Designing Emotion Regulation Support in Online Group Learning: Insights from an LLM-Based Support Agent

Designing Emotion Regulation Support in Online Group Learning: Insights from an LLM-Based Support Agent | 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 Designing Emotion Regulation Support in Online Group Learning: Insights from an LLM-Based Support Agent Erfan Jalili Jalal, Maartje Henderikx, Karel Kreijns, Rolands Klemke This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8443551/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 Online group learning (OGL) may be affected by socio-emotional challenges associated with social and interactional barriers in online settings, which can elicit negative emotions among learners. Effective emotion regulation (ER) appears to be a crucial factor in supporting productive collaboration. Recent advances in artificial intelligence (AI), particularly large language models (LLMs), offer potential avenues for ER support in OGL; however, empirical guidance on the design and implementation of such tools remains limited. To begin addressing this gap, the present study examined the use of a default GPT-4 chatbot implemented within an OGL setting as an ER support agent. Chatbot outputs and user experience survey responses were analyzed using a mixed-methods approach combining deductive content analysis, qualitative thematic analysis, and descriptive quantitative measures. Results indicated that most chatbot outputs contained theory-aligned ER components, with socially shared and co-regulated learning strategies occurring more frequently than individual-level ER strategies. User experience findings indicated moderate usability and mixed perceptions of the chatbot’s effectiveness, with qualitative feedback emphasizing the influence of delivery characteristics such as timing and verbosity of the chatbot’s responses. Taken together, the findings suggest that while default LLM-based agents may offer a feasible foundation for ER support in OGL, careful interaction design and theory-aligned refinement are critical for enhancing acceptability and practical value. Online Group Learning (OGL) Computer-supported Collaborative Learning (CSCL) Social Emotions Emotion Regulation Human-computer Interaction (HCI) Large Language Models (LLMs) Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Background Group learning has the potential to support both learning and social outcomes; however, students may encounter socio-emotional challenges that disrupt collaboration (Hadwin et al., 2017 ; Kreijns et al., 2024 ). Hassane et al. ( 2025 ) investigated socio-emotional challenges and social emotions in the context of OGL. The most common socio-emotional challenges identified included external constraints related to a learner’s personal life, different understandings of the task or concept, and the presence of group members who were not fully committed or engaged in free-riding. Most learners (96%) reported encountering socio-emotional challenges, with 86% experiencing multiple challenges during OGL (Hassane et al., 2025 ). According to this study, most students (74%) reported negative emotions, whereas only 26% reported positive emotions. Moreover, learners experienced nearly three times as many negative emotions in OGL as in face-to-face group learning (Hassane et al., 2025 ). Taken together, these findings indicate that socio-emotional challenges in OGL settings are frequently associated with negative social emotions, which can undermine productive collaboration. Hence, regulating negative emotions is essential for a productive learning environment (Hassane et al., 2025 ). In the context of collaborative learning, self-regulation (SRL) is crucial for effective learning, as it encompasses both cognitive learning processes and emotions (Harley et al., 2019 ). Particularly in online settings where social interaction is more challenging, ER becomes more challenging (Hassane et al., 2025 ; Kreijns et al., 2023 ). In the educational context, ER refers to becoming aware of emotional sensations and to monitoring the intensity and/or duration of emotional experiences to maintain productive engagement in learning activities (Gross, 2015 ; Hadwin et al. 2017 ). Hassane et al. ( 2025 ) examined learners’ ER in OGL through the lens of Gross’ ( 2015 ) process model of emotion regulation (PMER). The PMER outlines five families of ER strategies that individuals can use to influence their emotional experiences: situation selection, situation modification, attentional deployment, cognitive change, and response modulation (see Table 1 for definitions of each strategy family and illustrative examples in the context of OGL). Based on this model, Hassane et al. ( 2025 ) identified the strategies learners inherently employ when encountering socio-emotional challenges during OGL. These strategies may either support productive collaboration (e.g., focusing on manageable task elements or reappraising peer feedback positively) or hinder collaboration (e.g., avoiding group learning activities altogether). Table 1 Definitions of the Five ER Families and Illustrative Examples in Group Learning. Adapted from Hassane et al. ( 2025 ) Emotion Regulation Families Example during Group Learning Situation Selection: Taking actions that make it more (or less) likely that one will be in a situation that one expects will give rise to desirable (or undesirable) emotions (Gross, 2015 , p. 7). Avoid engaging in group learning activities altogether. Situation Modification: Taking actions that directly alter a situation to change its emotional impact (Gross, 2015 , p. 8). Initiating a conversation to address and diminish tension within the group Attentional Deployment: Directing one’s attention to influence one’s emotional response (Gross, 2015 , p. 8) Redirecting attention towards manageable task elements to reduce emotional overload. Cognitive Change: Modifying one’s appraisal of a situation to alter its emotional impact (Gross, 2015 , p. 9). Viewing repeated peer feedback as a chance to clarify your own thinking, rather than as constant criticism. Response Modulation: Directly influencing experiential, behavioral, or physiological components of the emotional response after the emotion is well developed (Gross, 2015 , p. 9). Forcing a smile during a tense group discussion to avoid escalating conflict. ER is integral to group learning (Järvenoja et al., 2023 ). However, Gross’s PMER merely emphasizes individual ER (self-regulation). Due to the social nature of collaboration, regulation extends beyond individual ER. Thus, two other regulation modes have been identified (Hadwin et al., 2017 ). These include co-regulation (CoRL), in which a learner supports peers in regulating their emotions by acting as a mediator, and socially shared regulation (SSRL), in which group members collectively manage emotional dynamics to support productive collaboration (Hadwin et al., 2017 ). Recent research highlighted positive correlations between different regulation modes in OGL and noted CoRL’s role in mediating peers’ negative emotions to support emotional stability and productive collaboration (Hassane et al., 2025 ). Given the importance of individual, co-, and socially shared emotion regulation for productive OGL, researchers have increasingly explored how technology can be used to support these regulatory processes (Järvenoja et al., 2020 ). In this context, advances in technology have further fueled the interest in technology-based ER support in OGL (Ngo et al., 2024 ; Nguyen et al., 2023 ; Rojas et al., 2022 ). A substantial body of CSCL research has proposed Group Awareness (GA) support that informs learners about other group members’ activities, knowledge, and emotions. (Henderikx and Kreijns, 2022 ; Kirschner et al., 2015 ; Miller and Hadwin, 2015 ; Shingjergji et al., 2026 ; Chen et al., 2024 ). Early GA tools illustrate how GA can promote interaction and support group processes. Group processes refer to how group members coordinate, monitor, and reflect on their collaborative work and social interactions. For example, Kirschner et al. ( 2015 ) introduced tools such as Radar and Reflector to enhance awareness of group members’ social and cognitive behaviors. These tools stimulate reflection on how the group is functioning, including participation patterns, collaboration quality, and the alignment between individual contributions and group goals. Chen et al. ( 2024 ) examined the effectiveness of GA support through a meta-analysis of 46 empirical studies. The results showed that GA support positively influences behavioral participation, cognitive development, and social emotion. However, the effects of GA support are minor for social emotion compared to the other two dimensions (Chen et al., 2024 ). Another form of ER support proposed for OGL is scripting. Scripting refers to the design of structured prompts or guidance that support learners at different phases of collaboration by directing their attention to essential aspects of the task and group interaction (Miller and Hadwin, 2015 ). Studies showed that scripting can support learners in setting goals related to emotions, such as maintaining positive feelings like optimism and confidence, and reducing negative emotions such as anxiety and stress during collaboration (Li et al., 2025 ; Miller and Hadwin, 2015 , 2024 ). More recently, advances in artificial intelligence have led to the development of AI-based ER support tools for OGL. These tools extend earlier awareness-based approaches by aiming to infer learners’ emotional states from biometric or behavioral cues, including facial expressions, posture, voice, and attention patterns, to enhance emotional awareness and provide more timely ER support (Ngo et al., 2024 ; Nguyen et al., 2023 ). AI-supported emotion detection can improve understanding of emotional regulation and socially shared regulation of learning in synchronous online learning environments (Ngo et al., 2024 ). Upon reviewing state-of-the-art studies on technological ER support in OGL (Hassane, 2025 ), most rely on awareness tools operating under the assumption that the regulation of the emotion will occur automatically once group members are aware of their emotions (Slovak et al., 2023 ). However, Kreijns et al. ( 2023 ) showed that learners do not necessarily regulate their emotions or select an effective ER strategy by themselves. This highlights the need for support that goes beyond awareness by implementing interventions that actively support the regulation process (Hassane, 2025 ). Studies in the context of Human-Computer Interaction (HCI) noted that effective technological interventions should explicitly incorporate validated theoretical frameworks (Kitson et al., 2024 ; Slovak et al., 2023 ). Upon this, they suggested a framework for designing ER interventions by emphasizing theory-informed approaches grounded in Gross’ PMER (Gross, 2015 ). Based on the background presented above, it follows that technologies designed to support ER in OGL should go beyond mere emotion awareness. Specifically, such technologies should provide ER support that is grounded in the following established ER modes: Individual ER : Supporting individual ER grounded in Gross’ PMER (Gross, 2015 ). Enhancing SSRL : Facilitating increased social interaction and collaborative activities through targeted group interventions (Hadwin et al., 2017 ; Järvenoja et al., 2023 ). Facilitating CoRL : Supporting learners to act as mediators toward assisting learners in their individual ER (Hadwin et al., 2017 ; Järvenoja et al., 2023 ). To take a first step toward designing such ER intervention tools for OGL, we developed a GPT-4–based chatbot deployed in an experimental OGL setting as an ER support agent. The chatbot was configured and positioned as a pedagogical agent that supports ER processes within the OGL environment. Through mixed-methods analysis, we examined 1. whether the general GPT-4-based chatbot interactions reflected OGL ER theories (Individual ER, CoRL, SSRL) in its responses and 2. how participants experienced its usability and usefulness. The findings provide evidence for improvements to design requirements for developing LLM-based pedagogical agents for ER in OGL. Methods Participants and context Participants were PhD students, educators, and researchers in technology-enhanced learning, ranging in age, gender, and experience level from junior to senior (n = 20). They were all participating in a workshop as part of the 18th EATEL Summer School on Technology-Enhanced Learning (JTELSS), where they learned how LLMs such as ChatGPT can be customized into interactive, conversational assistants. The first part of the workshop focused on introducing large language models (LLMs) for creating chatbots that can support and assist learners and educators in various educational settings, as well as on the basics of prompting. The second part of the workshop was dedicated to demonstrating and testing the first version of a chatbot that was explicitly designed for ER in OGL. Procedure At the beginning of the experiment, participants were introduced to the context and briefed on the tool they would be using: a chatbot named GroupBuddy. They were informed that GroupBuddy was designed to assist with ER and socio-emotional challenges, particularly in situations involving frustration, disengagement, or emerging conflict. Following this briefing, informed consent was obtained for the anonymous use of chat data for research purposes. Participants were then asked to work collaboratively in four groups. Their specific task was to design a prompt that could automate a pedagogical chatbot for a real-world use case of their choice. Each group should discuss potential applications and select a practical example, such as creating a chatbot for grading and personalized feedback. This activity required joint decision-making, idea negotiation, and coordination, thereby naturally creating opportunities to address the socio-emotional challenges typical of OGL. To simulate an OGL setting, participants were instructed to collaborate from different locations within the venue. The experiment lasted approximately 45 minutes, and at the end, a plenary discussion was held in which participants shared their impressions of GroupBuddy’s behavior and impact on group dynamics, as well as its role, tone, and usefulness during collaboration. Chatbot implementation The AI model and development GroupBuddy was developed using JavaScript and Node.js and connected to the OpenAI API. It utilizes the GPT-4 model (version 0314), which was selected for its 2024 state-of-the-art performance on language-understanding benchmarks and human-like dialogue capabilities (OpenAI et al., 2024; Ou et al., 2024). The source code is openly available on GitHub (see Data and Code Availability section). Full deployment specifications are provided in Appendix A. A zero-shot prompting (Y. Li, 2023) approach was employed, in which the model received instructional prompts without specific examples, relying instead on its pre-existing knowledge. The model was instructed to act as a pedagogical agent that supports group members experiencing negative emotions or socially challenging situations, such as differing goals and unequal participation, as described by Hassane et al. (2025). The chatbot was restricted to responding only to messages indicating negative emotions or group tension and was prevented from offering academic or task-related assistance. Platform and integration GroupBuddy was integrated into Discord (“Discord”, 2024) (see Fig. 1). Discord was chosen because it creates an engaging space for social interactions through rich affordances such as threaded conversations, ready-made emojis, animations, and file-sharing features (Almomani, 2024). Furthermore, the platform offers high technical flexibility for embedding conversational LLMs within its group messaging environment (channels). Within this environment, the chatbot operated on two levels. It could post messages on the group channel either proactively (when negative emotions appeared in the chat) or reactively (when users mentioned its name). Participants and context Participants were PhD students, educators, and researchers in technology-enhanced learning, ranging in age, gender, and experience level from junior to senior (n = 20). They were all participating in a workshop as part of the 18th EATEL Summer School on Technology-Enhanced Learning (JTELSS), where they learned how LLMs such as ChatGPT can be customized into interactive, conversational assistants. The first part of the workshop focused on introducing large language models (LLMs) for creating chatbots that can support and assist learners and educators in various educational settings, as well as on the basics of prompting. The second part of the workshop was dedicated to demonstrating and testing the first version of a chatbot that was explicitly designed for ER in OGL. Procedure At the beginning of the experiment, participants were introduced to the context and briefed on the tool they would be using: a chatbot named GroupBuddy. They were informed that GroupBuddy was designed to assist with ER and socio-emotional challenges, particularly in situations involving frustration, disengagement, or emerging conflict. Following this briefing, informed consent was obtained for the anonymous use of chat data for research purposes. Participants were then asked to work collaboratively in four groups. Their specific task was to design a prompt that could automate a pedagogical chatbot for a real-world use case of their choice. Each group should discuss potential applications and select a practical example, such as creating a chatbot for grading and personalized feedback. This activity required joint decision-making, idea negotiation, and coordination, thereby naturally creating opportunities to address the socio-emotional challenges typical of OGL. To simulate an OGL setting, participants were instructed to collaborate from different locations within the venue. The experiment lasted approximately 45 minutes, and at the end, a plenary discussion was held in which participants shared their impressions of GroupBuddy’s behavior and impact on group dynamics, as well as its role, tone, and usefulness during collaboration. Chatbot implementation The AI model and development GroupBuddy was developed using JavaScript and Node.js and connected to the OpenAI API. It utilizes the GPT-4 model (version 0314), which was selected for its 2024 state-of-the-art performance on language-understanding benchmarks and human-like dialogue capabilities (OpenAI et al., 2024; Ou et al., 2024). The source code is openly available on GitHub (see Data and Code Availability section). Full deployment specifications are provided in Appendix A. A zero-shot prompting (Y. Li, 2023) approach was employed, in which the model received instructional prompts without specific examples, relying instead on its pre-existing knowledge. The model was instructed to act as a pedagogical agent that supports group members experiencing negative emotions or socially challenging situations, such as differing goals and unequal participation, as described by Hassane et al. (2025). The chatbot was restricted to responding only to messages indicating negative emotions or group tension and was prevented from offering academic or task-related assistance. Platform and integration GroupBuddy was integrated into Discord (“Discord”, 2024) (see Fig. 1). Discord was chosen because it creates an engaging space for social interactions through rich affordances such as threaded conversations, ready-made emojis, animations, and file-sharing features (Almomani, 2024). Furthermore, the platform offers high technical flexibility for embedding conversational LLMs within its group messaging environment (channels). Within this environment, the chatbot operated on two levels. It could post messages on the group channel either proactively (when negative emotions appeared in the chat) or reactively (when users mentioned its name). Materials Chatbot interaction data Chat interactions from group members and GroupBuddy were automatically logged during the group activity. These chat logs were then used as a qualitative data source to examine whether the default GPT-4 model exhibited patterns consistent with established OGL ER modes, including individual ER (based on Gross’s PMER(Gross, 2015)), CoRL and SSR (Hadwin et al., 2017; Järvenoja et al., 2023). All logs were anonymized immediately after collection; no personal identifiers were retained. The final dataset comprised n = 75 chatbot messages across four group channels. Verbal feedback data The discussion at the end of the experiment was audio-recorded and later transcribed verbatim for analysis. Usability questionnaires To assess the usability of GroupBuddy, an adapted version of the System Usability Scale (SUS) (Brooke, 1996) was used. Fifteen participants completed this questionnaire, which contained ten closed-ended items rated on a 5-point Likert scale (1 = Completely disagree, 5 = Completely agree). The scale evaluated dimensions such as ease of use, complexity, consistency, and user confidence. Example statements included “I think that I would like to use the chatbot frequently,” “I felt very confident using the chatbot,” and “I thought the chatbot was easy to use”. SUS scoring followed standard procedures, yielding a total usability score ranging from 0 to 100 (Brooke, 1996). These scores were interpreted using established SUS benchmarks to classify perceived usability: 85 = excellent (Aaron, 2009). Usefulness questionnaires To evaluate the chatbot’s perceived usefulness , a custom questionnaire adapted from Davis (1989) was administered to the participants. The survey consisted of 8 closed-ended items rated on a 6-point Likert scale (1 = Completely disagree, 6 = Completely agree) that addressed emotions and support for collaboration, ER, and socio-emotional awareness. Sample items from the survey are “The chatbot functionality supported collaboration positively” and “The chatbot functionality enhanced socio-emotional awareness within the group.” In addition, two open-ended questions were included to gather qualitative feedback: “What did you find most positive about the chatbot functionality?” and “What would you like to see different in the chatbot functionality?” Data Analysis A mixed-methods approach was used to integrate quantitative and qualitative data, allowing triangulation between user evaluations, verbal feedback, and chatbot interaction logs. Quantitative data from the questionnaires were summarized descriptively, while qualitative data were analyzed thematically through theory-informed deductive and inductive content analysis. Quantitative analysis Data from the usability and usefulness questionnaires were used to calculate descriptive statistics; for each item, mean (M) and standard deviation (SD) were calculated to capture central tendencies and variation in participants’ responses. Item-level statistics were visualized to identify aspects of the chatbot that participants found particularly strong or in need of improvement. Given the small sample size and the exploratory nature of the study, no inferential statistical tests were performed. Instead, the analysis focused on identifying patterns and trends across measures to gain insights into the chatbot’s performance and user perceptions. Qualitative analysis Data from the open-ended survey questions, verbal feedback, and chatbot interaction logs were analyzed to capture participants’ subjective experiences and to evaluate the theoretical grounding of the chatbot’s responses. This triangulation was key to understanding both user experience and tool efficacy. Thematic analysis was applied to all three data sources (Braun and Clarke, 2006). After familiarization with the verbal and open-ended survey response data, responses were inductively coded and grouped into themes representing key aspects of participants' experiences. Representative quotes were selected to illustrate each resulting theme. In addition, the analysis of the chatbot interaction logs followed a separate theory-informed deductive coding process. This approach was used to determine the extent to which the chatbot’s responses aligned with established theories of ER in OGL, as discussed in the theoretical background. Each message was examined against the codebook (see Table 2 for a summary of the coding categories and Appendix B for the extended codebook) to determine whether it reflected one or more of the targeted ER modes. Corresponding codes were assigned to responses that aligned with theoretical definitions. This deductive coding process was conducted through a human–LLM collaborative method. The decision to use an LLM, specifically ChatGPT, as a second coder was based on its promising potential to serve as a researcher and accelerate coding qualitative data. Several studies indicate that LLMs can efficiently perform thematic analysis comparable to human coders (Jiang et al., 2021; Liu et al., 2025), with one study indicating an alignment with human coding up to 96.02% for broader themes (Liu et al., 2023). However, the most effective approach, as was used in this study, is the hybrid model, where AI supplements human expertise (Jiang et al., 2025) by providing efficiency while the human preserves critical interpretive judgment and ensures crucial human oversight (Parkington et al., 2025). This approach draws on procedures described by Chew et al. (2023) and Xiao et al. (2023). First, the (human) researcher created the theory-based codebook, formatted it as a prompt, and provided it to ChatGPT 01. Then, the LLM labeled a subset of 10% of the dataset using the provided codebook (O’Connor and Joffe, 2020), while the human researcher independently coded the same subset. Inter-rater reliability was calculated using Cohen’s κ to assess the consistency between human and LLM coding. It took two iterations of refining the codebook and the prompt to achieve a satisfactory level of agreement (κ = 0.85), after which the LLM coded the remaining data using the finalized codebook (see Appendix C for the whole prompt text). Table 2 Summary of coding categories for ER in chatbot responses, with corresponding theoretical definitions Results This section presents findings from the content analysis of chatbot responses, the usability and usefulness surveys, and the thematic analysis of participants’ qualitative feedback. Results are organized around two focal areas: (1) the theoretical alignment of chatbot responses with ER modes in OGL, (2) users’ perceptions of usability and usefulness, and emergent qualitative themes that inform future design improvements. Theory-based Intervention The chatbot response dataset comprised n = 75 groupBuddy outputs across the four Discord group channels. These outputs were produced during the collaborative activity period (~45 min, as per procedure), totaling ~180 group-minutes of interaction. This corresponds to approximately 1.67 chatbot messages per minute across all channels, or 0.42 messages per minute per group (~1 message every 2.4 minutes). Message length was variable (M = 118.4 words, SD = 78.5; range = 8:305), suggesting that the chatbot typically responded with multi-sentence supports. Of these n = 75 outputs, 62 (82.7%) contained at least one ER mode and were coded as an intervention using the regulation modes in Table 2. The remaining 13 (17.3%) responses consisted of greetings and acknowledgements and were left as ‘uncoded’. Examples of output coded as uncoded , SSRL , CoRL , and IER are provided in Table 6 in Appendix D. Across the 62 interventions, multi-label coding yielded 97 regulation-code instances (M = 1.56 codes per intervention). Half of the interventions were single-coded (50.0%), while 43.5% received two codes and 6.5% received three codes (SSRL + CoRL + individual ER). At the code occurrence level, SSRL accounted for 47.4%, CoRL for 39.2%, and individual ER for 13.4% (see Fig. 2). Usability and Usefulness Evaluation The overall SUS score, based on 15 item means, was estimated at 64.7, indicating a “OK' level of usability according to Aaron (2009). Figure 3 illustrates the distribution of SUS scores across participants, revealing that about one-third rated it at or above the industry-average benchmark of 68 (Brooke, 1996). Figure 4 displays mean scores and standard deviations for the ten SUS questions, categorized by themes such as frequent use, unnecessary complexity, and ease of use. Strengths were noted in ease of use (Q3, M = 4.13, SD = 0.83, 95% CI [3.67, 4.60]) and learnability (Q7, M = 3.80, SD = 0.86, 95% CI [3.32, 4.28]), with positive ratings also for integration (Q5, M = 3.60, SD = 1.06, 95% CI [3.02, 4.19]). Willingness to use frequently was low (Q1, M = 2.87, SD = 0.83, 95% CI [2.40, 3.33]), and confidence levels were moderate (Q9, M = 3.47, SD = 0.92, 95% CI [2.96, 3.97]). Conversely, perceptions of unnecessary complexity (Q2, M = 2.47, SD = 1.19, 95% CI [1.81, 3.12]) and learning required (Q10, M = 2.20, SD = 1.21, 95% CI [1.53, 2.87]) showed the greatest variability, indicating mixed user experiences (see Fig. 4) Perceived usefulness ratings are generally positive, especially for experienced fun and perceived advice quality. Fun (M = 4.43, SD = 1.22, 95% CI [3.82, 5.04]) and helpful general advice (M = 4.14, SD = 1.25, 95% CI [3.52, 4.76]) scores above the midpoint. Conversely, items related to collaboration and group-dynamics facilitation are more modest (collaboration M = 3.43, SD = 1.58, 95% CI [2.64, 4.22]; discussing group dynamics M = 3.29, SD = 1.49, 95% CI [2.55, 4.03]), with wider confidence intervals indicating varied experiences across teams. Pattern analysis in Figure 5 shows that fun/motivation and general advice tend to cluster at higher levels. At the same time, collaboration and socio-emotional dynamics items are closer to the scale's midpoint, indicating greater variation. Open-ended questions and post-session debrief We conducted a thematic analysis of two open-ended survey questions, “What did you find most positive about the chatbot functionality?” and “What would you like to see different in the chatbot functionality?,” completed by 16 participants, and of transcribed post-session verbal debrief feedback from an additional 5 participants. The combined qualitative data (n = 21) revealed a coherent set of themes, summarized in Tables 3 (perceived value and positive aspects) and 4 (suggested improvements and concerns). Regarding perceived value (Table 3), participants most often noted the helpfulness and coherence of advice or ideas (29%), along with an emotionally supportive tone described as calming, constructive, or sometimes motivating (24%). Less common themes included appreciation for proactive or easily accessible support when it was not overbearing (10%), novelty or enjoyment (10%), limited benefits for coordination or focus (5%), and the suitability for distributed or long-term group work contexts (5%). With respect to suggested improvements (Table 4), the most salient concerns were intrusiveness, particularly requests for fewer, better-timed messages or responses only when needed (33%), as well as shorter messages to reduce perceived verbosity (24%). Participants also noted issues related to timing or latency, such as a desire for faster responses or intervention at critical moments (14%), and to role clarity and conversational ordering, including ambiguity between moderator and peer roles and disruptions to group chat flow (14%). Less frequently reported but notable concerns included maintaining user control following explicit “stop” requests (5%), improving sensitivity by reducing oversensitivity to negative language or better handling of irony (5%), occasional accuracy issues (1 of 21; 5%), and minor feature requests, such as the inclusion of playful elements or GIFs (5%). Table 3 Positive/value themes across open-ended usefulness responses and post-session verbal feedback, pooled (n = 21). Themes are non-exclusive; prevalence is shown as a percentage. Selected representative quotes are included Theme Definition (coding rule) Representative quote Mentions (%) Helpful, coherent advice & ideas Useful, on-point answers; context understanding; idea generation “ useful… recommendations are good”; “good understanding of the context”; “coherent answers” 6 (29%) Emotion-supportive tone Calming/motivating; constructive presence; responds to emotions “tries to make peace”; “help to make… calm”; “motivating messages”; “reaction to our emotions” 5 (24%) Proactive/available support Takes initiative to help (perceived positively when not excessive) “initiative and proactive approaches” 2 (10%) Coordination/focus Nudge group to focus on specific aspects “coordinate the group and make it focus” 1 (5%) Novelty/enjoyment New, engaging, or “cute/fun” “different experience”; “answers were cute” 2 (10%) Potential fit for distributed use Perceived value for longer, distributed work “good for distributed group working… for a long time” 1 (5%) Table 4 Improvement/concern themes across open-ended usefulness responses and post-session debrief, pooled (n = 21). Themes are non-exclusive; prevalence shown as a percentage. Selected representative quotes included Theme Definition (coding rule) Representative quote Mentions (%) Intrusiveness/reply frequency & proactivity Too many messages; responds untagged; should reply only when needed; crowds out group talk “too much… responded to basically everything”; “less invasive”; “not too proactive… only when needed”; “responded too frequently—even when not tagged” 7 (33%) Verbosity/message length Texts too long; hard to read “answers were too long”; “shorter texts”; verbal: “generated text was long” 5 (24 %) Timing/latency Slow or mistimed; needs timely help at critical moments “appearance time… to answer”; “quick and dirty… responses”; “timely responses at critical moments” 3 (14%) Role clarity/conversation order Acts like peer vs. moderator; need ordering/structure verbal: “acted more like a group member than a moderator”; “make order in the group chat”; “didn’t enable group dynamics… long answers” 3 (14%) Respect for user control Keeps talking after being asked to stop verbal: “reacts to request of not responding… then starts responding again” 1 (5%) Sensitivity/irony detection Sensitivity/irony detection “oversensitive… cannot differentiate irony” 1 (5%) Accuracy Occasional correctness concerns “more accurate” 1 (5%) feature requests Suggesting new features “more funny aspects… using GIFs” 1 (5%) Discussion This exploratory study provides insights into how a default GPT-4 chatbot (zero-shot) performs when serving as an ER support agent in an OGL setting. Taken together, the findings offer insight into 1. the theoretical alignment of chatbot responses with ER modes in OGL and 2. how delivery characteristics shaped users’ perceptions of usability and usefulness, with implications for future chatbot design. Regarding theoretical alignment, the chatbot’s responses are most frequently aligned with SSRL, followed by CoRL, with individual ER occurring least often. The predominance of SSRL-like interactions suggests that a default GPT-4 model, operating under a zero-shot prompting approach and without explicit theory-guided constraints, is nevertheless capable of generating socially shared, group-oriented regulatory supports in collaborative learning contexts. One plausible explanation is that contemporary LLM assistants are pretrained on large-scale internet text and fine-tuned via RLHF (Reinforcement Learning from Human Feedback) to behave helpfully in dialogue (Bai et al., 2022 ; Ouyang et al., 2022 ). Recent work increasingly positions LLM-based agents as facilitators of collaborative learning and group discussion (Yang et al., 2025 ); accordingly, such systems may be more likely to produce group-oriented scaffolding than individualized ER coaching in multi-party OGL contexts (Lu et al., 2024 ). While socially shared support may be beneficial for collective regulation, it may also constrain the model’s capacity to provide more tailored individual ER strategies in OGL contexts without additional prompting or structural differentiation. Beyond theoretical alignment, several quantitative and qualitative indicators suggest that users’ perceptions of the chatbot were shaped more strongly by its delivery characteristics than by the content of its interventions alone. Chatbot outputs occurred relatively frequently (approximately once every 2.4 minutes per group) and were often lengthy (M = 118.4 words). These characteristics closely align with the most frequently reported usability concerns, particularly perceptions of intrusiveness and verbosity. Although participants frequently described the chatbot’s advice as coherent or supportive, these delivery features appear to have diminished its perceived usefulness for collaboration, helping explain why collaboration-related usefulness ratings were comparatively modest. Taken together, these findings suggest that future iterations of chatbot-based ER support in OGL should prioritize tighter control over unsolicited interventions. Design strategies such as refined sentiment thresholds, rate limiting, and persistent user control mechanisms (particularly those that reliably respect explicit “stop” requests) may help reduce perceived disruption while preserving supportive intent. Similarly, defaulting to concise, single-action messages (for example, brief validation paired with one concrete suggestion) may improve acceptability without substantially reducing supportive value. Channel configuration also appears to play a meaningful role in user experience. The exclusive use of a shared group channel likely contributed to perceptions of disruption, as public interventions can interfere with the natural flow of group interaction. Providing a private channel for individual ER support may reduce conversational crowding while enabling more personalized strategies. Individual ER often benefits from private disclosure, personalization, and iterative follow-up, all of which may be constrained in a public group chat where participants reasonably provide less personal context (Bazarova et al., 2015 ; Doré et al., 2016 ). Finally, recurrent concerns regarding role ambiguity underscore the importance of clearer role framing for ER chatbots in OGL. Participants’ feedback suggests the need to specify whether the chatbot functions as a moderator, a peer-like supporter, or a pedagogical agent, and to clarify when and how interventions will occur. More transparent role communication during onboarding and throughout use may help align user expectations with system behavior and reduce frustration related to unexpected or poorly timed interventions. Overall, the findings highlight that while a default GPT-4 chatbot can align with theoretically meaningful ER modes in OGL, its effectiveness and acceptability are strongly contingent on interaction design choices. Addressing delivery timing, verbosity, private channel placement, and role clarity appears critical for translating theoretically aligned ER support into practically helpful and non-disruptive learning support. Limitations This exploratory study is subject to several methodological and contextual limitations that influence the generalizability and interpretation of the findings. First, this research was conducted with a small convenience sample (N = 18) in a short experimental OGL session (approx. 45 minutes). Consequently, the generalizability of the results, particularly concerning the long-term benefit and varied user experiences with the chatbot, is limited. Future research should employ larger, more diverse samples over more extended testing periods to validate and generalize these preliminary findings. A second limitation is that using the LLM (ChatGPT o1) for the deductive coding process introduces challenges to replicability. Since that specific LLM version is no longer publicly available, exact methodological replication is limited. This highlights the necessity for transparency in reporting LLM methodology and emphasizes the need for critical reflection on findings generated using non-static AI tools. Future studies should focus on robust, version-agnostic prompting strategies to enhance methodological durability and replicability, although this will likely be a challenge due to the rapid continuous enhancement of AI models. Third, this study examined the theoretical alignment of chatbot responses with different emotion regulation (ER) modes in OGL. However, the analysis focused on chatbot outputs in isolation and did not examine them in relation to surrounding user conversations. Consequently, we did not assess the contextual appropriateness or correctness of the chatbot interventions in combination with learners’ expressed sentiments in the Discord group chats. Future research should adopt context-sensitive analyses that link chatbot responses to preceding and subsequent learner interactions, including sentiment dynamics, to evaluate the situational relevance, timing, and effectiveness of ER support in authentic collaborative settings. A further limitation concerns the interpretation of participants’ evaluations of the chatbot’s content and support. Although users expressed a range of positive and critical perspectives, they were not selected or trained to assess ER strategies from a theoretical or clinical standpoint. As a result, these evaluations likely reflect subjective user experiences rather than objective judgments of ER quality or theoretical appropriateness. Finally, the study identified specific boundary conditions related to the hybrid OGL setting. Because groups were co-located while using an online chat platform, face-to-face cues sometimes substituted for the chatbot's chat-based mediation. Additionally, participants perceived group-level interventions from GroupBuddy as 'crowding' the conversation. This suggests that the perceived marginal value of a chat-embedded agent diminishes when co-presence is high and public posts are frequent or lengthy. Future experiments should test chatbots in actual OGL settings where digital mediation is the primary mode of interaction, thereby reducing the influence of co-presence effects. Also, implementing a private channel to offer confidential, individual support alongside group-level interventions may mitigate participants' perceived issue of 'crowding' public conversation. Conclusion and future directions This exploratory study examined how a zero-shot GPT-4 chatbot functioned as an emotion ER support agent within an OGL setting. The findings suggest that large language model–based agents may provide a feasible foundation for ER support in OGL settings. However, their effectiveness appears contingent on careful alignment between the theoretical intent and the system's behavior. In particular, managing unsolicited interventions, promoting concise and context-sensitive messaging, and differentiating between group-level and individual-level support channels appear critical for reducing disruption while preserving supportive value. Future development studies should move beyond exploratory testing to systematically examine how theory-guided prompting strategies, adaptive intervention thresholds, and the inclusion of private support channels influence the effectiveness of ER support. Studies employing larger, more diverse samples, more extended interaction periods, and fully online OGL settings are needed to assess the durability and generalizability of these findings. In addition, future work should incorporate expert evaluations of ER quality alongside user feedback to more rigorously assess theoretical adequacy and instructional value. Collectively, such efforts will be essential for advancing the responsible and effective integration of LLM-based Emotion Regulation support for Online Group Learning. Declarations Funding The authors declare that no funds, grants, or other support were received during the preparation of this manuscript. Competing Interests The authors have no relevant financial or non-financial interests to disclose. Author Contributions All authors contributed to the conceptualization and design of the study. Material preparation, data collection, and analysis were performed by Erfan Jalili Jalal and Maartje Henderikx. The manuscript was written by Erfan Jalili Jalal and Maartje Henderikx. Maartje Henderikx, Karel Kreijns, and Rolands Klemke commented on previous versions of the manuscript. All authors read and approved the final manuscript. Ethics Approval The study did not involve medical or sensitive personal data. The data were gathered in accordance with ethical standards; participants volunteered, and the research study was approved by [blinded for review]. Consent to Participate Informed consent was obtained from all individual participants included in the study. Consent for Publication Not applicable; no personal or identifiable information is published. Data and Code Availability Due to institutional regulations, the datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request. The chatbot source code and prompt templates used in this study are openly available at: https://github.com/ejalili/groupBuddy Authors and Affiliations Erfan Jalili Jalal [email protected] Erfan Jalili Jalal 1 . Maartje Henderix 1 . Karel Kreijns 2 . Rolands Klemke 2,3 . 1. Research in Distance Education (RIDE) Lab. Open Universiteit, 6401 DL Heerlen, the Netherlands 2. Department of Learning & Instruction, Faculty of Psychology, 6401 DL Heerlen, the Netherlands 3. Cologne Game Lab, TH Köln, Cologne, Germany The authors declare that no funds, grants, or other support were received during the preparation of this manuscript. Competing Interests The authors have no relevant financial or non-financial interests to disclose. Author Contributions All authors contributed to the conceptualization and design of the study. Material preparation, data collection, and analysis were performed by Erfan Jalili Jalal and Maartje Henderikx. The manuscript was written by Erfan Jalili Jalal and Maartje Henderikx. Maartje Henderikx, Karel Kreijns, and Rolands Klemke commented on previous versions of the manuscript. All authors read and approved the final manuscript. Ethics Approval The study did not involve medical or sensitive personal data. The data were gathered in accordance with ethical standards; participants volunteered, and the research study was approved by [blinded for review]. Consent to Participate Informed consent was obtained from all individual participants included in the study. Consent for Publication Not applicable; no personal or identifiable information is published. Data and Code Availability Due to institutional regulations, the datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request. The chatbot source code and prompt templates used in this study are openly available at: https://github.com/ejalili/groupBuddy Authors and Affiliations Erfan Jalili Jalal [email protected] Erfan Jalili Jalal 1 . Maartje Henderix 1 . Karel Kreijns 2 . Rolands Klemke 2,3 . 1. Research in Distance Education (RIDE) Lab. Open Universiteit, 6401 DL Heerlen, the Netherlands 2. Department of Learning & Instruction, Faculty of Psychology, 6401 DL Heerlen, the Netherlands 3. Cologne Game Lab, TH Köln, Cologne, Germany References Aaron, B. (2009). Determining what individual SUS scores mean: adding an adjective rating scale. Journal of usability studies , 4 , 3. Almomani, G. (2024). Exploring digital technology use in socially and technologically complex contexts: the case of Discord. [Doctoral Thesis, University of Edinburgh]. https://doi.org/10.7488/era/4881 Bai, Y., Jones, A., Ndousse, K., Askell, A., Chen, A., DasSarma, N., et al. (2022). Training a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback. arXiv . https://doi.org/10.48550/arXiv.2204.05862 Bakhtiar, A., Webster, E. A., & Hadwin, A. F. (2018). Regulation and socio-emotional interactions in a positive and a negative group climate. Metacognition and Learning , 13 , 57–90. https://doi.org/10.1007/s11409-017-9178-x Bazarova, N. N., Choi, Y. H., Schwanda Sosik, V., Cosley, D., & Whitlock, J. (2015). Social Sharing of Emotions on Facebook: Channel Differences, Satisfaction, and Replies. In Proceedings of the 18th ACM Conference on Computer Supported Cooperative Work & Social Computing (pp. 154–164). New York, NY, USA: Association for Computing Machinery. https://doi.org/10.1145/2675133.2675297 Braun, V., & Clarke, V. (2006). Using thematic analysis in psychology. Qualitative Research in Psychology , 3 (2), 77–101. https://doi.org/10.1191/1478088706qp063oa Brooke, J. (1996). SUS-A quick and dirty usability scale. Usability evaluation in industry , 189 (194), 4–7. Chen, D., Zhang, Y., Luo, H., Zhu, Z., Ma, J., & Lin, Y. (2024). Effects of group awareness support in CSCL on students’ learning performance: A three-level meta-analysis. International Journal of Computer-Supported Collaborative Learning , 19 (1), 97–129. https://doi.org/10.1007/s11412-024-09418-3 Chew, R., Bollenbacher, J., Wenger, M., Speer, J., & Kim, A. (2023). LLM-Assisted Content Analysis: Using Large Language Models to Support Deductive Coding. arXiv. https://doi.org/10.48550/ARXIV.2306.14924 Davis, F. D. (1989). Perceived Usefulness, Perceived Ease of Use, and User Acceptance of Information Technology. MIS Quarterly , 13 (3), 319–340. https://doi.org/10.2307/249008 Discord. (2024). Discord - Group Chat That’s All Fun & Games . https://discord.com/ Doré, B. P., Silvers, J. A., & Ochsner, K. N. (2016). Toward a Personalized Science of Emotion Regulation. Social and personality psychology compass , 10 (4), 171–187. https://doi.org/10.1111/spc3.12240 Gross, J. J. (2015). Emotion Regulation: Current Status and Future Prospects. Psychological Inquiry , 26 (1), 1–26. https://doi.org/10.1080/1047840x.2014.940781 Hadwin, A., Järvelä, S., & Miller, M. (2017). Self-Regulation, Co-Regulation, and Shared Regulation in Collaborative Learning Environments. In D. H. Schunk & J. A. Greene (Eds.), Handbook of Self-Regulation of Learning and Performance (2nd ed., pp. 83–106). Routledge. https://doi.org/10.4324/9781315697048-6 Harley, J. M., Pekrun, R., Taxer, J. L., & Gross, J. J. (2019). Emotion Regulation in Achievement Situations: An Integrated Model. Educational Psychologist , 54 (2), 106–126. https://doi.org/10.1080/00461520.2019.1587297 Hassane, S. (2025). Exploring socio-emotional regulation in online group learning . [Doctoral Thesis, Open Universiteit]. Open University. https://doi.org/10.71583/20251212sh Hassane, S., Henderikx, M., Järvenoja, H., Beckers, J., & Kreijns, K. (2025). The biotope of online group learning: group genesis and persistence seen through the lens of socio-emotional challenges and regulation. European Journal of Psychology of Education , 40 (2), 1–31. https://doi.org/10.1007/s10212-025-00967-1 Henderikx, M. A., & Kreijns, C. (2022). Development and evaluation of the “Feedback and Reflection in Online Collaborative Learning” (FROCOLE) app. In INTED2022 Proceedings: 16th International Technology, Education and Development Conference (pp. 1582–1590). IATED Academy. https://doi.org/10.21125/inted.2022.0474 Järvenoja, H., Malmberg, J., Törmänen, T., Mänty, K., Haataja, E., Ahola, S., & Järvelä, S. (2020). A Collaborative Learning Design for Promoting and Analyzing Adaptive Motivation and Emotion Regulation in the Science Classroom. Frontiers in Education , 5 . https://doi.org/10.3389/feduc.2020.00111 Järvenoja, H., Näykki, P., & Törmänen, T. (2019). Emotional regulation in collaborative learning: when do higher education students activate group level regulation in the face of challenges? Studies in Higher Education , 44 (10), 1747–1757. https://doi.org/10.1080/03075079.2019.1665318 Järvenoja, H., Törmänen, T., Järvelä, S., & Shubina, T. (2023). Motivation and emotion regulation in collaborative learning contexts. In Motivation and Emotion in Learning and Teaching across Educational Contexts (pp. 100–113). Routledge. Järvenoja, H., Volet, S., & Järvelä, S. (2013). Regulation of emotions in socially challenging learning situations: an instrument to measure the adaptive and social nature of the regulation process. Educational Psychology , 33 (1), 31–58. https://doi.org/10.1080/01443410.2012.742334 Jiang, J. A., Wade, K., Fiesler, C., & Brubaker, J. R. (2021). Supporting Serendipity: Opportunities and Challenges for Human-AI Collaboration in Qualitative Analysis. Proceedings of the ACM on Human-Computer Interaction , 5 (CSCW1), 1–23. https://doi.org/10.1145/3449168 Kirschner, P. a., Kreijns, K., Phielix, C., & Fransen, J. (2015). Awareness of cognitive and social behaviour in a CSCL environment. Journal of Computer Assisted Learning , 31 (1), 59–77. https://doi.org/10.1111/jcal.12084 Kitson, A., Slovak, P., & Antle, A. N. (2024). Supporting Cognitive Reappraisal With Digital Technology: A Content Analysis and Scoping Review of Challenges, Interventions, and Future Directions. In Proceedings of the CHI Conference on Human Factors in Computing Systems (pp. 1–17). https://doi.org/10.1145/3613904.3642488 Kreijns, K., Weidlich, J., & Kirschner, P. A. (2023). Pitfalls of Social Interaction in Online Group Learning. In Z. Yan (Ed.), The Cambridge Handbook of Cyber Behavior (1st ed., pp. 588–630). Cambridge University Press. https://doi.org/10.1017/9781107165250.025 Kreijns, K., Yau, J., Weidlich, J., & Weinberger, A. (2024). Towards a comprehensive framework of social presence for online, hybrid, and blended learning. Frontiers in Education , 8 . https://doi.org/10.3389/feduc.2023.1286594 Li, X., Hu, W., & Li, Y. (2025). Effect of adaptable and non-adaptable collaboration scripts through conversational agents on student’s engagement in online collaborative learning. Educational technology research and development , 73 (2), 875–907. https://doi.org/10.1007/s11423-025-10448-3 Li, Y. (2023). A Practical Survey on Zero-shot Prompt Design for In-context Learning. In Proceedings of the Conference Recent Advances in Natural Language Processing - Large Language Models for Natural Language Processings (pp. 641–647). https://doi.org/10.26615/978-954-452-092-2_069 Liu, X., Zambrano, A. F., Baker, R. S., Barany, A., Ocumpaugh, J., Zhang, J., et al. (2025). Qualitative Coding with GPT-4: Where it Works Better. Journal of Learning Analytics , 12 (1), 169–185. https://doi.org/10.18608/jla.2025.8575 Lu, Y., Aleta, A., Du, C., Shi, L., & Moreno, Y. (2024). LLMs and generative agent-based models for complex systems research. Physics of Life Reviews , 51 , 283–293. https://doi.org/10.1016/j.plrev.2024.10.013 Miller, M., & Hadwin, A. (2015). Scripting and awareness tools for regulating collaborative learning: Changing the landscape of support in CSCL. Computers in Human Behavior , 52 , 573–588. https://doi.org/10.1016/j.chb.2015.01.050 Miller, M., & Hadwin, A. F. (2024). Comparing the effectiveness of CSCL scripts for shared task perceptions in socially shared regulation of collaborative learning. International Journal of Computer-Supported Collaborative Learning , 19 (4), 455–478. https://doi.org/10.1007/s11412-024-09434-3 Ngo, D., Nguyen, A., Dang, B., & Ngo, H. (2024). Facial Expression Recognition for Examining Emotional Regulation in Synchronous Online Collaborative Learning. International Journal of Artificial Intelligence in Education , 34 (3), 650–669. https://doi.org/10.1007/s40593-023-00378-7 Nguyen, A., Li, Y., Järvelä, S., Liu, Y., & Zhao, G. (2023). Emotional Synchrony Through Regulatory Triggers in Collaborative Learning: A Facial Expression Recognition Study.In Proceedings of the 17th International Conference of the Learning Sciences (ICLS 2023) (pp. 1038–1041) . International Society of the Learning Sciences. https://doi.org/10.22318/icls2023.547768 O’Connor, C., & Joffe, H. (2020). Intercoder Reliability in Qualitative Research: Debates and Practical Guidelines. International Journal of Qualitative Methods , 19 . https://doi.org/10.1177/1609406919899220 OpenAI, Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., et al. (2024). GPT-4 Technical Report. arXiv. https://doi.org/10.48550/arXiv.2303.08774 Ou, J., Lu, J., Liu, C., Tang, Y., Zhang, F., Zhang, D., & Gai, K. (2024). DialogBench: Evaluating LLMs as Human-like Dialogue Systems. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) (pp. 6137–6170). https://doi.org/10.18653/v1/2024.naacl-long.341 Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C. L., Mishkin, P., et al. (2022). Training language models to follow instructions with human feedback. arXiv. https://doi.org/10.48550/arXiv.2203.02155 Parkington, K., Teferra, B. G., Rouleau-Tang, M., Perivolaris, A., Rueda, A., Dubrowski, A., et al. (2025). Human vs. LLM-Based Thematic Analysis for Digital Mental Health Research: Proof-of-Concept Comparative Study. arXi v. https://doi.org/10.48550/arXiv.2507.08002 Pekrun, R., Muis, K. R., Frenzel, A. C., & Goetz, T. (2017). Emotions at School (1st ed.). Routledge. https://doi.org/10.4324/9781315187822 Rojas, M., Nussbaum, M., Guerrero, O., Chiuminatto, P., Greiff, S., Del Rio, R., & Alvares, D. (2022). Integrating a collaboration script and group awareness to support group regulation and emotions towards collaborative problem solving. International Journal of Computer-Supported Collaborative Learning , 17 (1), 135–168. https://doi.org/10.1007/s11412-022-09362-0 Sentiment. (2024). Sentiment (Version 5.0.2) [Computer Software]. npm. https://www.npmjs.com/package/sentiment Shingjergji, K., Iren, D., Urlings, C., & Klemke, R. (2026). Affective computing in online higher education: A systematic literature review. Computers and Education: Artificial Intelligence , 10 , 100499. https://doi.org/10.1016/j.caeai.2025.100499 Slovak, P., Antle, A., Theofanopoulou, N., Daudén Roquet, C., Gross, J., & Isbister, K. (2023). Designing for Emotion Regulation Interventions: An Agenda for HCI Theory and Research. ACM Transactions on Computer-Human Interaction , 30 (1), 1–51. https://doi.org/10.1145/3569898 Weidlich, J., Yau, J., & Kreijns, K. (2024). Social presence and psychological distance: A construal level account for online distance learning. Education and Information Technologies , 29 (1), 401–423. https://doi.org/10.1007/s10639-023-12289-0 Xiao, Z., Yuan, X., Liao, Q. V., Abdelghani, R., & Oudeyer, P.-Y. (2023). Supporting Qualitative Analysis with Large Language Models: Combining Codebook with GPT-3 for Deductive Coding. In 28th International Conference on Intelligent User Interfaces (pp. 75–78). Association for Computing Machinery. https://doi.org/10.1145/3581754.3584136 Yang, Q., Yang, Y., An, S., Hao, T., & Xu, G. (2025). LLM-based collaborative agents with pedagogy-guided interaction modeling for timely instructive feedback generation in task-oriented group discussions. In Proceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence (pp. 9972–9980). Montreal, Canada. https://doi.org/10.24963/ijcai.2025/1108 Additional Declarations No competing interests reported. Supplementary Files Appendixs.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-8443551","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":566593508,"identity":"4a95face-8064-4a64-a42e-685591bf405c","order_by":0,"name":"Erfan Jalili Jalal","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA+0lEQVRIie3OsWrDMBCAYR0H50XY6xkNeQVnCYYm+FViDJ46dOyQwV00lWTto3R0MHjSAwS6FPICKYEuhVJLLt0sr4XoH04g9EkSIhT6l8H7MEjEiE3LIhdx1Njt3EMwc4QQLGFBsrXbPEcsgsadJN76SaI6uD6IeEERPLX5jheUnnuG12mS7itUL4KW2n2s56VWdc1gpklmklZJQTASYtDqfjXMaVIYxK+BFCP55kKnxk8yiWRfKR1JNZeapZ+wQbqTGVUj2XOlZV3lpYckz4hv8rHfHJKuu/LnenOIuuPpQ68nye/3erfg39XbGTC0cxMu8ydDoVDoBvsBHmk/tOLJsYAAAAAASUVORK5CYII=","orcid":"","institution":"Open University of the Netherlands","correspondingAuthor":true,"prefix":"","firstName":"Erfan","middleName":"Jalili","lastName":"Jalal","suffix":""},{"id":566593512,"identity":"838f5634-7a94-40ba-af10-b1448c8a4d9d","order_by":1,"name":"Maartje Henderikx","email":"","orcid":"","institution":"Open University of the Netherlands","correspondingAuthor":false,"prefix":"","firstName":"Maartje","middleName":"","lastName":"Henderikx","suffix":""},{"id":566593513,"identity":"a3092b97-2f85-4cda-b6c0-7d2ca030f569","order_by":2,"name":"Karel Kreijns","email":"","orcid":"","institution":"Open University of the Netherlands","correspondingAuthor":false,"prefix":"","firstName":"Karel","middleName":"","lastName":"Kreijns","suffix":""},{"id":566593514,"identity":"967580c5-cf6c-4c9b-a4d1-f5c170fcee17","order_by":3,"name":"Rolands Klemke","email":"","orcid":"","institution":"Open University of the Netherlands","correspondingAuthor":false,"prefix":"","firstName":"Rolands","middleName":"","lastName":"Klemke","suffix":""}],"badges":[],"createdAt":"2025-12-24 14:38:15","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8443551/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8443551/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":99796742,"identity":"f59e035e-60f7-409b-9b65-6bdaa08a479b","added_by":"auto","created_at":"2026-01-08 13:43:20","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":284749,"visible":true,"origin":"","legend":"","description":"","filename":"ArticleLLMbasedSupportAgent23122025submittedVersion.docx","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/4562a70a01ee526fc87c1be7.docx"},{"id":99797450,"identity":"e1631985-0ef8-4909-85ac-29a1dd0da8c8","added_by":"auto","created_at":"2026-01-08 13:45:51","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6328,"visible":true,"origin":"","legend":"","description":"","filename":"ab6aa9eb31d549b1a05bf0dc9826d1f1.json","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/c5a70753bd6edac32089d2e2.json"},{"id":99797600,"identity":"c51c2cc6-3d86-4bb8-99ae-1ea74f2b0ad1","added_by":"auto","created_at":"2026-01-08 13:46:05","extension":"xml","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":168314,"visible":true,"origin":"","legend":"","description":"","filename":"ab6aa9eb31d549b1a05bf0dc9826d1f11enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/f765594fa974379d4643a326.xml"},{"id":99797133,"identity":"c4156186-4b7c-4097-bcbe-d9041ee7d45e","added_by":"auto","created_at":"2026-01-08 13:45:17","extension":"eps","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":597,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage3.eps","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/875ac8892e320c0b376e3140.eps"},{"id":99797084,"identity":"d925f200-739d-46bc-a327-b9ce98da8f51","added_by":"auto","created_at":"2026-01-08 13:44:33","extension":"eps","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":129409,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage4.eps","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/aed810d03b2995c32989e787.eps"},{"id":99797049,"identity":"c91b6b90-369b-43c3-ab29-7f7906816aa3","added_by":"auto","created_at":"2026-01-08 13:44:29","extension":"png","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":128274,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/8cb5d6571aad1d82fa0383aa.png"},{"id":99718741,"identity":"cdb1c33a-d790-4560-a302-56e2cb1dec62","added_by":"auto","created_at":"2026-01-07 15:04:46","extension":"png","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1536,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/de409b319cc90467d09201fd.png"},{"id":99718745,"identity":"a3b99423-44ea-49eb-be98-8ee9b6c5a4ef","added_by":"auto","created_at":"2026-01-07 15:04:46","extension":"png","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":26313,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/2226d1449f9ed2582146b498.png"},{"id":99796834,"identity":"8e3c0a0f-ef7d-4aad-84d0-654a788d14b6","added_by":"auto","created_at":"2026-01-08 13:43:46","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1078,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/ca20b60f799c805a1b95acee.png"},{"id":99718747,"identity":"cdf971f5-31b0-44e5-a253-9c9f0a06d2d5","added_by":"auto","created_at":"2026-01-07 15:04:46","extension":"xml","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":167169,"visible":true,"origin":"","legend":"","description":"","filename":"ab6aa9eb31d549b1a05bf0dc9826d1f11structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/f45151e3688d78b54e376abf.xml"},{"id":99798065,"identity":"d8af0429-3124-477d-a6a0-bed62dd14f05","added_by":"auto","created_at":"2026-01-08 13:47:10","extension":"html","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":180759,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/9f9780c2b7cdace2a14c8de5.html"},{"id":99718732,"identity":"fc300d6b-6486-4f54-9721-6f5974537d3d","added_by":"auto","created_at":"2026-01-07 15:04:46","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":274970,"visible":true,"origin":"","legend":"\u003cp\u003eExample of an interaction with GroupBuddy on the Discord platform: a user tagged GroupBuddy and asked for advice, and GroupBuddy responded\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/0b78d9793111a57e1df715e7.png"},{"id":99718738,"identity":"327ed74e-b700-4fd0-91a0-85b6d61bc093","added_by":"auto","created_at":"2026-01-07 15:04:46","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":24166,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of chatbot responses (n = 62) across regulation categories, with SSRL being the most frequently supported mode.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/b1b8e68e26093007457e0b3a.png"},{"id":99718736,"identity":"eda41486-3cbb-4ab1-b12f-32fc9a78a652","added_by":"auto","created_at":"2026-01-07 15:04:46","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":40779,"visible":true,"origin":"","legend":"\u003cp\u003eDistribution of SUS scores across participants (n=15). The dashed line marks the industry average usability threshold of 68, placing the chatbot in the “OK” range and revealing participant-level variability\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/77e31a0b672681181845e499.png"},{"id":99797363,"identity":"a289bbe8-b887-4c6b-af93-ef37e870c713","added_by":"auto","created_at":"2026-01-08 13:45:40","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":81011,"visible":true,"origin":"","legend":"\u003cp\u003eItem-level mean ratings (±SD) for the ten SUS questions. Each label includes the item’s underlying usability theme (e.g., confidence, learn quickly, unnecessary complexity). Ratings were given on a 5-point scale (1 = Completely disagree, 5 = Completely agree)\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/26f27d7e65750a0f162b7e84.png"},{"id":99797125,"identity":"d0db1de5-d80a-4c9c-8f1f-a3d5167feece","added_by":"auto","created_at":"2026-01-08 13:45:16","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":59440,"visible":true,"origin":"","legend":"\u003cp\u003ePerceived usefulness by item (n = 18; 1 = Completely disagree, 6 = Completely agree). Bars show item means with ±SD error bars. Items include motivation/fun, general advice, collaboration support, emotion-regulation support, and socio-emotional awareness\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/35378d7b76eda52367c04f36.png"},{"id":105035567,"identity":"c075bed2-c3c3-4aec-b65f-04197ab5518d","added_by":"auto","created_at":"2026-03-20 07:26:16","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1543079,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/92b2b2cb-d56f-4cbb-8662-e1a16a4b2fc3.pdf"},{"id":99718734,"identity":"2e1dfb93-5587-441e-8f71-21b90880acef","added_by":"auto","created_at":"2026-01-07 15:04:46","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":25432,"visible":true,"origin":"","legend":"","description":"","filename":"Appendixs.docx","url":"https://assets-eu.researchsquare.com/files/rs-8443551/v1/0285e9dec6c25510470ab4a4.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Designing Emotion Regulation Support in Online Group Learning: Insights from an LLM-Based Support Agent","fulltext":[{"header":"Background","content":"\u003cp\u003eGroup learning has the potential to support both learning and social outcomes; however, students may encounter socio-emotional challenges that disrupt collaboration (Hadwin et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Kreijns et al., \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Hassane et al. (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) investigated socio-emotional challenges and social emotions in the context of OGL. The most common socio-emotional challenges identified included external constraints related to a learner\u0026rsquo;s personal life, different understandings of the task or concept, and the presence of group members who were not fully committed or engaged in free-riding. Most learners (96%) reported encountering socio-emotional challenges, with 86% experiencing multiple challenges during OGL (Hassane et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). According to this study, most students (74%) reported negative emotions, whereas only 26% reported positive emotions. Moreover, learners experienced nearly three times as many negative emotions in OGL as in face-to-face group learning (Hassane et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Taken together, these findings indicate that socio-emotional challenges in OGL settings are frequently associated with negative social emotions, which can undermine productive collaboration. Hence, regulating negative emotions is essential for a productive learning environment (Hassane et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn the context of collaborative learning, self-regulation (SRL) is crucial for effective learning, as it encompasses both cognitive learning processes and emotions (Harley et al., \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Particularly in online settings where social interaction is more challenging, ER becomes more challenging (Hassane et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Kreijns et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). In the educational context, ER refers to becoming aware of emotional sensations and to monitoring the intensity and/or duration of emotional experiences to maintain productive engagement in learning activities (Gross, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Hadwin et al. \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Hassane et al. (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) examined learners\u0026rsquo; ER in OGL through the lens of Gross\u0026rsquo; (\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2015\u003c/span\u003e) process model of emotion regulation (PMER). The PMER outlines five families of ER strategies that individuals can use to influence their emotional experiences: situation selection, situation modification, attentional deployment, cognitive change, and response modulation (see Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e for definitions of each strategy family and illustrative examples in the context of OGL). Based on this model, Hassane et al. (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) identified the strategies learners inherently employ when encountering socio-emotional challenges during OGL. These strategies may either support productive collaboration (e.g., focusing on manageable task elements or reappraising peer feedback positively) or hinder collaboration (e.g., avoiding group learning activities altogether).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eDefinitions of the Five ER Families and Illustrative Examples in Group Learning.\u003c/p\u003e \u003cdiv class=\"Credit\"\u003e\u003cp\u003eAdapted from Hassane et al. (\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e)\u003c/p\u003e\u003c/div\u003e\u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"2\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEmotion Regulation Families\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eExample during Group Learning\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSituation Selection: Taking actions that make it more (or less) likely that one will be in a situation that one expects will give rise to desirable (or undesirable) emotions (Gross, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2015\u003c/span\u003e, p. 7).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAvoid engaging in group learning activities altogether.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSituation Modification: Taking actions that directly alter a situation to change its emotional impact (Gross, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2015\u003c/span\u003e, p. 8).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eInitiating a conversation to address and diminish tension within the group\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAttentional Deployment: Directing one\u0026rsquo;s attention to influence one\u0026rsquo;s emotional response (Gross, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2015\u003c/span\u003e, p. 8)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eRedirecting attention towards manageable task elements to reduce emotional overload.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCognitive Change: Modifying one\u0026rsquo;s appraisal of a situation to alter its emotional impact (Gross, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2015\u003c/span\u003e, p. 9).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eViewing repeated peer feedback as a chance to clarify your own thinking, rather than as constant criticism.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eResponse Modulation: Directly influencing experiential, behavioral, or physiological components of the emotional response after the emotion is well developed (Gross, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2015\u003c/span\u003e, p. 9).\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eForcing a smile during a tense group discussion to avoid escalating conflict.\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eER is integral to group learning (J\u0026auml;rvenoja et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). However, Gross\u0026rsquo;s PMER merely emphasizes \u003cem\u003eindividual ER\u003c/em\u003e (self-regulation). Due to the social nature of collaboration, regulation extends beyond individual ER. Thus, two other regulation modes have been identified (Hadwin et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). These include \u003cem\u003eco-regulation\u003c/em\u003e (CoRL), in which a learner supports peers in regulating their emotions by acting as a mediator, and \u003cem\u003esocially shared regulation\u003c/em\u003e (SSRL), in which group members collectively manage emotional dynamics to support productive collaboration (Hadwin et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2017\u003c/span\u003e). Recent research highlighted positive correlations between different regulation modes in OGL and noted CoRL\u0026rsquo;s role in mediating peers\u0026rsquo; negative emotions to support emotional stability and productive collaboration (Hassane et al., \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Given the importance of individual, co-, and socially shared emotion regulation for productive OGL, researchers have increasingly explored how technology can be used to support these regulatory processes (J\u0026auml;rvenoja et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn this context, advances in technology have further fueled the interest in technology-based ER support in OGL (Ngo et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Nguyen et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Rojas et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). A substantial body of CSCL research has proposed Group Awareness (GA) support that informs learners about other group members\u0026rsquo; activities, knowledge, and emotions. (Henderikx and Kreijns, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Kirschner et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Miller and Hadwin, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Shingjergji et al., \u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e2026\u003c/span\u003e; Chen et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Early GA tools illustrate how GA can promote interaction and support group processes. Group processes refer to how group members coordinate, monitor, and reflect on their collaborative work and social interactions. For example, Kirschner et al. (\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2015\u003c/span\u003e) introduced tools such as Radar and Reflector to enhance awareness of group members\u0026rsquo; social and cognitive behaviors. These tools stimulate reflection on how the group is functioning, including participation patterns, collaboration quality, and the alignment between individual contributions and group goals. Chen et al. (\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) examined the effectiveness of GA support through a meta-analysis of 46 empirical studies. The results showed that GA support positively influences behavioral participation, cognitive development, and social emotion. However, the effects of GA support are minor for social emotion compared to the other two dimensions (Chen et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eAnother form of ER support proposed for OGL is scripting. Scripting refers to the design of structured prompts or guidance that support learners at different phases of collaboration by directing their attention to essential aspects of the task and group interaction (Miller and Hadwin, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). Studies showed that scripting can support learners in setting goals related to emotions, such as maintaining positive feelings like optimism and confidence, and reducing negative emotions such as anxiety and stress during collaboration (Li et al., \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Miller and Hadwin, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2015\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eMore recently, advances in artificial intelligence have led to the development of AI-based ER support tools for OGL. These tools extend earlier awareness-based approaches by aiming to infer learners\u0026rsquo; emotional states from biometric or behavioral cues, including facial expressions, posture, voice, and attention patterns, to enhance emotional awareness and provide more timely ER support (Ngo et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Nguyen et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). AI-supported emotion detection can improve understanding of emotional regulation and socially shared regulation of learning in synchronous online learning environments (Ngo et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2024\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eUpon reviewing state-of-the-art studies on technological ER support in OGL (Hassane, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), most rely on awareness tools operating under the assumption that the regulation of the emotion will occur automatically once group members are aware of their emotions (Slovak et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). However, Kreijns et al. (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) showed that learners do not necessarily regulate their emotions or select an effective ER strategy by themselves. This highlights the need for support that goes beyond awareness by implementing interventions that actively support the regulation process (Hassane, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Studies in the context of Human-Computer Interaction (HCI) noted that effective technological interventions should explicitly incorporate validated theoretical frameworks (Kitson et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Slovak et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Upon this, they suggested a framework for designing ER interventions by emphasizing theory-informed approaches grounded in Gross\u0026rsquo; PMER (Gross, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2015\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eBased on the background presented above, it follows that technologies designed to support ER in OGL should go beyond mere emotion awareness. Specifically, such technologies should provide ER support that is grounded in the following established ER modes:\u003c/p\u003e \u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eIndividual ER\u003c/b\u003e: Supporting individual ER grounded in Gross\u0026rsquo; PMER (Gross, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2015\u003c/span\u003e).\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eEnhancing SSRL\u003c/b\u003e: Facilitating increased social interaction and collaborative activities through targeted group interventions (Hadwin et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; J\u0026auml;rvenoja et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003e \u003cb\u003eFacilitating CoRL\u003c/b\u003e: Supporting learners to act as mediators toward assisting learners in their individual ER (Hadwin et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; J\u0026auml;rvenoja et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2023\u003c/span\u003e).\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e \u003cp\u003eTo take a first step toward designing such ER intervention tools for OGL, we developed a GPT-4\u0026ndash;based chatbot deployed in an experimental OGL setting as an ER support agent. The chatbot was configured and positioned as a pedagogical agent that supports ER processes within the OGL environment.\u003c/p\u003e \u003cp\u003eThrough mixed-methods analysis, we examined 1. whether the general GPT-4-based chatbot interactions reflected OGL ER theories (Individual ER, CoRL, SSRL) in its responses and 2. how participants experienced its usability and usefulness. The findings provide evidence for improvements to design requirements for developing LLM-based pedagogical agents for ER in OGL.\u003c/p\u003e"},{"header":"Methods","content":"\u003ch2\u003eParticipants and context\u003c/h2\u003e\n\u003cp\u003eParticipants were PhD students, educators, and researchers in technology-enhanced learning, ranging in age, gender, and experience level from junior to senior (n = 20). They were all participating in a workshop as part of the 18th EATEL Summer School on Technology-Enhanced Learning (JTELSS), where they learned how LLMs such as ChatGPT can be customized into interactive, conversational assistants.\u003c/p\u003e\n\u003cp\u003eThe first part of the workshop focused on introducing large language models (LLMs) for creating chatbots that can support and assist learners and educators in various educational settings, as well as on the basics of prompting. The second part of the workshop was dedicated to demonstrating and testing the first version of a chatbot that was explicitly designed for ER in OGL.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eProcedure\u003c/h2\u003e\n\u003cp\u003eAt the beginning of the experiment, participants were introduced to the context and briefed on the tool they would be using: a chatbot named GroupBuddy. They were informed that GroupBuddy was designed to assist with ER and socio-emotional challenges, particularly in situations involving frustration, disengagement, or emerging conflict. Following this briefing, informed consent was obtained for the anonymous use of chat data for research purposes.\u003c/p\u003e\n\u003cp\u003eParticipants were then asked to work collaboratively in four groups. Their specific task was to design a prompt that could automate a pedagogical chatbot for a real-world use case of their choice. Each group should discuss potential applications and select a practical example, such as creating a chatbot for grading and personalized feedback. This activity required joint decision-making, idea negotiation, and coordination, thereby naturally creating opportunities to address the socio-emotional challenges typical of OGL. To simulate an OGL setting, participants were instructed to collaborate from different locations within the venue. The experiment lasted approximately 45 minutes, and at the end, a plenary discussion was held in which participants shared their impressions of GroupBuddy\u0026rsquo;s behavior and impact on group dynamics, as well as its role, tone, and usefulness during collaboration.\u003c/p\u003e\n\u003ch2\u003eChatbot implementation\u003c/h2\u003e\n\u003ch3\u003eThe AI model and development\u003c/h3\u003e\n\u003cp\u003eGroupBuddy was developed using JavaScript and Node.js and connected to the OpenAI API. It utilizes the GPT-4 model (version 0314), which was selected for its 2024 state-of-the-art performance on language-understanding benchmarks and human-like dialogue capabilities (OpenAI et al., 2024; Ou et al., 2024). The source code is openly available on GitHub (see Data and Code Availability section). Full deployment specifications are provided in Appendix A.\u003c/p\u003e\n\u003cp\u003eA zero-shot prompting (Y. Li, 2023) approach was employed, in which the model received instructional prompts without specific examples, relying instead on its pre-existing knowledge. The model was instructed to act as a pedagogical agent that supports group members experiencing negative emotions or socially challenging situations, such as differing goals and unequal participation, as described by Hassane et al. (2025). The chatbot was restricted to responding only to messages indicating negative emotions or group tension and was prevented from offering academic or task-related assistance.\u003c/p\u003e\n\u003ch3\u003ePlatform and integration\u003c/h3\u003e\n\u003cp\u003eGroupBuddy was integrated into Discord (\u0026ldquo;Discord\u0026rdquo;, 2024) (see Fig. 1). Discord was chosen because it creates an engaging space for social interactions through rich affordances such as threaded conversations, ready-made emojis, animations, and file-sharing features (Almomani, 2024). Furthermore, the platform offers high technical flexibility for embedding conversational LLMs within its group messaging environment (channels).\u003c/p\u003e\n\u003cp\u003eWithin this environment, the chatbot operated on two levels. It could post messages on the group channel either proactively (when negative emotions appeared in the chat) or reactively (when users mentioned its name).\u003c/p\u003e\n\u003ch2\u003eParticipants and context\u003c/h2\u003e\n\u003cp\u003eParticipants were PhD students, educators, and researchers in technology-enhanced learning, ranging in age, gender, and experience level from junior to senior (n = 20). They were all participating in a workshop as part of the 18th EATEL Summer School on Technology-Enhanced Learning (JTELSS), where they learned how LLMs such as ChatGPT can be customized into interactive, conversational assistants.\u003c/p\u003e\n\u003cp\u003eThe first part of the workshop focused on introducing large language models (LLMs) for creating chatbots that can support and assist learners and educators in various educational settings, as well as on the basics of prompting. The second part of the workshop was dedicated to demonstrating and testing the first version of a chatbot that was explicitly designed for ER in OGL.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eProcedure\u003c/h2\u003e\n\u003cp\u003eAt the beginning of the experiment, participants were introduced to the context and briefed on the tool they would be using: a chatbot named GroupBuddy. They were informed that GroupBuddy was designed to assist with ER and socio-emotional challenges, particularly in situations involving frustration, disengagement, or emerging conflict. Following this briefing, informed consent was obtained for the anonymous use of chat data for research purposes.\u003c/p\u003e\n\u003cp\u003eParticipants were then asked to work collaboratively in four groups. Their specific task was to design a prompt that could automate a pedagogical chatbot for a real-world use case of their choice. Each group should discuss potential applications and select a practical example, such as creating a chatbot for grading and personalized feedback. This activity required joint decision-making, idea negotiation, and coordination, thereby naturally creating opportunities to address the socio-emotional challenges typical of OGL. To simulate an OGL setting, participants were instructed to collaborate from different locations within the venue. The experiment lasted approximately 45 minutes, and at the end, a plenary discussion was held in which participants shared their impressions of GroupBuddy\u0026rsquo;s behavior and impact on group dynamics, as well as its role, tone, and usefulness during collaboration.\u003c/p\u003e\n\u003ch2\u003eChatbot implementation\u003c/h2\u003e\n\u003ch3\u003eThe AI model and development\u003c/h3\u003e\n\u003cp\u003eGroupBuddy was developed using JavaScript and Node.js and connected to the OpenAI API. It utilizes the GPT-4 model (version 0314), which was selected for its 2024 state-of-the-art performance on language-understanding benchmarks and human-like dialogue capabilities (OpenAI et al., 2024; Ou et al., 2024). The source code is openly available on GitHub (see Data and Code Availability section). Full deployment specifications are provided in Appendix A.\u003c/p\u003e\n\u003cp\u003eA zero-shot prompting (Y. Li, 2023) approach was employed, in which the model received instructional prompts without specific examples, relying instead on its pre-existing knowledge. The model was instructed to act as a pedagogical agent that supports group members experiencing negative emotions or socially challenging situations, such as differing goals and unequal participation, as described by Hassane et al. (2025). The chatbot was restricted to responding only to messages indicating negative emotions or group tension and was prevented from offering academic or task-related assistance.\u003c/p\u003e\n\u003ch3\u003ePlatform and integration\u003c/h3\u003e\n\u003cp\u003eGroupBuddy was integrated into Discord (\u0026ldquo;Discord\u0026rdquo;, 2024) (see Fig. 1). Discord was chosen because it creates an engaging space for social interactions through rich affordances such as threaded conversations, ready-made emojis, animations, and file-sharing features (Almomani, 2024). Furthermore, the platform offers high technical flexibility for embedding conversational LLMs within its group messaging environment (channels).\u003c/p\u003e\n\u003cp\u003eWithin this environment, the chatbot operated on two levels. It could post messages on the group channel either proactively (when negative emotions appeared in the chat) or reactively (when users mentioned its name).\u003c/p\u003e\n\u003ch2\u003eMaterials\u003c/h2\u003e\n\u003ch3\u003eChatbot interaction data\u003c/h3\u003e\n\u003cp\u003eChat interactions from group members and GroupBuddy were automatically logged during the group activity. These chat logs were then used as a qualitative data source to examine whether the default GPT-4 model exhibited patterns consistent with established OGL ER modes, including individual ER (based on Gross\u0026rsquo;s PMER(Gross, 2015)), CoRL and SSR (Hadwin et al., 2017; J\u0026auml;rvenoja et al., 2023). All logs were anonymized immediately after collection; no personal identifiers were retained. The final dataset comprised n = 75 chatbot messages across four group channels.\u003c/p\u003e\n\u003ch3\u003eVerbal feedback data\u003c/h3\u003e\n\u003cp\u003eThe discussion at the end of the experiment was audio-recorded and later transcribed verbatim for analysis.\u003c/p\u003e\n\u003ch3\u003eUsability questionnaires\u003c/h3\u003e\n\u003cp\u003eTo assess the \u003cem\u003eusability\u003c/em\u003e of GroupBuddy, an adapted version of the \u003cem\u003eSystem Usability Scale\u0026nbsp;\u003c/em\u003e(SUS) (Brooke, 1996) was used. Fifteen participants completed this questionnaire, which contained ten closed-ended items rated on a 5-point Likert scale (1 = Completely disagree, 5 = Completely agree). The scale evaluated dimensions such as ease of use, complexity, consistency, and user confidence. Example statements included \u0026ldquo;I think that I would like to use the chatbot frequently,\u0026rdquo; \u0026ldquo;I felt very confident using the chatbot,\u0026rdquo; and \u0026ldquo;I thought the chatbot was easy to use\u0026rdquo;. SUS scoring followed standard procedures, yielding a total usability score ranging from 0 to 100 (Brooke, 1996). These scores were interpreted using established SUS benchmarks to classify perceived usability: \u0026lt;51 = poor, 51 70 = OK, 71-84 = good, \u0026gt; 85 = excellent (Aaron, 2009).\u003c/p\u003e\n\u003ch3\u003eUsefulness questionnaires\u0026nbsp;\u003c/h3\u003e\n\u003cp\u003eTo evaluate the chatbot\u0026rsquo;s perceived \u003cem\u003eusefulness\u003c/em\u003e, a custom questionnaire adapted from Davis (1989) was administered to the participants. The survey consisted of 8 closed-ended items rated on a 6-point Likert scale (1 = Completely disagree, 6 = Completely agree) that addressed emotions and support for collaboration, ER, and socio-emotional awareness. Sample items from the survey are \u0026ldquo;The chatbot functionality supported collaboration positively\u0026rdquo; and \u0026ldquo;The chatbot functionality enhanced socio-emotional awareness within the group.\u0026rdquo; In addition, two open-ended questions were included to gather qualitative feedback: \u0026ldquo;What did you find most positive about the chatbot functionality?\u0026rdquo; and \u0026ldquo;What would you like to see different in the chatbot functionality?\u0026rdquo;\u003c/p\u003e\n\u003ch2\u003eData Analysis\u003c/h2\u003e\n\u003cp\u003eA mixed-methods approach was used to integrate quantitative and qualitative data, allowing triangulation between user evaluations, verbal feedback, and chatbot interaction logs. Quantitative data from the questionnaires were summarized descriptively, while qualitative data were analyzed thematically through theory-informed deductive and inductive content analysis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eQuantitative analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData from the usability and usefulness questionnaires were used to calculate descriptive statistics; for each item, mean (M) and standard deviation (SD) were calculated to capture central tendencies and variation in participants\u0026rsquo; responses. Item-level statistics were visualized to identify aspects of the chatbot that participants found particularly strong or in need of improvement. Given the small sample size and the exploratory nature of the study, no inferential statistical tests were performed. Instead, the analysis focused on identifying patterns and trends across measures to gain insights into the chatbot\u0026rsquo;s performance and user perceptions.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eQualitative analysis\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData from the open-ended survey questions, verbal feedback, and chatbot interaction logs were analyzed to capture participants\u0026rsquo; subjective experiences and to evaluate the theoretical grounding of the chatbot\u0026rsquo;s responses. This triangulation was key to understanding both user experience and tool efficacy.\u003c/p\u003e\n\u003cp\u003eThematic analysis was applied to all three data sources (Braun and Clarke, 2006). After familiarization with the verbal and open-ended survey response data, responses were inductively coded and grouped into themes representing key aspects of participants\u0026apos; experiences. Representative quotes were selected to illustrate each resulting theme.\u003c/p\u003e\n\u003cp\u003eIn addition, the analysis of the chatbot interaction logs followed a separate theory-informed deductive coding process. This approach was used to determine the extent to which the chatbot\u0026rsquo;s responses aligned with established theories of ER in OGL, as discussed in the theoretical background. Each message was examined against the codebook (see Table 2 for a summary of the coding categories and Appendix B for the extended codebook) to determine whether it reflected one or more of the targeted ER modes. Corresponding codes were assigned to responses that aligned with theoretical definitions.\u003c/p\u003e\n\u003cp\u003eThis deductive coding process was conducted through a human\u0026ndash;LLM collaborative method. The decision to use an LLM, specifically ChatGPT, as a second coder was based on its promising potential to serve as a researcher and accelerate coding qualitative data. Several studies indicate that LLMs can efficiently perform thematic analysis comparable to human coders (Jiang et al., 2021; Liu et al., 2025), with one study indicating an alignment with human coding up to 96.02% for broader themes (Liu et al., 2023). However, the most effective approach, as was used in this study, is the hybrid model, where AI supplements human expertise (Jiang et al., 2025) by providing efficiency while the human preserves critical interpretive judgment and ensures crucial human oversight (Parkington et al., 2025).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThis approach draws on procedures described by Chew et al. (2023) and Xiao et al. (2023). First, the (human) researcher created the theory-based codebook, formatted it as a prompt, and provided it to ChatGPT 01. Then, the LLM labeled a subset of 10% of the dataset using the provided codebook (O\u0026rsquo;Connor and Joffe, 2020), while the human researcher independently coded the same subset. Inter-rater reliability was calculated using Cohen\u0026rsquo;s \u0026kappa; to assess the consistency between human and LLM coding. It took two iterations of refining the codebook and the prompt to achieve a satisfactory level of agreement (\u0026kappa; = 0.85), after which the LLM coded the remaining data using the finalized codebook (see Appendix C for the whole prompt text).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2\u003c/strong\u003e Summary of coding categories for ER in chatbot responses, with corresponding theoretical definitions\u003c/p\u003e\n\u003cp\u003e\u003cimg src=\"data:image/png;base64,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\" width=\"718\" height=\"387\"\u003e\u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eThis section presents findings from the content analysis of chatbot responses, the usability and usefulness surveys, and the thematic analysis of participants\u0026rsquo; qualitative feedback. Results are organized around two focal areas: (1) the theoretical alignment of chatbot responses with ER modes in OGL, (2) users\u0026rsquo; perceptions of usability and usefulness, and emergent qualitative themes that inform future design improvements.\u003c/p\u003e\n\u003ch2\u003eTheory-based Intervention\u003c/h2\u003e\n\u003cp\u003eThe chatbot response dataset comprised n = 75 groupBuddy outputs across the four Discord group channels. These outputs were produced during the collaborative activity period (~45 min, as per procedure), totaling ~180 group-minutes of interaction. This corresponds to approximately 1.67 chatbot messages per minute across all channels, or 0.42 messages per minute per group (~1 message every 2.4 minutes). Message length was variable (M = 118.4 words, SD = 78.5; range = 8:305), suggesting that the chatbot typically responded with multi-sentence supports.\u003c/p\u003e\n\u003cp\u003eOf these n = 75 outputs, 62 (82.7%) contained at least one ER mode and were coded as an intervention using the regulation modes in Table 2. The remaining 13 (17.3%) responses consisted of greetings and acknowledgements and were left as \u0026lsquo;uncoded\u0026rsquo;. Examples of output coded as \u003cem\u003euncoded\u003c/em\u003e, \u003cem\u003eSSRL\u003c/em\u003e, \u003cem\u003eCoRL\u003c/em\u003e, and \u003cem\u003eIER\u003c/em\u003e are provided in Table 6 in Appendix D.\u003c/p\u003e\n\u003cp\u003eAcross the 62 interventions, multi-label coding yielded 97 regulation-code instances (M = 1.56 codes per intervention). Half of the interventions were single-coded (50.0%), while 43.5% received two codes and 6.5% received three codes (SSRL + CoRL + individual ER). At the code occurrence level, SSRL accounted for 47.4%, CoRL for 39.2%, and individual ER for 13.4% (see Fig. 2).\u003c/p\u003e\n\u003ch2\u003eUsability and Usefulness Evaluation\u003c/h2\u003e\n\u003cp\u003eThe overall SUS score, based on 15 item means, was estimated at 64.7, indicating a \u0026ldquo;OK\u0026apos; level of usability according to Aaron (2009). Figure 3 illustrates the distribution of SUS scores across participants, revealing that about one-third rated it at or above the industry-average benchmark of 68 (Brooke, 1996). Figure 4 displays mean scores and standard deviations for the ten SUS questions, categorized by themes such as frequent use, unnecessary complexity, and ease of use. Strengths were noted in ease of use (Q3, M = 4.13, SD = 0.83, 95% CI [3.67, 4.60]) and learnability (Q7, M = 3.80, SD = 0.86, 95% CI [3.32, 4.28]), with positive ratings also for integration (Q5, M = 3.60, SD = 1.06, 95% CI [3.02, 4.19]). Willingness to use frequently was low (Q1, M = 2.87, SD = 0.83, 95% CI [2.40, 3.33]), and confidence levels were moderate (Q9, M = 3.47, SD = 0.92, 95% CI [2.96, 3.97]). Conversely, perceptions of unnecessary complexity (Q2, M = 2.47, SD = 1.19, 95% CI [1.81, 3.12]) and learning required (Q10, M = 2.20, SD = 1.21, 95% CI [1.53, 2.87]) showed the greatest variability, indicating mixed user experiences (see Fig. 4)\u003c/p\u003e\n\u003cp\u003ePerceived usefulness ratings are generally positive, especially for experienced fun and perceived advice quality. Fun (M = 4.43, SD = 1.22, 95% CI [3.82, 5.04]) and helpful general advice (M = 4.14, SD = 1.25, 95% CI [3.52, 4.76]) scores above the midpoint. Conversely, items related to collaboration and group-dynamics facilitation are more modest (collaboration M = 3.43, SD = 1.58, 95% CI [2.64, 4.22]; discussing group dynamics M = 3.29, SD = 1.49, 95% CI [2.55, 4.03]), with wider confidence intervals indicating varied experiences across teams. Pattern analysis in Figure 5 shows that fun/motivation and general advice tend to cluster at higher levels. At the same time, collaboration and socio-emotional dynamics items are closer to the scale\u0026apos;s midpoint, indicating greater variation.\u003c/p\u003e\n\u003ch2\u003eOpen-ended questions and post-session debrief\u003c/h2\u003e\n\u003cp\u003eWe conducted a thematic analysis of two open-ended survey questions, \u0026ldquo;What did you find most positive about the chatbot functionality?\u0026rdquo; and \u0026ldquo;What would you like to see different in the chatbot functionality?,\u0026rdquo; completed by 16 participants, and of transcribed post-session verbal debrief feedback from an additional 5 participants. The combined qualitative data (n = 21) revealed a coherent set of themes, summarized in Tables 3 (perceived value and positive aspects) and 4 (suggested improvements and concerns).\u003c/p\u003e\n\u003cp\u003eRegarding perceived value (Table 3), participants most often noted the helpfulness and coherence of advice or ideas (29%), along with an emotionally supportive tone described as calming, constructive, or sometimes motivating (24%). Less common themes included appreciation for proactive or easily accessible support when it was not overbearing (10%), novelty or enjoyment (10%), limited benefits for coordination or focus (5%), and the suitability for distributed or long-term group work contexts (5%).\u003c/p\u003e\n\u003cp\u003eWith respect to suggested improvements (Table 4), the most salient concerns were intrusiveness, particularly requests for fewer, better-timed messages or responses only when needed (33%), as well as shorter messages to reduce perceived verbosity (24%). Participants also noted issues related to timing or latency, such as a desire for faster responses or intervention at critical moments (14%), and to role clarity and conversational ordering, including ambiguity between moderator and peer roles and disruptions to group chat flow (14%). Less frequently reported but notable concerns included maintaining user control following explicit \u0026ldquo;stop\u0026rdquo; requests (5%), improving sensitivity by reducing oversensitivity to negative language or better handling of irony (5%), occasional accuracy issues (1 of 21; 5%), and minor feature requests, such as the inclusion of playful elements or GIFs (5%).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3\u0026nbsp;\u003c/strong\u003ePositive/value themes across open-ended usefulness responses and post-session verbal feedback, pooled (n = 21). Themes are non-exclusive; prevalence is shown as a percentage. Selected representative quotes are included\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"624\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTheme\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 182px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n \u003cp\u003e\u003cstrong\u003eDefinition (coding rule)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRepresentative quote\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMentions (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHelpful, coherent advice \u0026amp; ideas\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 182px;\"\u003e\n \u003cp\u003eUseful, on-point answers; context understanding; idea generation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003e\u003cins cite=\"mailto:Kreijns,%20Karel\" datetime=\"2025-11-18T09:17\"\u003e\u0026ldquo;\u003c/ins\u003euseful\u0026hellip; recommendations are good\u0026rdquo;; \u0026ldquo;good understanding of the context\u0026rdquo;; \u0026ldquo;coherent answers\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e6 (29%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eEmotion-supportive tone\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 182px;\"\u003e\n \u003cp\u003eCalming/motivating; constructive presence; responds to emotions\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003e\u0026ldquo;tries to make peace\u0026rdquo;; \u0026ldquo;help to make\u0026hellip; calm\u0026rdquo;; \u0026ldquo;motivating messages\u0026rdquo;; \u0026ldquo;reaction to our emotions\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e5 (24%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eProactive/available support\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 182px;\"\u003e\n \u003cp\u003eTakes initiative to help (perceived positively when not excessive)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003e\u0026ldquo;initiative and proactive approaches\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e2 (10%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eCoordination/focus\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 182px;\"\u003e\n \u003cp\u003eNudge group to focus on specific aspects\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003ctable border=\"0\" cellspacing=\"3\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\u003cbr\u003e\u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003ctable border=\"0\" cellspacing=\"3\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd\u003e\n \u003cp\u003e\u0026ldquo;coordinate the group and make it focus\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n \u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e1 (5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNovelty/enjoyment\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 182px;\"\u003e\n \u003cp\u003eNew, engaging, or \u0026ldquo;cute/fun\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003e\u0026ldquo;different experience\u0026rdquo;; \u0026ldquo;answers were cute\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e2 (10%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePotential fit for distributed use\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 182px;\"\u003e\n \u003cp\u003ePerceived value for longer, distributed work\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 173px;\"\u003e\n \u003cp\u003e\u0026ldquo;good for distributed group working\u0026hellip; for a long time\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 142px;\"\u003e\n \u003cp\u003e1 (5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4\u0026nbsp;\u003c/strong\u003eImprovement/concern themes across open-ended usefulness responses and post-session debrief, pooled (n = 21). Themes are non-exclusive; prevalence shown as a percentage. Selected representative quotes included\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTheme\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 194px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDefinition (coding rule)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 166px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRepresentative quote\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 136px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMentions (%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eIntrusiveness/reply frequency \u0026amp; proactivity\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 194px;\"\u003e\n \u003cp\u003eToo many messages; responds untagged; should reply only when needed; crowds out group talk\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 166px;\"\u003e\n \u003cp\u003e\u0026ldquo;too much\u0026hellip; responded to basically everything\u0026rdquo;; \u0026ldquo;less invasive\u0026rdquo;; \u0026ldquo;not too proactive\u0026hellip; only when needed\u0026rdquo;; \u0026ldquo;responded too frequently\u0026mdash;even when not tagged\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 136px;\"\u003e\n \u003cp\u003e7 (33%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVerbosity/message length\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 194px;\"\u003e\n \u003cp\u003eTexts too long; hard to read\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 166px;\"\u003e\n \u003cp\u003e\u0026ldquo;answers were too long\u0026rdquo;; \u0026ldquo;shorter texts\u0026rdquo;; verbal: \u0026ldquo;generated text was long\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 136px;\"\u003e\n \u003cp\u003e5 (24 %)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eTiming/latency\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 194px;\"\u003e\n \u003cp\u003eSlow or mistimed; needs timely help at critical moments\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 166px;\"\u003e\n \u003cp\u003e\u0026ldquo;appearance time\u0026hellip; to answer\u0026rdquo;; \u0026ldquo;quick and dirty\u0026hellip; responses\u0026rdquo;; \u0026ldquo;timely responses at critical moments\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 136px;\"\u003e\n \u003cp\u003e3 (14%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRole clarity/conversation order\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 194px;\"\u003e\n \u003cp\u003eActs like peer vs. moderator; need ordering/structure\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 166px;\"\u003e\n \u003cp\u003everbal: \u0026ldquo;acted more like a group member than a moderator\u0026rdquo;; \u0026ldquo;make order in the group chat\u0026rdquo;; \u0026ldquo;didn\u0026rsquo;t enable group dynamics\u0026hellip; long answers\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 136px;\"\u003e\n \u003cp\u003e3 (14%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRespect for user control\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 194px;\"\u003e\n \u003cp\u003eKeeps talking after being asked to stop\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 166px;\"\u003e\n \u003cp\u003everbal: \u0026ldquo;reacts to request of not responding\u0026hellip; then starts responding again\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 136px;\"\u003e\n \u003cp\u003e1 (5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSensitivity/irony detection\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 194px;\"\u003e\n \u003cp\u003eSensitivity/irony detection\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 166px;\"\u003e\n \u003cp\u003e\u0026ldquo;oversensitive\u0026hellip; cannot differentiate irony\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 136px;\"\u003e\n \u003cp\u003e1 (5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 194px;\"\u003e\n \u003cp\u003eOccasional correctness concerns\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 166px;\"\u003e\n \u003cp\u003e\u0026ldquo;more accurate\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 136px;\"\u003e\n \u003cp\u003e1 (5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 128px;\"\u003e\n \u003cp\u003e\u003cstrong\u003efeature requests\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 194px;\"\u003e\n \u003cp\u003eSuggesting new features\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 166px;\"\u003e\n \u003cp\u003e\u0026ldquo;more funny aspects\u0026hellip; using GIFs\u0026rdquo;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 136px;\"\u003e\n \u003cp\u003e1 (5%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis exploratory study provides insights into how a default GPT-4 chatbot (zero-shot) performs when serving as an ER support agent in an OGL setting. Taken together, the findings offer insight into 1. the theoretical alignment of chatbot responses with ER modes in OGL and 2. how delivery characteristics shaped users’ perceptions of usability and usefulness, with implications for future chatbot design.\u003c/p\u003e \u003cp\u003eRegarding theoretical alignment, the chatbot’s responses are most frequently aligned with SSRL, followed by CoRL, with individual ER occurring least often. The predominance of SSRL-like interactions suggests that a default GPT-4 model, operating under a zero-shot prompting approach and without explicit theory-guided constraints, is nevertheless capable of generating socially shared, group-oriented regulatory supports in collaborative learning contexts. One plausible explanation is that contemporary LLM assistants are pretrained on large-scale internet text and fine-tuned via RLHF (Reinforcement Learning from Human Feedback) to behave helpfully in dialogue (Bai et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Ouyang et al., \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Recent work increasingly positions LLM-based agents as facilitators of collaborative learning and group discussion (Yang et al., \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2025\u003c/span\u003e); accordingly, such systems may be more likely to produce group-oriented scaffolding than individualized ER coaching in multi-party OGL contexts (Lu et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). While socially shared support may be beneficial for collective regulation, it may also constrain the model’s capacity to provide more tailored individual ER strategies in OGL contexts without additional prompting or structural differentiation.\u003c/p\u003e \u003cp\u003eBeyond theoretical alignment, several quantitative and qualitative indicators suggest that users’ perceptions of the chatbot were shaped more strongly by its delivery characteristics than by the content of its interventions alone. Chatbot outputs occurred relatively frequently (approximately once every 2.4 minutes per group) and were often lengthy (M = 118.4 words). These characteristics closely align with the most frequently reported usability concerns, particularly perceptions of intrusiveness and verbosity. Although participants frequently described the chatbot’s advice as coherent or supportive, these delivery features appear to have diminished its perceived usefulness for collaboration, helping explain why collaboration-related usefulness ratings were comparatively modest.\u003c/p\u003e \u003cp\u003eTaken together, these findings suggest that future iterations of chatbot-based ER support in OGL should prioritize tighter control over unsolicited interventions. Design strategies such as refined sentiment thresholds, rate limiting, and persistent user control mechanisms (particularly those that reliably respect explicit “stop” requests) may help reduce perceived disruption while preserving supportive intent. Similarly, defaulting to concise, single-action messages (for example, brief validation paired with one concrete suggestion) may improve acceptability without substantially reducing supportive value.\u003c/p\u003e \u003cp\u003eChannel configuration also appears to play a meaningful role in user experience. The exclusive use of a shared group channel likely contributed to perceptions of disruption, as public interventions can interfere with the natural flow of group interaction. Providing a private channel for individual ER support may reduce conversational crowding while enabling more personalized strategies. Individual ER often benefits from private disclosure, personalization, and iterative follow-up, all of which may be constrained in a public group chat where participants reasonably provide less personal context (Bazarova et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Doré et al., \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2016\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFinally, recurrent concerns regarding role ambiguity underscore the importance of clearer role framing for ER chatbots in OGL. Participants’ feedback suggests the need to specify whether the chatbot functions as a moderator, a peer-like supporter, or a pedagogical agent, and to clarify when and how interventions will occur. More transparent role communication during onboarding and throughout use may help align user expectations with system behavior and reduce frustration related to unexpected or poorly timed interventions.\u003c/p\u003e \u003cp\u003eOverall, the findings highlight that while a default GPT-4 chatbot can align with theoretically meaningful ER modes in OGL, its effectiveness and acceptability are strongly contingent on interaction design choices. Addressing delivery timing, verbosity, private channel placement, and role clarity appears critical for translating theoretically aligned ER support into practically helpful and non-disruptive learning support.\u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eLimitations\u003c/h2\u003e \u003cp\u003eThis exploratory study is subject to several methodological and contextual limitations that influence the generalizability and interpretation of the findings. First, this research was conducted with a small convenience sample (N = 18) in a short experimental OGL session (approx. 45 minutes). Consequently, the generalizability of the results, particularly concerning the long-term benefit and varied user experiences with the chatbot, is limited. Future research should employ larger, more diverse samples over more extended testing periods to validate and generalize these preliminary findings.\u003c/p\u003e \u003cp\u003eA second limitation is that using the LLM (ChatGPT o1) for the deductive coding process introduces challenges to replicability. Since that specific LLM version is no longer publicly available, exact methodological replication is limited. This highlights the necessity for transparency in reporting LLM methodology and emphasizes the need for critical reflection on findings generated using non-static AI tools. Future studies should focus on robust, version-agnostic prompting strategies to enhance methodological durability and replicability, although this will likely be a challenge due to the rapid continuous enhancement of AI models.\u003c/p\u003e \u003cp\u003eThird, this study examined the theoretical alignment of chatbot responses with different emotion regulation (ER) modes in OGL. However, the analysis focused on chatbot outputs in isolation and did not examine them in relation to surrounding user conversations. Consequently, we did not assess the contextual appropriateness or correctness of the chatbot interventions in combination with learners’ expressed sentiments in the Discord group chats. Future research should adopt context-sensitive analyses that link chatbot responses to preceding and subsequent learner interactions, including sentiment dynamics, to evaluate the situational relevance, timing, and effectiveness of ER support in authentic collaborative settings.\u003c/p\u003e \u003cp\u003eA further limitation concerns the interpretation of participants’ evaluations of the chatbot’s content and support. Although users expressed a range of positive and critical perspectives, they were not selected or trained to assess ER strategies from a theoretical or clinical standpoint. As a result, these evaluations likely reflect subjective user experiences rather than objective judgments of ER quality or theoretical appropriateness.\u003c/p\u003e \u003cp\u003eFinally, the study identified specific boundary conditions related to the hybrid OGL setting. Because groups were co-located while using an online chat platform, face-to-face cues sometimes substituted for the chatbot's chat-based mediation. Additionally, participants perceived group-level interventions from GroupBuddy as 'crowding' the conversation. This suggests that the perceived marginal value of a chat-embedded agent diminishes when co-presence is high and public posts are frequent or lengthy. Future experiments should test chatbots in actual OGL settings where digital mediation is the primary mode of interaction, thereby reducing the influence of co-presence effects. Also, implementing a private channel to offer confidential, individual support alongside group-level interventions may mitigate participants' perceived issue of 'crowding' public conversation.\u003c/p\u003e \u003c/div\u003e "},{"header":"Conclusion and future directions","content":"\u003cp\u003eThis exploratory study examined how a zero-shot GPT-4 chatbot functioned as an emotion ER support agent within an OGL setting. The findings suggest that large language model–based agents may provide a feasible foundation for ER support in OGL settings. However, their effectiveness appears contingent on careful alignment between the theoretical intent and the system's behavior. In particular, managing unsolicited interventions, promoting concise and context-sensitive messaging, and differentiating between group-level and individual-level support channels appear critical for reducing disruption while preserving supportive value.\u003c/p\u003e\u003cp\u003eFuture development studies should move beyond exploratory testing to systematically examine how theory-guided prompting strategies, adaptive intervention thresholds, and the inclusion of private support channels influence the effectiveness of ER support. Studies employing larger, more diverse samples, more extended interaction periods, and fully online OGL settings are needed to assess the durability and generalizability of these findings. In addition, future work should incorporate expert evaluations of ER quality alongside user feedback to more rigorously assess theoretical adequacy and instructional value. Collectively, such efforts will be essential for advancing the responsible and effective integration of LLM-based Emotion Regulation support for Online Group Learning.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare that no funds, grants, or other support were received during the preparation of this manuscript.\u003c/p\u003e\n\u003ch2\u003eCompeting Interests\u003c/h2\u003e\n\u003cp\u003eThe authors have no relevant financial or non-financial interests to disclose.\u003c/p\u003e\n\u003ch2\u003eAuthor Contributions\u003c/h2\u003e\n\u003cp\u003eAll authors contributed to the conceptualization and design of the study. Material preparation, data collection, and analysis were performed by Erfan Jalili Jalal and Maartje Henderikx. The manuscript was written by Erfan Jalili Jalal and Maartje Henderikx. Maartje Henderikx, Karel Kreijns, and Rolands Klemke commented on previous versions of the manuscript. All authors read and approved the final manuscript.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eEthics Approval\u003c/h2\u003e\n\u003cp\u003eThe study did not involve medical or sensitive personal data. The data were gathered in accordance with ethical standards; participants volunteered, and the research study was approved by [blinded for review].\u003c/p\u003e\n\u003ch2\u003eConsent to Participate\u003c/h2\u003e\n\u003cp\u003eInformed consent was obtained from all individual participants included in the study.\u003c/p\u003e\n\u003ch2\u003eConsent for Publication\u003c/h2\u003e\n\u003cp\u003eNot applicable; no personal or identifiable information is published.\u003c/p\u003e\n\u003ch2\u003eData and Code Availability\u003c/h2\u003e\n\u003cp\u003eDue to institutional regulations, the datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003eThe chatbot source code and prompt templates used in this study are openly available at:\u003c/p\u003e\n\u003cp\u003ehttps://github.com/ejalili/groupBuddy\u003c/p\u003e\n\u003ch1\u003eAuthors and Affiliations\u003c/h1\u003e\n\u003cp\u003e\u003cimg width=\"19\" height=\"19\" src=\"data:image/png;base64,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\" alt=\"Envelope outline\"\u003e\u0026nbsp;Erfan Jalili Jalal\u003c/p\u003e\n\u003cp\[email protected]\u003c/p\u003e\n\u003cp\u003eErfan Jalili Jalal\u003csup\u003e1\u003c/sup\u003e. Maartje Henderix\u003csup\u003e1\u003c/sup\u003e. Karel Kreijns\u003csup\u003e2\u003c/sup\u003e. Rolands Klemke\u003csup\u003e2,3\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e1. Research in Distance Education (RIDE) Lab. Open Universiteit, 6401 DL Heerlen, the Netherlands\u003c/p\u003e\n\u003cp\u003e2. \u0026nbsp;Department of Learning \u0026amp; Instruction, Faculty of Psychology, 6401 DL Heerlen, the Netherlands\u003c/p\u003e\n\u003cp\u003e3. Cologne Game Lab, TH K\u0026ouml;ln, Cologne, Germany \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe authors declare that no funds, grants, or other support were received during the preparation of this manuscript.\u003c/p\u003e\n\u003ch2\u003eCompeting Interests\u003c/h2\u003e\n\u003cp\u003eThe authors have no relevant financial or non-financial interests to disclose.\u003c/p\u003e\n\u003ch2\u003eAuthor Contributions\u003c/h2\u003e\n\u003cp\u003eAll authors contributed to the conceptualization and design of the study. Material preparation, data collection, and analysis were performed by Erfan Jalili Jalal and Maartje Henderikx. The manuscript was written by Erfan Jalili Jalal and Maartje Henderikx. Maartje Henderikx, Karel Kreijns, and Rolands Klemke commented on previous versions of the manuscript. All authors read and approved the final manuscript.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eEthics Approval\u003c/h2\u003e\n\u003cp\u003eThe study did not involve medical or sensitive personal data. The data were gathered in accordance with ethical standards; participants volunteered, and the research study was approved by [blinded for review].\u003c/p\u003e\n\u003ch2\u003eConsent to Participate\u003c/h2\u003e\n\u003cp\u003eInformed consent was obtained from all individual participants included in the study.\u003c/p\u003e\n\u003ch2\u003eConsent for Publication\u003c/h2\u003e\n\u003cp\u003eNot applicable; no personal or identifiable information is published.\u003c/p\u003e\n\u003ch2\u003eData and Code Availability\u003c/h2\u003e\n\u003cp\u003eDue to institutional regulations, the datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.\u003c/p\u003e\n\u003cp\u003eThe chatbot source code and prompt templates used in this study are openly available at:\u003c/p\u003e\n\u003cp\u003ehttps://github.com/ejalili/groupBuddy\u003c/p\u003e\n\u003ch1\u003eAuthors and Affiliations\u003c/h1\u003e\n\u003cp\u003e\u003cimg width=\"19\" height=\"19\" src=\"data:image/png;base64,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\" alt=\"Envelope outline\"\u003e\u0026nbsp;Erfan Jalili Jalal\u003c/p\u003e\n\u003cp\[email protected]\u003c/p\u003e\n\u003cp\u003eErfan Jalili Jalal\u003csup\u003e1\u003c/sup\u003e. Maartje Henderix\u003csup\u003e1\u003c/sup\u003e. Karel Kreijns\u003csup\u003e2\u003c/sup\u003e. Rolands Klemke\u003csup\u003e2,3\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003e1. Research in Distance Education (RIDE) Lab. Open Universiteit, 6401 DL Heerlen, the Netherlands\u003c/p\u003e\n\u003cp\u003e2. \u0026nbsp;Department of Learning \u0026amp; Instruction, Faculty of Psychology, 6401 DL Heerlen, the Netherlands\u003c/p\u003e\n\u003cp\u003e3. Cologne Game Lab, TH K\u0026ouml;ln, Cologne, Germany \u0026nbsp;\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAaron, B. (2009). Determining what individual SUS scores mean: adding an adjective rating scale. \u003cem\u003eJournal of usability studies\u003c/em\u003e, \u003cem\u003e4\u003c/em\u003e, 3.\u003c/li\u003e\n\u003cli\u003eAlmomani, G. (2024). \u003cem\u003eExploring digital technology use in socially and technologically complex contexts: the case of Discord.\u003c/em\u003e [Doctoral Thesis, University of Edinburgh]. https://doi.org/10.7488/era/4881\u003c/li\u003e\n\u003cli\u003eBai, Y., Jones, A., Ndousse, K., Askell, A., Chen, A., DasSarma, N., et al. (2022). \u003cem\u003eTraining a Helpful and Harmless Assistant with Reinforcement Learning from Human Feedback. arXiv\u003c/em\u003e. https://doi.org/10.48550/arXiv.2204.05862\u003c/li\u003e\n\u003cli\u003eBakhtiar, A., Webster, E. A., \u0026amp; Hadwin, A. F. (2018). Regulation and socio-emotional interactions in a positive and a negative group climate. \u003cem\u003eMetacognition and Learning\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e, 57\u0026ndash;90. https://doi.org/10.1007/s11409-017-9178-x\u003c/li\u003e\n\u003cli\u003eBazarova, N. N., Choi, Y. H., Schwanda Sosik, V., Cosley, D., \u0026amp; Whitlock, J. (2015). Social Sharing of Emotions on Facebook: Channel Differences, Satisfaction, and Replies. In \u003cem\u003eProceedings of the 18th ACM Conference on Computer Supported Cooperative Work \u0026amp; Social Computing\u003c/em\u003e (pp. 154\u0026ndash;164). New York, NY, USA: Association for Computing Machinery. https://doi.org/10.1145/2675133.2675297\u003c/li\u003e\n\u003cli\u003eBraun, V., \u0026amp; Clarke, V. (2006). Using thematic analysis in psychology. \u003cem\u003eQualitative Research in Psychology\u003c/em\u003e, \u003cem\u003e3\u003c/em\u003e(2), 77\u0026ndash;101. https://doi.org/10.1191/1478088706qp063oa\u003c/li\u003e\n\u003cli\u003eBrooke, J. (1996). SUS-A quick and dirty usability scale. \u003cem\u003eUsability evaluation in industry\u003c/em\u003e, \u003cem\u003e189\u003c/em\u003e(194), 4\u0026ndash;7.\u003c/li\u003e\n\u003cli\u003eChen, D., Zhang, Y., Luo, H., Zhu, Z., Ma, J., \u0026amp; Lin, Y. (2024). Effects of group awareness support in CSCL on students\u0026rsquo; learning performance: A three-level meta-analysis. \u003cem\u003eInternational Journal of Computer-Supported Collaborative Learning\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e(1), 97\u0026ndash;129. https://doi.org/10.1007/s11412-024-09418-3\u003c/li\u003e\n\u003cli\u003eChew, R., Bollenbacher, J., Wenger, M., Speer, J., \u0026amp; Kim, A. (2023). \u003cem\u003eLLM-Assisted Content Analysis: Using Large Language Models to Support Deductive Coding. arXiv.\u003c/em\u003e https://doi.org/10.48550/ARXIV.2306.14924\u003c/li\u003e\n\u003cli\u003eDavis, F. D. (1989). Perceived Usefulness, Perceived Ease of Use, and User Acceptance of Information Technology. \u003cem\u003eMIS Quarterly\u003c/em\u003e, \u003cem\u003e13\u003c/em\u003e(3), 319\u0026ndash;340. https://doi.org/10.2307/249008\u003c/li\u003e\n\u003cli\u003eDiscord. (2024). \u003cem\u003eDiscord - Group Chat That\u0026rsquo;s All Fun \u0026amp; Games\u003c/em\u003e. https://discord.com/\u003c/li\u003e\n\u003cli\u003eDor\u0026eacute;, B. P., Silvers, J. A., \u0026amp; Ochsner, K. N. (2016). Toward a Personalized Science of Emotion Regulation. \u003cem\u003eSocial and personality psychology compass\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e(4), 171\u0026ndash;187. https://doi.org/10.1111/spc3.12240\u003c/li\u003e\n\u003cli\u003eGross, J. J. (2015). Emotion Regulation: Current Status and Future Prospects. \u003cem\u003ePsychological Inquiry\u003c/em\u003e, \u003cem\u003e26\u003c/em\u003e(1), 1\u0026ndash;26. https://doi.org/10.1080/1047840x.2014.940781\u003c/li\u003e\n\u003cli\u003eHadwin, A., J\u0026auml;rvel\u0026auml;, S., \u0026amp; Miller, M. (2017). Self-Regulation, Co-Regulation, and Shared Regulation in Collaborative Learning Environments. In D. H. Schunk \u0026amp; J. A. Greene (Eds.), \u003cem\u003eHandbook of Self-Regulation of Learning and Performance\u003c/em\u003e (2nd ed., pp. 83\u0026ndash;106). Routledge. https://doi.org/10.4324/9781315697048-6\u003c/li\u003e\n\u003cli\u003eHarley, J. M., Pekrun, R., Taxer, J. L., \u0026amp; Gross, J. J. (2019). Emotion Regulation in Achievement Situations: An Integrated Model. \u003cem\u003eEducational Psychologist\u003c/em\u003e, \u003cem\u003e54\u003c/em\u003e(2), 106\u0026ndash;126. https://doi.org/10.1080/00461520.2019.1587297\u003c/li\u003e\n\u003cli\u003eHassane, S. (2025). \u003cem\u003eExploring socio-emotional regulation in online group learning\u003c/em\u003e. [Doctoral Thesis, Open Universiteit]. Open University. https://doi.org/10.71583/20251212sh\u003c/li\u003e\n\u003cli\u003eHassane, S., Henderikx, M., J\u0026auml;rvenoja, H., Beckers, J., \u0026amp; Kreijns, K. (2025). The biotope of online group learning: group genesis and persistence seen through the lens of socio-emotional challenges and regulation. \u003cem\u003eEuropean Journal of Psychology of Education\u003c/em\u003e, \u003cem\u003e40\u003c/em\u003e(2), 1\u0026ndash;31. https://doi.org/10.1007/s10212-025-00967-1\u003c/li\u003e\n\u003cli\u003eHenderikx, M. A., \u0026amp; Kreijns, C. (2022). Development and evaluation of the \u0026ldquo;Feedback and Reflection in Online Collaborative Learning\u0026rdquo; (FROCOLE) app. \u003cem\u003eIn INTED2022 Proceedings: 16th International Technology, Education and Development Conference\u003c/em\u003e (pp. 1582\u0026ndash;1590). IATED Academy. https://doi.org/10.21125/inted.2022.0474\u003c/li\u003e\n\u003cli\u003eJ\u0026auml;rvenoja, H., Malmberg, J., T\u0026ouml;rm\u0026auml;nen, T., M\u0026auml;nty, K., Haataja, E., Ahola, S., \u0026amp; J\u0026auml;rvel\u0026auml;, S. (2020). A Collaborative Learning Design for Promoting and Analyzing Adaptive Motivation and Emotion Regulation in the Science Classroom. \u003cem\u003eFrontiers in Education\u003c/em\u003e, \u003cem\u003e5\u003c/em\u003e. https://doi.org/10.3389/feduc.2020.00111\u003c/li\u003e\n\u003cli\u003eJ\u0026auml;rvenoja, H., N\u0026auml;ykki, P., \u0026amp; T\u0026ouml;rm\u0026auml;nen, T. (2019). Emotional regulation in collaborative learning: when do higher education students activate group level regulation in the face of challenges? \u003cem\u003eStudies in Higher Education\u003c/em\u003e, \u003cem\u003e44\u003c/em\u003e(10), 1747\u0026ndash;1757. https://doi.org/10.1080/03075079.2019.1665318\u003c/li\u003e\n\u003cli\u003eJ\u0026auml;rvenoja, H., T\u0026ouml;rm\u0026auml;nen, T., J\u0026auml;rvel\u0026auml;, S., \u0026amp; Shubina, T. (2023). Motivation and emotion regulation in collaborative learning contexts. In \u003cem\u003eMotivation and Emotion in Learning and Teaching across Educational Contexts\u003c/em\u003e (pp. 100\u0026ndash;113). Routledge.\u003c/li\u003e\n\u003cli\u003eJ\u0026auml;rvenoja, H., Volet, S., \u0026amp; J\u0026auml;rvel\u0026auml;, S. (2013). Regulation of emotions in socially challenging learning situations: an instrument to measure the adaptive and social nature of the regulation process. \u003cem\u003eEducational Psychology\u003c/em\u003e, \u003cem\u003e33\u003c/em\u003e(1), 31\u0026ndash;58. https://doi.org/10.1080/01443410.2012.742334\u003c/li\u003e\n\u003cli\u003eJiang, J. A., Wade, K., Fiesler, C., \u0026amp; Brubaker, J. R. (2021). Supporting Serendipity: Opportunities and Challenges for Human-AI Collaboration in Qualitative Analysis. \u003cem\u003eProceedings of the ACM on Human-Computer Interaction\u003c/em\u003e, \u003cem\u003e5\u003c/em\u003e(CSCW1), 1\u0026ndash;23. https://doi.org/10.1145/3449168\u003c/li\u003e\n\u003cli\u003eKirschner, P. a., Kreijns, K., Phielix, C., \u0026amp; Fransen, J. (2015). Awareness of cognitive and social behaviour in a CSCL environment. \u003cem\u003eJournal of Computer Assisted Learning\u003c/em\u003e, \u003cem\u003e31\u003c/em\u003e(1), 59\u0026ndash;77. https://doi.org/10.1111/jcal.12084\u003c/li\u003e\n\u003cli\u003eKitson, A., Slovak, P., \u0026amp; Antle, A. N. (2024). Supporting Cognitive Reappraisal With Digital Technology: A Content Analysis and Scoping Review of Challenges, Interventions, and Future Directions. In \u003cem\u003eProceedings of the CHI Conference on Human Factors in Computing Systems\u003c/em\u003e (pp. 1\u0026ndash;17). https://doi.org/10.1145/3613904.3642488\u003c/li\u003e\n\u003cli\u003eKreijns, K., Weidlich, J., \u0026amp; Kirschner, P. A. (2023). Pitfalls of Social Interaction in Online Group Learning. \u003cem\u003eIn Z. Yan (Ed.), The Cambridge Handbook of Cyber Behavior\u003c/em\u003e (1st ed., pp. 588\u0026ndash;630). Cambridge University Press. https://doi.org/10.1017/9781107165250.025\u003c/li\u003e\n\u003cli\u003eKreijns, K., Yau, J., Weidlich, J., \u0026amp; Weinberger, A. (2024). Towards a comprehensive framework of social presence for online, hybrid, and blended learning. \u003cem\u003eFrontiers in Education\u003c/em\u003e, \u003cem\u003e8\u003c/em\u003e. https://doi.org/10.3389/feduc.2023.1286594\u003c/li\u003e\n\u003cli\u003eLi, X., Hu, W., \u0026amp; Li, Y. (2025). Effect of adaptable and non-adaptable collaboration scripts through conversational agents on student\u0026rsquo;s engagement in online collaborative learning. \u003cem\u003eEducational technology research and development\u003c/em\u003e, \u003cem\u003e73\u003c/em\u003e(2), 875\u0026ndash;907. https://doi.org/10.1007/s11423-025-10448-3\u003c/li\u003e\n\u003cli\u003eLi, Y. (2023). A Practical Survey on Zero-shot Prompt Design for In-context Learning. In \u003cem\u003eProceedings of the Conference Recent Advances in Natural Language Processing - Large Language Models for Natural Language Processings\u003c/em\u003e (pp. 641\u0026ndash;647). https://doi.org/10.26615/978-954-452-092-2_069\u003c/li\u003e\n\u003cli\u003eLiu, X., Zambrano, A. F., Baker, R. S., Barany, A., Ocumpaugh, J., Zhang, J., et al. (2025). Qualitative Coding with GPT-4: Where it Works Better. \u003cem\u003eJournal of Learning Analytics\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(1), 169\u0026ndash;185. https://doi.org/10.18608/jla.2025.8575\u003c/li\u003e\n\u003cli\u003eLu, Y., Aleta, A., Du, C., Shi, L., \u0026amp; Moreno, Y. (2024). LLMs and generative agent-based models for complex systems research. \u003cem\u003ePhysics of Life Reviews\u003c/em\u003e, \u003cem\u003e51\u003c/em\u003e, 283\u0026ndash;293. https://doi.org/10.1016/j.plrev.2024.10.013\u003c/li\u003e\n\u003cli\u003eMiller, M., \u0026amp; Hadwin, A. (2015). Scripting and awareness tools for regulating collaborative learning: Changing the landscape of support in CSCL. \u003cem\u003eComputers in Human Behavior\u003c/em\u003e, \u003cem\u003e52\u003c/em\u003e, 573\u0026ndash;588. https://doi.org/10.1016/j.chb.2015.01.050\u003c/li\u003e\n\u003cli\u003eMiller, M., \u0026amp; Hadwin, A. F. (2024). Comparing the effectiveness of CSCL scripts for shared task perceptions in socially shared regulation of collaborative learning. \u003cem\u003eInternational Journal of Computer-Supported Collaborative Learning\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e(4), 455\u0026ndash;478. https://doi.org/10.1007/s11412-024-09434-3\u003c/li\u003e\n\u003cli\u003eNgo, D., Nguyen, A., Dang, B., \u0026amp; Ngo, H. (2024). Facial Expression Recognition for Examining Emotional Regulation in Synchronous Online Collaborative Learning. \u003cem\u003eInternational Journal of Artificial Intelligence in Education\u003c/em\u003e, \u003cem\u003e34\u003c/em\u003e(3), 650\u0026ndash;669. https://doi.org/10.1007/s40593-023-00378-7\u003c/li\u003e\n\u003cli\u003eNguyen, A., Li, Y., J\u0026auml;rvel\u0026auml;, S., Liu, Y., \u0026amp; Zhao, G. (2023). Emotional Synchrony Through Regulatory Triggers in Collaborative Learning: A Facial Expression Recognition Study.In\u003cem\u003e Proceedings of the 17th International Conference of the Learning Sciences (ICLS 2023)\u003c/em\u003e \u003cem\u003e(pp. 1038\u0026ndash;1041)\u003c/em\u003e. International Society of the Learning Sciences. https://doi.org/10.22318/icls2023.547768\u003c/li\u003e\n\u003cli\u003eO\u0026rsquo;Connor, C., \u0026amp; Joffe, H. (2020). Intercoder Reliability in Qualitative Research: Debates and Practical Guidelines. \u003cem\u003eInternational Journal of Qualitative Methods\u003c/em\u003e, \u003cem\u003e19\u003c/em\u003e. https://doi.org/10.1177/1609406919899220\u003c/li\u003e\n\u003cli\u003eOpenAI, Achiam, J., Adler, S., Agarwal, S., Ahmad, L., Akkaya, I., et al. (2024). \u003cem\u003eGPT-4 Technical Report. arXiv.\u003c/em\u003e https://doi.org/10.48550/arXiv.2303.08774\u003c/li\u003e\n\u003cli\u003eOu, J., Lu, J., Liu, C., Tang, Y., Zhang, F., Zhang, D., \u0026amp; Gai, K. (2024). DialogBench: Evaluating LLMs as Human-like Dialogue Systems. In \u003cem\u003eProceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)\u003c/em\u003e (pp. 6137\u0026ndash;6170). https://doi.org/10.18653/v1/2024.naacl-long.341\u003c/li\u003e\n\u003cli\u003eOuyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C. L., Mishkin, P., et al. (2022). \u003cem\u003eTraining language models to follow instructions with human feedback. arXiv.\u003c/em\u003e https://doi.org/10.48550/arXiv.2203.02155\u003c/li\u003e\n\u003cli\u003eParkington, K., Teferra, B. G., Rouleau-Tang, M., Perivolaris, A., Rueda, A., Dubrowski, A., et al. (2025). \u003cem\u003eHuman vs. LLM-Based Thematic Analysis for Digital Mental Health Research: Proof-of-Concept Comparative Study. arXi\u003c/em\u003ev. https://doi.org/10.48550/arXiv.2507.08002\u003c/li\u003e\n\u003cli\u003ePekrun, R., Muis, K. R., Frenzel, A. C., \u0026amp; Goetz, T. (2017). \u003cem\u003eEmotions at School\u003c/em\u003e (1st ed.). Routledge. https://doi.org/10.4324/9781315187822\u003c/li\u003e\n\u003cli\u003eRojas, M., Nussbaum, M., Guerrero, O., Chiuminatto, P., Greiff, S., Del Rio, R., \u0026amp; Alvares, D. (2022). Integrating a collaboration script and group awareness to support group regulation and emotions towards collaborative problem solving. \u003cem\u003eInternational Journal of Computer-Supported Collaborative Learning\u003c/em\u003e, \u003cem\u003e17\u003c/em\u003e(1), 135\u0026ndash;168. https://doi.org/10.1007/s11412-022-09362-0\u003c/li\u003e\n\u003cli\u003eSentiment. (2024). \u003cem\u003eSentiment \u003c/em\u003e(Version 5.0.2) [Computer Software]. npm. https://www.npmjs.com/package/sentiment\u003c/li\u003e\n\u003cli\u003eShingjergji, K., Iren, D., Urlings, C., \u0026amp; Klemke, R. (2026). Affective computing in online higher education: A systematic literature review. \u003cem\u003eComputers and Education: Artificial Intelligence\u003c/em\u003e, \u003cem\u003e10\u003c/em\u003e, 100499. https://doi.org/10.1016/j.caeai.2025.100499\u003c/li\u003e\n\u003cli\u003eSlovak, P., Antle, A., Theofanopoulou, N., Daud\u0026eacute;n Roquet, C., Gross, J., \u0026amp; Isbister, K. (2023). Designing for Emotion Regulation Interventions: An Agenda for HCI Theory and Research. \u003cem\u003eACM Transactions on Computer-Human Interaction\u003c/em\u003e, \u003cem\u003e30\u003c/em\u003e(1), 1\u0026ndash;51. https://doi.org/10.1145/3569898\u003c/li\u003e\n\u003cli\u003eWeidlich, J., Yau, J., \u0026amp; Kreijns, K. (2024). Social presence and psychological distance: A construal level account for online distance learning. \u003cem\u003eEducation and Information Technologies\u003c/em\u003e, \u003cem\u003e29\u003c/em\u003e(1), 401\u0026ndash;423. https://doi.org/10.1007/s10639-023-12289-0\u003c/li\u003e\n\u003cli\u003eXiao, Z., Yuan, X., Liao, Q. V., Abdelghani, R., \u0026amp; Oudeyer, P.-Y. (2023). Supporting Qualitative Analysis with Large Language Models: Combining Codebook with GPT-3 for Deductive Coding. In \u003cem\u003e28th International Conference on Intelligent User Interfaces\u003c/em\u003e (pp. 75\u0026ndash;78). Association for Computing Machinery. https://doi.org/10.1145/3581754.3584136\u003c/li\u003e\n\u003cli\u003eYang, Q., Yang, Y., An, S., Hao, T., \u0026amp; Xu, G. (2025). LLM-based collaborative agents with pedagogy-guided interaction modeling for timely instructive feedback generation in task-oriented group discussions. In \u003cem\u003eProceedings of the Thirty-Fourth International Joint Conference on Artificial Intelligence\u003c/em\u003e (pp. 9972\u0026ndash;9980). Montreal, Canada. https://doi.org/10.24963/ijcai.2025/1108\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":true,"hideJournal":true,"highlight":"","institution":"","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":"Online Group Learning (OGL), Computer-supported Collaborative Learning (CSCL), Social Emotions, Emotion Regulation, Human-computer Interaction (HCI), Large Language Models (LLMs)","lastPublishedDoi":"10.21203/rs.3.rs-8443551/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8443551/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eOnline group learning (OGL) may be affected by socio-emotional challenges associated with social and interactional barriers in online settings, which can elicit negative emotions among learners. Effective emotion regulation (ER) appears to be a crucial factor in supporting productive collaboration. Recent advances in artificial intelligence (AI), particularly large language models (LLMs), offer potential avenues for ER support in OGL; however, empirical guidance on the design and implementation of such tools remains limited. To begin addressing this gap, the present study examined the use of a default GPT-4 chatbot implemented within an OGL setting as an ER support agent. Chatbot outputs and user experience survey responses were analyzed using a mixed-methods approach combining deductive content analysis, qualitative thematic analysis, and descriptive quantitative measures. Results indicated that most chatbot outputs contained theory-aligned ER components, with socially shared and co-regulated learning strategies occurring more frequently than individual-level ER strategies. User experience findings indicated moderate usability and mixed perceptions of the chatbot\u0026rsquo;s effectiveness, with qualitative feedback emphasizing the influence of delivery characteristics such as timing and verbosity of the chatbot\u0026rsquo;s responses. Taken together, the findings suggest that while default LLM-based agents may offer a feasible foundation for ER support in OGL, careful interaction design and theory-aligned refinement are critical for enhancing acceptability and practical value.\u003c/p\u003e","manuscriptTitle":"Designing Emotion Regulation Support in Online Group Learning: Insights from an LLM-Based Support Agent","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-07 15:04:41","doi":"10.21203/rs.3.rs-8443551/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":"490a8d81-1182-4beb-bd1a-816bb4b34933","owner":[],"postedDate":"January 7th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-03-19T09:42:23+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-07 15:04:41","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8443551","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8443551","identity":"rs-8443551","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}