shifts and reflect core bodily processes underpinning participants\u0026rsquo; emotional experiences during the task, whether interacting with a human or chatbot. This alignment between mood reports and physiological activity validates the self-reported data, showing that the emotional states were reflected in real-time bodily activity \u0026ndash; indicating that these experiences were not merely subjective impressions, but embodied and measurable.\u003c/p\u003e\u003cp\u003eImportantly, the predictive value of physiological markers, particularly Pulse and Temperature AUC, illustrates that beyond group-level differences, the pattern and emotional meaning of physiological activation differ across individuals. Higher pulse AUC predicted more negative mood states, while higher temperature AUC predicted increases in positive emotions. These results indicate that what participants felt during the session is not just explained by group averages, but also by how their bodies responded within each condition. This reinforces the importance of looking beyond group-level comparisons to examine within-individual variation and its emotional significance. Such insights provide meaningful implications for the development of emotionally intelligent GenAI tools. For instance, designers can use physiological research to better understand how different interaction styles or system prompts affect emotional and cognitive responses. Insights from physiological patterns, such as the association between higher heartrate and emotional strain, or between higher skin temperature and positive engagement can inform the development of chatbots that better support learners\u0026rsquo; sustained engagement more effectively.\u003c/p\u003e\u003cp\u003eWhile physiological data provided insights into participant\u0026rsquo;s emotional and cognitive states, it is important to note that in future studies of this nature, physiological responses alone is unlikely to reliably distinguish between the chatbot and human conditions. Physiological metrics, though valuable should be interpreted alongside subjective and behavioural measures to gain a more holistic understanding of user experiences. Relying solely on physiological data may obscure subtle differences that may only emerge through self-report or qualitative data, while relying solely on subjective data may overlook meaningful physiological patterns.\u003c/p\u003e\u003c/div\u003e"},{"header":"6. Conclusion","content":"\u003cp\u003eWhile chatbots are not necessarily superior to human teachers, they are also not demonstrably worse in many respects. Both groups in our study experienced emotional gains, reflected in improved mood, and exhibited largely similar physiological responses, with only modest differences in pulse and temperature AUCs. Perceived effectiveness as expressed through their questionnaire responses was also high across both conditions. Importantly, the chatbot environment still supported key aspects of cognitive and motivational engagement.\u003c/p\u003e\u003cp\u003ePedagogically, the findings reaffirm the enduring value of the human teacher in engaging and motivating the students. Participants consistently rated the human teacher more favourably across multiple dimensions \u0026ndash; comfort, enjoyment, learning gains, and ease of communication \u0026ndash; emphasising the irreplaceable role of human empathy, adaptability, and non-verbal reassurance in teaching. Notwithstanding, the study also makes a strong case for the educational viability of GenAI chatbots, particularly as cognitive partners that can stimulate reflection and critical thinking. Despite limitations in emotional warmth, the chatbot condition still supported high levels of engagement and intellectual challenge, as evidenced by both self-reports and physiological data. This suggests that, when designed thoughtfully, chatbots can serve as scalable pedagogical aides that can support the teacher\u0026rsquo;s design of the learning experience. Its value is especially apparent under circumstances where it is advantageous to have round-the-clock availability, personalised and objective tutoring. They can help mitigate the practical and time constraints that often limit human teaching capacity, despite being unable to fully replicate the nuances of ideal human interaction.\u003c/p\u003e\u003cp\u003eOur study contributes to the understanding of the relationship between learning and emotion in the use of GenAI for education. Our study is premised on the recognition that learning is not only a cognitive activity but also a deeply emotional and physiological experience. The observed mood shifts, particularly the greater positive emotional gains in the human teacher condition, highlight the importance of relational dynamics and perceived emotional presence in shaping students\u0026rsquo; receptivity and engagement.\u003c/p\u003e\u003cp\u003eOur study employs a multimodal, mixed-methods approach that integrates self-reported mood and continuous physiological tracking. This design enhances the reliability of emotional data by triangulating subjective experiences with physiological indicators such as electrodermal activity (EDA), pulse, and skin temperature. In doing so, it mitigates common limitations of self-report measures \u0026ndash; such as social desirability bias and retrospective inaccuracies. The integration of physiological data also offers empirical evidence that emotional states are not only subjectively experienced but physiologically embodied. The predictive power of physiological markers \u0026ndash; particularly skin temperature and heartrate \u0026ndash; strengthens the case for multimodal learning theories on embodied teaching (Lim, \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) and learning (Barsalou, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2008\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe use of AUC and growth curve modelling of physiological data offers methodological contribution in capturing sustained emotional and cognitive engagement over time, rather than relying on static point measurements. These techniques illuminate not just whether physiological differences exist, but how they unfold across learning episodes. The nuanced analyses, which control for baseline individual variability and contextual factors like age and gender, ensure rigour in interpreting affective responses attributed to interaction. Our study also contributes towards a replicable methodological template for future studies aiming to assess affective receptivity and engagement in human\u0026ndash;AI interaction, particularly in educational contexts where physiological data can complement more traditional outcome measures.\u003c/p\u003e\u003cp\u003eWhile the study has found that the GenAI chatbot was able to increase students\u0026rsquo; positive moods, thereby motivating them in their learning, a limitation in this study relates to the design of the chatbot we used. Our chatbot was programmed to use a Socratic questioning approach to promote critical thinking; however, some participants reported feeling mentally exhausted or frustrated as it was atypical to the usual chatbots they had used which would offer them answers directly. Future iterations of the chatbot could incorporate adaptive mechanisms to detect circular or stalled conversations and provide more direct information or examples when necessary, which may help reduce students\u0026rsquo; feelings of frustration. Feedback from the participants also suggests that the chatbot would benefit from emotional augmentation features, such as detecting and responding to signs of user frustration in real time (Arguel et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2019\u003c/span\u003e), adjusting its tone to match that of participants, explicitly acknowledging their effort or confusion, in order to generate greater perceived feelings of empathy and support.\u003c/p\u003e\u003cp\u003eUltimately, it is essential to recognise that AI chatbots and human teachers are fundamentally different (e.g., communication styles, adaptability, sensitivity to nuance). Finding the right fit between learner needs and the type of facilitator will likely be complex, and there is unlikely to be a single, universal solution. An important value that GenAI affords is the myriad of ways that the chatbots can be designed for different educational purposes and audiences. For example, some contexts may benefit from highly structured, informative chatbots, while others might require more conversational agents that are capable of providing emotional support and nuanced dialogue. Future work could explore how different learners respond to varying levels of cognitive and emotional support from GenAI chatbots, to inform the development of personalised chatbot designs that can effectively support a variety of pedagogical innovations.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAbedianpour, S., \u0026amp; Omidvari, A. (2018). Brainstorming strategy and writing performance: Effects and attitudes. \u003cem\u003eJournal of Language Teaching and Research\u003c/em\u003e, \u003cem\u003e9\u003c/em\u003e(5), 1084-1095.\u003c/li\u003e\n\u003cli\u003eAkbulut, F. P. (2022). Evaluating the effects of the autonomic nervous system and sympathetic activity on emotional states. \u003cem\u003eIstanbul Ticaret \u0026Uuml;niversitesi Fen Bilimleri Dergisi\u003c/em\u003e, \u003cem\u003e21\u003c/em\u003e(41), 156-169.\u003c/li\u003e\n\u003cli\u003eAl-Khatib, B. A. (2012). The effect of using brainstorming strategy in developing creative problem solving skills among female students in Princess Alia University College. \u003cem\u003eAmerican International Journal of Contemporary Research\u003c/em\u003e, \u003cem\u003e2\u003c/em\u003e(10), 29-38.\u003c/li\u003e\n\u003cli\u003eAmoush, K. H. (2015). The impact of employing brainstorming strategy on improving writing performance of English major students at Balqa Applied University in Jordan. \u003cem\u003eJournal of Education and Practice\u003c/em\u003e, \u003cem\u003e6\u003c/em\u003e(35), 88-92.\u003c/li\u003e\n\u003cli\u003eArguel, A., Lockyer, L., Kennedy, G., Lodge, J. M., \u0026amp; Pachman, M. (2019). Seeking optimal confusion: a review on epistemic emotion management in interactive digital learning environments. \u003cem\u003eInteractive Learning Environments\u003c/em\u003e, \u003cem\u003e27\u003c/em\u003e(2), 200-210.\u003c/li\u003e\n\u003cli\u003eAristizabal-Tique, V. H., Henao-P\u0026eacute;rez, M., L\u0026oacute;pez-Medina, D. C., Zambrano-Cruz, R., \u0026amp; D\u0026iacute;az-Londo\u0026ntilde;o, G. (2023). Facial thermal and blood perfusion patterns of human emotions: Proof-of-Concept. \u003cem\u003eJournal of thermal biology\u003c/em\u003e, \u003cem\u003e112\u003c/em\u003e, 103464.\u003c/li\u003e\n\u003cli\u003eBarrett, L. F., \u0026amp; Simmons, W. K. (2015). Interoceptive predictions in the brain. \u003cem\u003eNature reviews neuroscience\u003c/em\u003e, \u003cem\u003e16\u003c/em\u003e(7), 419-429.\u003c/li\u003e\n\u003cli\u003eBarsalou, L. W. (2008). Grounded cognition. Annual Review of Psychology, 59, 617\u0026ndash;645. https://doi.org/10.1146/annurev.psych.59.103006.093639\u003c/li\u003e\n\u003cli\u003eBates, D., Kliegl, R., Vasishth, S., \u0026amp; Baayen, H. (2015). Parsimonious mixed models. \u003cem\u003earXiv preprint arXiv:1506.04967\u003c/em\u003e.\u003c/li\u003e\n\u003cli\u003eBlascovich, J. (2013). Challenge and threat appraisal. In \u003cem\u003eHandbook of approach and avoidance motivation\u003c/em\u003e (pp. 431-445). Psychology Press.\u003c/li\u003e\n\u003cli\u003eBoekaerts, M. (2010). The crucial role of motivation and emotion in classroom learning. \u003cem\u003eThe nature of learning: Using research to inspire practice\u003c/em\u003e, 91-111\u003c/li\u003e\n\u003cli\u003eBoucsein, W. (2011). Applications of electrodermal recording. In Electrodermal activity (pp. 259-523). Boston, MA: Springer US.\u003c/li\u003e\n\u003cli\u003eCevher, A. Y., \u0026amp; Yildirim, S. (2023). Design and Use of Chatbots for Educational Purposes: A Study on Student Opinions. \u003cem\u003eIleri Teknolojilerde \u0026Ccedil;alismalar Dergisi\u003c/em\u003e, \u003cem\u003e1\u003c/em\u003e(2), 76-83.\u003c/li\u003e\n\u003cli\u003eChandel, P., \u0026amp; Lim, F. V. (2025). Generative AI and literacy development in the language classroom: A systematic review of literature. \u003cem\u003eUbiquitous Learning\u003c/em\u003e, \u003cem\u003e18\u003c/em\u003e(2), 31.\u003c/li\u003e\n\u003cli\u003eCraig, A. D. (2002). \u003cem\u003eHow do you feel? Interoception: the sense of the physiological condition of the body.\u003c/em\u003e Nature Reviews Neuroscience, 3(8), 655\u0026ndash;666. https://doi.org/10.1038/nrn894\u003c/li\u003e\n\u003cli\u003eCrossley, S. A., Muldner, K., \u0026amp; McNamara, D. S. (2016). Idea generation in student writing: Computational assessments and links to successful writing. \u003cem\u003eWritten Communication\u003c/em\u003e, \u003cem\u003e33\u003c/em\u003e(3), 328-354.\u003c/li\u003e\n\u003cli\u003eD\u0026apos;Mello, S. K., \u0026amp; Graesser, A. (2012). Language and discourse are powerful signals of student emotions during tutoring. \u003cem\u003eIEEE Transactions on Learning Technologies\u003c/em\u003e, \u003cem\u003e5\u003c/em\u003e(4), 304-317.\u003c/li\u003e\n\u003cli\u003eD\u0026rsquo;Mello, S., Lehman, B., Pekrun, R., \u0026amp; Graesser, A. (2014). Confusion can be beneficial for learning. \u003cem\u003eLearning and Instruction\u003c/em\u003e, \u003cem\u003e29\u003c/em\u003e, 153-170.\u003c/li\u003e\n\u003cli\u003eDemeure, V., Niewiadomski, R., \u0026amp; Pelachaud, C. (2011). How is believability of a virtual agent related to warmth, competence, personification, and embodiment? Presence. Teleoperators \u0026amp; Virtual Environments, 20(5), 431\u0026ndash;448. https://doi.org/10.1162/ PRES_a_00065\u003c/li\u003e\n\u003cli\u003eDarnell, D. K., \u0026amp; Krieg, P. A. (2019). Student engagement, assessed using heart rate, shows no reset following active learning sessions in lectures. \u003cem\u003ePloS one\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(12), e0225709.\u003c/li\u003e\n\u003cli\u003eDowning, V. R., Cooper, K. M., Cala, J. M., Gin, L. E., \u0026amp; Brownell, S. E. (2020). Fear of negative evaluation and student anxiety in community college active-learning science courses. \u003cem\u003eCBE\u0026mdash;life sciences education\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e(2), ar20.\u003c/li\u003e\n\u003cli\u003eDuffy, M. C., Lajoie, S. P., Pekrun, R., \u0026amp; Lachapelle, K. (2020). Emotions in medical education: Examining the validity of the Medical Emotion Scale (MES) across authentic medical learning environments. \u003cem\u003eLearning and Instruction\u003c/em\u003e, \u003cem\u003e70\u003c/em\u003e, 101150.\u003c/li\u003e\n\u003cli\u003eEricsson, E., Sofkova Hashemi, S., \u0026amp; Lundin, J. (2023). Fun and frustrating: Students\u0026rsquo; perspectives on practising speaking English with virtual humans. \u003cem\u003eCogent Education\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e(1), 2170088.\u003c/li\u003e\n\u003cli\u003eFang, C., Liu, Z., Li, Q., \u0026amp; Liu, S. (2018, December). Sensor Data-Driven Emotion Perception in Physical Learning Spaces-A Review and Prospect. In \u003cem\u003e2018 Seventh International Conference of Educational Innovation through Technology (EITT)\u003c/em\u003e (pp. 1-5). IEEE.\u003c/li\u003e\n\u003cli\u003eGhabanchi, Z., \u0026amp; Behrooznia, S. (2014). The impact of brainstorming on reading comprehension and critical thinking ability of EFL learners. \u003cem\u003eProcedia-Social and Behavioral Sciences\u003c/em\u003e, \u003cem\u003e98\u003c/em\u003e, 513-521.\u003c/li\u003e\n\u003cli\u003eGorga, D., \u0026amp; Schneider, D. K. (2009). Computer-based learning environments with emotional agents. In \u003cem\u003eHandbook of research on synthetic emotions and sociable robotics: new applications in affective computing and artificial intelligence\u003c/em\u003e (pp. 413-441). IGI Global Scientific Publishing.\u003c/li\u003e\n\u003cli\u003eGouizi, K., Bereksi Reguig, F., \u0026amp; Maaoui, C. (2011). Emotion recognition from physiological signals. \u003cem\u003eJournal of medical engineering \u0026amp; technology\u003c/em\u003e, \u003cem\u003e35\u003c/em\u003e(6-7), 300-307.\u003c/li\u003e\n\u003cli\u003eGuo, K., \u0026amp; Li, D. (2024). Understanding EFL students\u0026rsquo; use of self-made AI chatbots as personalized writing assistance tools: A mixed methods study. \u003cem\u003eSystem\u003c/em\u003e, \u003cem\u003e124\u003c/em\u003e, 103362.\u003c/li\u003e\n\u003cli\u003eHarley, J. M. (2016). Measuring emotions: A survey of cutting edge methodologies used in computer-based learning environment research. \u003cem\u003eEmotions, technology, design, and learning\u003c/em\u003e, 89-114.\u003c/li\u003e\n\u003cli\u003eHarris-Watson, A. M., Larson, L. E., Lauharatanahirun, N., DeChurch, L. A., \u0026amp; Contractor, N. S. (2023). Social perception in Human-AI teams: Warmth and competence predict receptivity to AI teammates. \u003cem\u003eComputers in Human Behavior\u003c/em\u003e, \u003cem\u003e145\u003c/em\u003e, 107765.\u003c/li\u003e\n\u003cli\u003eHashempour, Z., Rostampour, M., \u0026amp; Behjat, F. (2015). The effect of brainstorming as a pre-writing strategy on EFL advanced learners\u0026acirc;\u0026euro;\u0026trade; writing ability. \u003cem\u003eJournal of Applied Linguistics and Language Research\u003c/em\u003e, \u003cem\u003e2\u003c/em\u003e(1), 86-99.\u003c/li\u003e\n\u003cli\u003eHayashi, N., Someya, N., Endo, M. Y., Miura, A., \u0026amp; Fukuba, Y. (2006). Vasoconstriction and blood flow responses in visceral arteries to mental task in humans. \u003cem\u003eExperimental physiology\u003c/em\u003e, \u003cem\u003e91\u003c/em\u003e(1), 215-220.\u003c/li\u003e\n\u003cli\u003eHeine, T., Lenis, G., Reichensperger, P., Beran, T., Doessel, O., \u0026amp; Deml, B. (2017). Electrocardiographic features for the measurement of drivers\u0026apos; mental workload. \u003cem\u003eApplied ergonomics\u003c/em\u003e, \u003cem\u003e61\u003c/em\u003e, 31-43.\u003c/li\u003e\n\u003cli\u003eHorvers, A., Tombeng, N., Bosse, T., Lazonder, A. W., \u0026amp; Molenaar, I. (2021). Detecting emotions through electrodermal activity in learning contexts: A systematic review. \u003cem\u003eSensors\u003c/em\u003e, \u003cem\u003e21\u003c/em\u003e(23), 7869.\u003c/li\u003e\n\u003cli\u003eJeon, J., \u0026amp; Lee, S. (2024). Can learners benefit from chatbots instead of humans? A systematic review of human-chatbot comparison research in language education. \u003cem\u003eEducation and Information Technologies\u003c/em\u003e, \u003cem\u003e29\u003c/em\u003e(17), 23329-23360.\u003c/li\u003e\n\u003cli\u003eKaranjakwut, C., \u0026amp; Charunsri, K. (2025). Transforming AI Chatbots for a Brainstorming Teaching Technique of Process Writing. \u003cem\u003eMalaysian Online Journal of Educational Technology\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(1), 1-18.\u003c/li\u003e\n\u003cli\u003eKellogg, R. T. (1990). Effectiveness of prewriting strategies as a function of task demands. \u003cem\u003eThe American Journal of Psychology\u003c/em\u003e, 327-342.\u003c/li\u003e\n\u003cli\u003eKetonen, E. E., Salonen, V., Lonka, K., \u0026amp; Salmela-Aro, K. (2023). Can you feel the excitement? Physiological correlates of students\u0026apos; self-reported emotions. \u003cem\u003eBritish Journal of Educational Psychology\u003c/em\u003e, \u003cem\u003e93\u003c/em\u003e, 113-129.\u003c/li\u003e\n\u003cli\u003eKreibig, S. D. (2010). Autonomic nervous system activity in emotion: A review. \u003cem\u003eBiological psychology\u003c/em\u003e, \u003cem\u003e84\u003c/em\u003e(3), 394-421.\u003c/li\u003e\n\u003cli\u003eLabadze, L., Grigolia, M., \u0026amp; Machaidze, L. (2023). Role of AI chatbots in education: systematic literature review. \u003cem\u003eInternational journal of Educational Technology in Higher education\u003c/em\u003e, \u003cem\u003e20\u003c/em\u003e(1), 56.\u003c/li\u003e\n\u003cli\u003eLajoie, S. P. (2014). Multimedia learning of cognitive processes. In R. E. Mayer (Ed.), \u003cem\u003eThe Cambridge handbook of multimedia learning\u003c/em\u003e (2nd ed., pp. 623\u0026ndash;646). Cambridge University Press. https://doi.org/10.1017/CBO9781139547369.031.\u003c/li\u003e\n\u003cli\u003eLee, U., Jeong, Y., Koh, J., Byun, G., Lee, Y., Hwang, Y., ... \u0026amp; Lim, C. (2024). Can ChatGPT be a debate partner? Developing ChatGPT-based application \u0026ldquo;DEBO\u0026rdquo; for debate education, findings and limitations. \u003cem\u003eEducational Technology \u0026amp; Society\u003c/em\u003e, \u003cem\u003e27\u003c/em\u003e(2), 321-346.\u003c/li\u003e\n\u003cli\u003eLi, L., Gow, A. D. I., \u0026amp; Zhou, J. (2020). The role of positive emotions in education: A neuroscience perspective. \u003cem\u003eMind, Brain, and Education\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(3), 220-234.\u003c/li\u003e\n\u003cli\u003eLim, F. V. (2020). \u003cem\u003eDesigning learning with embodied teaching: Perspectives from multimodality\u003c/em\u003e. Routledge.\u003c/li\u003e\n\u003cli\u003eLoderer, K., Pekrun, R., \u0026amp; Lester, J. C. (2020). Beyond cold technology: A systematic review and meta-analysis on emotions in technology-based learning environments. \u003cem\u003eLearning and instruction\u003c/em\u003e, \u003cem\u003e70\u003c/em\u003e, 101162.\u003c/li\u003e\n\u003cli\u003eMahapatra, S. (2024). Impact of ChatGPT on ESL students\u0026rsquo; academic writing skills: A mixed methods intervention study. \u003cem\u003eSmart Learning Environments\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e(1), 9.\u003c/li\u003e\n\u003cli\u003eMalmberg, J., Haataja, E., Sepp\u0026auml;nen, T., \u0026amp; J\u0026auml;rvel\u0026auml;, S. (2019). Are we together or not? The temporal interplay of monitoring, physiological arousal and physiological synchrony during a collaborative exam. \u003cem\u003eInternational Journal of Computer-Supported Collaborative Learning\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(4), 467-490.\u003c/li\u003e\n\u003cli\u003eMayer, R. E. (2020). Searching for the role of emotions in e-learning. \u003cem\u003eLearning and Instruction\u003c/em\u003e, \u003cem\u003e70\u003c/em\u003e, 101213.\u003c/li\u003e\n\u003cli\u003eMcNeal, K. S., Zhong, M., Soltis, N. A., Doukopoulos, L., Johnson, E. T., Courtney, S., ... \u0026amp; Porch, M. (2020). Biosensors show promise as a measure of student engagement in a large introductory biology course. \u003cem\u003eCBE\u0026mdash;Life Sciences Education\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e(4), ar50.\u003c/li\u003e\n\u003cli\u003eMehigan, T., \u0026amp; Pitt, I. (2019). Engaging learners through emotion in artificially intelligent environments. In \u003cem\u003eEDULEARN19 Proceedings\u003c/em\u003e (pp. 5661-5668). IATED.\u003c/li\u003e\n\u003cli\u003eNugraha, D., \u0026amp; Indihadi, D. (2019, October). Brainstorming in scientific writing in elementary school. In \u003cem\u003eJournal of Physics: Conference Series\u003c/em\u003e (Vol. 1318, No. 1, p. 012026). IOP Publishing.\u003c/li\u003e\n\u003cli\u003eO\u0026apos;Mealia, S. (2011). How Can Prewriting Strategies Benefit Students?.\u003cem\u003eEducation Masters.\u003c/em\u003e Paper 14.\u003cbr\u003ehttps://fisherpub.sjf.edu/education_ETD_masters/14\u003c/li\u003e\n\u003cli\u003eOudeyer, P. Y., Gottlieb, J., \u0026amp; Lopes, M. (2016). Intrinsic motivation, curiosity, and learning: Theory and applications in educational technologies. \u003cem\u003eProgress in brain research\u003c/em\u003e, \u003cem\u003e229\u003c/em\u003e, 257-284.\u003c/li\u003e\n\u003cli\u003ePiaget, J. (2005/1950). \u003cem\u003eThe psychology of intelligence\u003c/em\u003e. Routledge.\u003c/li\u003e\n\u003cli\u003ePlass, J. L., \u0026amp; Kaplan, U. (2016). Emotional design in digital media for learning. In \u003cem\u003eEmotions, technology, design, and learning\u003c/em\u003e (pp. 131-161). Academic Press.\u003c/li\u003e\n\u003cli\u003ePekrun, R. (2020). Self-report is indispensable to assess students\u0026rsquo; learning. \u003cem\u003eFrontline learning research\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e(3), 185-193.\u003c/li\u003e\n\u003cli\u003eRao, Z. (2007). Training in brainstorming and developing writing skills. \u003cem\u003eELT journal\u003c/em\u003e, \u003cem\u003e61\u003c/em\u003e(2), 100-106.\u003c/li\u003e\n\u003cli\u003eSeery, M. D. (2011). Challenge or threat? Cardiovascular indexes of resilience and vulnerability to potential stress in humans. \u003cem\u003eNeuroscience \u0026amp; Biobehavioral Reviews\u003c/em\u003e, \u003cem\u003e35\u003c/em\u003e(7), 1603-1610.\u003c/li\u003e\n\u003cli\u003eSeli, P., Wammes, J. D., Risko, E. F., \u0026amp; Smilek, D. (2016). On the relation between motivation and retention in educational contexts: The role of intentional and unintentional mind wandering. \u003cem\u003ePsychonomic bulletin \u0026amp; review\u003c/em\u003e, \u003cem\u003e23\u003c/em\u003e(4), 1280-1287.\u003c/li\u003e\n\u003cli\u003eShirvani, M., \u0026amp; Porkar, R. (2021). How do EFL students perceive brainstorming in L2 writing classes. \u003cem\u003eTheory and Practice in Language Studies\u003c/em\u003e, \u003cem\u003e11\u003c/em\u003e(12), 1602-1609.\u003c/li\u003e\n\u003cli\u003eSoltis, N. A., McNeal, K. S., Atkins, R. M., \u0026amp; Maudlin, L. C. (2020). A novel approach to measuring student engagement while using an augmented reality sandbox. \u003cem\u003eJournal of Geography in Higher Education\u003c/em\u003e, \u003cem\u003e44\u003c/em\u003e(4), 512-531.\u003c/li\u003e\n\u003cli\u003eStefano, G. B., Stefano, J. M., \u0026amp; Esch, T. (2008). Anticipatory stress response: a significant commonality in stress, relaxation, pleasure and love responses. \u003cem\u003eMedical Science Monitor\u003c/em\u003e, \u003cem\u003e14\u003c/em\u003e(2), RA17.\u003c/li\u003e\n\u003cli\u003eSumithra, M., Buvaneswar, B., Jessica Judith, S., \u0026amp; Punitha, R. (2022). Innovation for better education system using artificial intelligence. \u003cem\u003eJournal of Cognitive Human-Computer Interaction\u003c/em\u003e, \u003cem\u003e2\u003c/em\u003e(1), 19-28.\u003c/li\u003e\n\u003cli\u003eRichardson, M. P., Strange, B. A., \u0026amp; Dolan, R. J. (2004). Encoding of emotional memories depends on amygdala and hippocampus and their interactions. \u003cem\u003eNature neuroscience\u003c/em\u003e, \u003cem\u003e7\u003c/em\u003e(3), 278-285.\u003c/li\u003e\n\u003cli\u003eRyan, A. M., Gheen, M. H., \u0026amp; Midgley, C. (1998). Why do some students avoid asking for help? An examination of the interplay among students\u0026apos; academic efficacy, teachers\u0026apos; social\u0026ndash;emotional role, and the classroom goal structure. \u003cem\u003eJournal of educational psychology\u003c/em\u003e, \u003cem\u003e90\u003c/em\u003e(3), 528.\u003c/li\u003e\n\u003cli\u003eTyng, C. M., Amin, H. U., Saad, M. N., \u0026amp; Malik, A. S. (2017). The influences of emotion on learning and memory. \u003cem\u003eFrontiers in psychology\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e, 235933.\u003c/li\u003e\n\u003cli\u003eUnin, N. (2016). Brainstorming as a Way to Approach Student-centered Learning in the ESL Classroom. \u003cem\u003eProcedia-Social and Behavioral Sciences\u003c/em\u003e, \u003cem\u003e224\u003c/em\u003e, 605-612.\u003c/li\u003e\n\u003cli\u003eVogel, S., \u0026amp; Schwabe, L. (2016). Learning and memory under stress: implications for the classroom. \u003cem\u003enpj Science of Learning\u003c/em\u003e, \u003cem\u003e1\u003c/em\u003e(1), 1-10.\u003c/li\u003e\n\u003cli\u003eWang, J., \u0026amp; Fan, W. (2025). The effect of ChatGPT on students\u0026rsquo; learning performance, learning perception, and higher-order thinking: insights from a meta-analysis. \u003cem\u003eHumanities and Social Sciences Communications\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(1), 1-21.\u003c/li\u003e\n\u003cli\u003eWang, L., Yu, J., Diao, X., Zhang, Y., Miao, Y., \u0026amp; He, W. (2024). The chain mediating effects of resilience and perceived social support in the relationship between perceived stress and depression in patients with COVID-19. \u003cem\u003eFrontiers in Psychology\u003c/em\u003e, \u003cem\u003e15\u003c/em\u003e, 1400267.\u003c/li\u003e\n\u003cli\u003eXiao, Y., \u0026amp; Zhi, Y. (2023). An exploratory study of EFL learners\u0026rsquo; use of ChatGPT for language learning tasks: Experience and perceptions. \u003cem\u003eLanguages\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e(3), 212.\u003c/li\u003e\n\u003cli\u003eZhang, Q., Siraj, S. B., \u0026amp; Abdul Razak, R. B. (2025). Effects of AI chatbots on EFL students\u0026rsquo; critical thinking skills and intrinsic motivation in argumentative writing. \u003cem\u003eInnovation in Language Learning and Teaching\u003c/em\u003e, 1-29.\u003c/li\u003e\n\u003cli\u003eZimmerman, B. J. (2002). Becoming a self-regulated learner: An overview. \u003cem\u003eTheory into practice\u003c/em\u003e, \u003cem\u003e41\u003c/em\u003e(2), 64-70.\u003c/li\u003e\n\u003cli\u003eZull, J. E. (2006). Key aspects of how the brain learns. In S. Johnson \u0026amp; K. Taylor (Eds.), \u003cem\u003eThe neuroscience of adult learning\u003c/em\u003e (pp. 3\u0026ndash;9).\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"National Institute of Education, Nanyang Technological University","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"customised generative AI chatbots, multimodal learning analytics, emotional responses, physiological responses, embodied learning","lastPublishedDoi":"10.21203/rs.3.rs-7785914/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7785914/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAs generative AI (Gen AI) chatbots become more common as learning partners, questions remain about students\u0026rsquo; emotional and physiological responses to them. This study used a multimodal design to compare university students\u0026rsquo; experiences during a 25‑minute brainstorming session with either a human teacher or Gen AI chatbot. Thirty participants wore EmbracePlus sensors to record heart rate, electrodermal activity (EDA), and skin temperature while completing the task, and completed mood questionnaires before and after brainstorming. Analyses compared mood change scores (controlling for age and gender) and examined physiological data for both temporal patterns and total activation (area-under-the-curve; AUC). While both groups reported improved mood, students brainstorming with a human teacher showed greater gains in positive mood, whereas the chatbot group reported increased stress and discouragement, and exhibited higher cumulative cardiovascular activation. Although physiological change trajectories did not differ by condition, specific AUC measures were associated with mood: higher pulse AUC was linked to negative moods, and higher skin temperature AUC to positive moods. These findings suggest that while human facilitation produces stronger emotional benefits, GenAI chatbots can sustain comparable physiological engagement and serve as valuable complementary tools. Physiological signals also reveal distinctive patterns between bodily states and learning experiences, underscoring the value of integrating multimodal data into research on AI‑mediated education.\u003c/p\u003e","manuscriptTitle":"Moods, Bots, and Bodies: University Students’ Emotional and Physiological responses to Human vs. GenAI Chatbots","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-08 09:16:53","doi":"10.21203/rs.3.rs-7785914/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"76d075bf-8297-4bb7-97b7-2cf740966935","owner":[],"postedDate":"October 8th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":55792885,"name":"Educational Psychology"},{"id":55792886,"name":"Artificial Intelligence and Machine Learning"},{"id":55792887,"name":"Psychology"}],"tags":[],"updatedAt":"2025-10-08T09:16:53+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-08 09:16:53","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7785914","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7785914","identity":"rs-7785914","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}