Students’ psychological biases towards teacher and AI-generated feedback: An experiment study

Students’ psychological biases towards teacher and AI-generated feedback: An experiment study | 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 Students’ psychological biases towards teacher and AI-generated feedback: An experiment study Haiyan YU This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8942475/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 22 Apr, 2026 Read the published version in BMC Psychology → Version 1 posted 10 You are reading this latest preprint version Abstract Background While AI-generated feedback has been promoted as a supplement to teacher feedback, recent empirical studies have consistently revealed that students prefer teacher feedback to AI feedback. It remains unclear whether such preferences reflect the inherent quality of the feedback or stem from psychological biases toward the feedback source. Method This study investigates the existence of psychological biases towards teacher and AI feedback. An experiment was conducted with 52 advanced learners of Chinese as a second language (CSL). The subjects were randomly divided into two groups: AI-label group and teacher-label group. All participants received identical AI-generated feedback on an essay, but were led to believe it came from either an AI system (AI-label group) or an experienced teacher (teacher-label group). Data included participants’ perceptions of the feedback measured across seven psychological dimensions (coverage, accuracy, elaboration, utility, cost, interest, and intention), as well as their subsequent textual revisions coded by type and frequency. Results Participants in the teacher-label group reported consistently more positive perceptions of the feedback across all dimensions, although these differences did not reach statistical significance. However, the AI-label group made significantly more textual revisions, particularly replacements, than the teacher-label group ( p = .017). This reveals a dissociation between cognitive appraisal of feedback and behavioral engagement with it. Conclusion These findings provide empirical evidence of a psychological bias against AI-generated feedback, wherein the same feedback is perceived less favorably when attributed to AI. However, this negative bias does not translate into reduced behavioral engagement; instead, learners interact more actively with AI-attributed feedback, potentially due to reduced social-evaluative concerns and enhanced autonomy. The study contributes to the psychology of human-AI interaction in educational contexts and highlights the need to consider both cognitive and socio-affective mechanisms in understanding feedback engagement. subjective bias feedback source AI feedback teacher feedback international students Chinese as a second language Figures Figure 1 Figure 2 1 Introduction The rise of generative artificial intelligence (GenAI) has opened up promising new possibilities for enhancing second language (L2) writing pedagogy (Luo & Yusuf, 2025 ; Rong et al., 2025 ; Zhan et al., 2025 ; Zhan & Yan, 2025 ; Zhang et al., 2025 ; Zou et al., 2025 ). AI-generated feedback is increasingly recognized for its potential to offer fast, consistent, and scalable support to L2 writers (Steiss et al., 2024 ; Woo et al., 2024 ; Yang & Li, 2024 ). In response to these affordances, scholars have started examining whether AI feedback can effectively supplement or even replace teacher feedback (Yan et al., 2026 ). However, the output from AI has been viewed less than satisfactory by its end users (Man, 2026 ). Recent studies consistently show that students tend to value and engage more deeply with teacher feedback than with AI feedback, despite recognizing AI’s advantages such as accessibility, immediacy, and comprehensiveness (Henderson et al., 2025 ; Zou et al., 2025 ). From a psychological perspective, this preference pattern raises a fundamental question: Do students’ differential responses to AI and teacher feedback reflect genuine differences in feedback quality, or do they stem from pre-existing psychological biases toward the feedback source? A key limitation in prior comparative studies lies in the lack of control over feedback content. It remains uncertain whether differences in students’ preferences and subsequent revisions are driven by the actual content of the feedback or by subjective perceptions of its source. Understanding learners’ perceptions of AI-generated feedback and their subsequent revisions is critical because feedback from automated systems, whether traditional tools like Grammarly or newer conversational agents such as ChatGPT, has often been perceived as focusing primarily on surface-level issues and lacking contextual specificity (Shi & Aryadoust, 2024 ; Thi & Nikolov, 2021 ; Zhang et al., 2025 ). Such negative perceptions may be amplified when AI systems are used for languages other than English, where available linguistic resources are comparatively limited (Shadiev & Feng, 2024 ). For example, while ChatGPT has outperformed teachers in detecting errors and providing detailed, balanced feedback on English writing, it may underperform in other low-resource languages, where its comments were less balanced and occasionally inaccurate (Fokides & Peristeraki, 2024 ). These results underscore the need to investigate the performance of AI feedback and its perceptions beyond English. The suboptimal quality of AI-generated feedback may lead to persistent negative perceptions, to the extent that simply labeling feedback as AI-generated can bias students’ evaluations of feedback message (Brummernhenrich et al., 2025 ; Ruwe & Kuklick, 2025 ). Despite recent interest in the topic of subjective bias in feedback perceptions, the few relevant studies have focused on perceived trustworthiness, with other dimensions of feedback perceptions underexplored. The present paper addresses this research gap by exploring whether a psychological bias exists in students’ perceptions of AI feedback within the context of learning Chinese as a second language, and whether such bias influences their subsequent behavioral engagement with the feedback. Two groups of learners receive identical feedback produced by an AI system but are informed of different feedback sources (AI vs. teacher). Specifically, this paper answers two questions: 1)Does the label of the feedback source influence students’ perceptions of the feedback? 2)Does the label of the feedback source influence students’ textual revisions in response to the feedback? 2 Literature review 2.1 The quality and functions of automated feedback Feedback has been identified as a key support for student learning (Hattie & Timperley, 2007 ). However, in many low-resource educational contexts, quality teacher feedback is often unavailable or delayed. Automated writing evaluation (AWE) systems emerged to address this problem by providing scalable, consistent, and timely feedback to learners (Almusharraf & Alotaibi, 2022 ; Shi & Aryadoust, 2024 ). Despite these affordances, AWE feedback has often been viewed as inferior to teacher feedback. Research on established AWE tools such as Pigai , Grammarly , and Criterion reveals persistent limitations in their ability to identify and explain errors accurately (Ding & Zou, 2024 ). For instance, Bai and Hu ( 2016 ) found that Pigai produced correct suggestions in only 58.61% of grammar cases and 21.83% of collocation cases, confirming AWE’s tendency to emphasize surface-level issues over higher-order concerns such as idea, coherence, and style (Thi & Nikolov, 2021 ; Zou et al., 2025 ). Recently, GenAI marks a turning point in AWE feedback. Tools such as ChatGPT are believed to move beyond rule-based error detection to emulate human-like evaluation of ideas, argumentation, and discourse structure (Mizumoto et al., 2024 ). Comparative studies, however, show mixed results about the quality of GenAI feedback (Chen et al., 2024 ). Steiss et al. ( 2024 ) compared feedback from human evaluators and ChatGPT 3.5 on a corpus of English essays and found that human feedback scored higher on most quality dimensions, including accuracy, prioritization, and supportive tone. They noted that ChatGPT, though more consistent in referencing assessment criteria, occasionally produced inaccurate or contradictory comments. Similarly, Fokides and Peristeraki ( 2024 ) observed that while ChatGPT surpassed teachers in error detection for English essays, it underperformed in Greek writing due to inaccurate flagging and less attention to mechanics. These findings suggest that GenAI’s capacity to generate high-quality feedback varies across languages, especially those with limited training resources. Overall, this body of research reveals both the potential and the limitations of AI-mediated feedback. While AI can provide high-volume, rapid responses, it still faces challenges with linguistic nuance, contextual understanding, and cross-linguistic equity. 2.2 Potential and uptake of automated feedback AWE has the potential to promote L2 writing. AWE systems can promote noticing, provide metalinguistic explanations, and encourage self-directed learning (Barrot, 2021 ; Kim, 2024 ). Empirical evidence shows that learners who receive AWE feedback often outperform those who do not, particularly in grammatical accuracy (e.g., Barrot, 2021 ). Yet, findings comparing AWE and teacher feedback remain inconclusive. Some studies report stronger effects of AWE on writing quality (e.g., Hassanzadeh & Fotoohnejad, 2021 ), while others find no significant difference (e.g., Escalante et al., 2023 ) or highlight complementary effects when AWE is combined with teacher input (Han & Sari, 2022 ; Zhang et al., 2025 ). These results suggest that AWE is most useful as a collaborative tool rather than a replacement for teachers (Henderson et al., 2025 ). However, feedback is not useful until it is properly utilized (Winstone et al., 2021 ). While AI can provide timely and scalable feedback, there is no guarantee that students can make good use of the feedback provided, especially given the large amount of feedback. Feedback can be easily interpreted as criticism even if the feedback focuses on the areas for improvement. Excessive critical feedback can induce anxiety and impede feedback uptake (Koltovskaia, 2022 ; Sun & Fan, 2022 ). Studies consistently show low uptake rates of automated feedback: learners adopt only about half of the system’s suggestions (Bai & Hu, 2016 ; Tian & Zhou, 2020 ). For example, Koltovskaia ( 2020 ) found that their participants corrected about 57% of the errors flagged by Grammarly. Low uptake has been attributed not only to technical limitations, such as inaccuracy and lack of metalinguistic explanation (Barrot, 2021 ; Guo et al., 2022 ), but also to learners’ perceptions of the credibility and usefulness of AWE feedback (Link et al., 2022 ). For instance, Zou et al. ( 2025 ) found that students incorporated more revisions based on teacher feedback than on AI-generated feedback. These findings highlight that feedback uptake is not only a cognitive process but also a socially and affectively mediated one, dependent on learners’ trust and perceived value of the feedback source (Ranalli, 2021 ). 2.3 Feedback source and feedback perceptions Research has shown that the source of feedback can impact students’ perceptions of the feedback they receive (Van der Kleij & Lipnevich, 2020 ). This influence may arise from the characteristics of the feedback itself or from the credibility of its source (van de Ridder et al., 2015 ). In the case of automated feedback, both students and teachers have reported mixed perceptions regarding its characteristics (Fu et al., 2024 ). On one hand, some students perceive automated feedback as more timely and detailed than teacher feedback (Oneill & Russell, 2019 ). Automated feedback helps students address surface-level issues such as typos and grammatical errors while providing guidance for improvement and textual revisions (Allen & Mizumoto, 2024 ). On the other hand, students often perceive AWE feedback as inferior to teacher feedback because it may lack accuracy and explicitness and has limited capacity to offer in-depth feedback on content, coherence, style, and overall precision (Fu et al., 2024 ; Sun & Fan, 2022 ; Wang & Han, 2022 ). Students tend to rate teacher feedback as significantly more accurate and helpful than computer-generated feedback, even though the feedback was identical in content (Lipnevich & Smith, 2014 ). The mixed perceptions associated with traditional automated feedback systems also apply to AI. Recent research has examined how students perceive feedback generated by AI in comparison to teacher feedback. For example, Escalante et al. ( 2023 ) examined the use of ChatGPT-4 among 48 tertiary EFL learners over six weeks. Their study suggested that students generally preferred teacher feedback over AI feedback. Similarly, Zou et al. ( 2025 ) investigated how Chinese EFL students responded to feedback on their English writing when it was provided by teachers versus generated by ChatGPT. Students perceived teacher feedback as more helpful than AI feedback; however, only the difference in perceived usefulness of feedback for language use was statistically significant. Overall, teacher feedback was preferred over AI feedback. Unlike these two experiments, Henderson et al. ( 2025 ) administered a survey of 6,960 students from four Australian universities. They found that half of the students sought feedback from GenAI, and that students perceived teacher feedback to be more useful and more trustworthy. The findings from these recent studies suggest that students may question the authority of AI tools because they are not human experts who are the real judges of student performance (Zhan & Yan, 2025 ). User perceptions of feedback can play a key role in feedback utilization (Winstone et al., 2021 ). Students evaluate the quality of feedback and weigh the costs and benefits of seeking and implementing it (Gu, 2025 ). Their subjective perceptions of AWE feedback influence their subsequent feedback uptake. Although artificial intelligence has become more powerful than ever, and AI-generated feedback appears comparable to human-generated feedback (Fokides & Peristeraki, 2024 ), language learners may still harbor lingering biases regarding AWE feedback and its functions (Fu et al., 2024 ; Sun & Fan, 2022 ). Students might assume that AI-generated feedback is less accurate than teacher feedback and develop a sense of distrust toward it (Fu et al., 2024 ; Sun & Fan, 2022 ). Without knowing the source of feedback, the experiment by Ruwe and Kuklick ( 2025 ) showed that participants rated feedback labeled as AI-generated as less trustworthy than feedback labeled as teacher-generated. Such stereotypes not only influence students’ judgements regarding the authority and relevance of AI feedback but also potentially influence students’ response to AI feedback. While the potential biased view of AI feedback has far-reaching implications, limited research has been conducted to confirm its existence. To address this gap, the present paper reports on an experiment that aims to determine whether CSL learners hold biased perceptions of AI-generated feedback and whether the designated source of the feedback affects their revisions. 3 Method 3.1 Participants and the context A total of 52 CSL learners were invited to participate in the experiment. All participants were international students studying Chinese at a public university in China. They came from multiple countries: Indonesia (24), Myanmar (4), Thailand (4), Venezuela (3), the Dominican Republic (2), Mongolia (2), Peru (2), Japan (2), Vietnam (2), Belarus (1), Russia (1), France (1), Kyrgyzstan (1), Canada (1), Malaysia (1), and South Africa (1). Seventy-five percent of the participants (39) are descendants of overseas Chinese. All participants demonstrated advanced Chinese proficiency, as evidenced by a score of ≥ 220 on the HSK, an international standardized test for Chinese language proficiency. They had normal or corrected-to-normal visual acuity. They were informed of the experimental procedures and gave written consent to participate. But they were naïve to the purpose of the experiment, i.e., to test for subjective bias in feedback perceptions. Each participant received a small gift as a token of appreciation upon completing the experiment. The sample size was determined based on guidelines proposed by Cohen ( 2013 ), which recommends a minimum statistical power of 0.8 and a corresponding effect size for robust results. Using G*Power software, we estimated that a total sample size of 50 participants was required for a one-factor, two-level between-subjects design. To account for potential attrition, we recruited 52 participants, comprising 18 males and 34 females, with an average age of 21.90 years ( SD = 2.22). Participants were randomly allocated to two groups. The first group (hereafter Group 1) was informed that the feedback they processed was generated by an AI system. The second group (hereafter Group 2) was told that the feedback they processed was composed by an experienced teacher. 3.2 Materials The materials include an essay and the feedback comments that an AI system generated for the essay. The essay was selected through a systematic procedure. First, three essays, representative of high, medium, and low proficiency levels, were randomly selected from the Dynamic Composition Corpus for HSK, developed by Beijing Language and Culture University. The three essays were evaluated by five senior CSL instructors, each with at least seven years of teaching experience, to assess their suitability for the experiment. The essay titled The Impact of Smoking on Personal Health and Public Welfare was chosen because it was typical of the proficiency level of the participating students. The selected essay was presented to two advanced Chinese learners and an experienced CSL teacher, who were invited to revise it. Their textual changes were used to ensure that the essay requires revisions. The essay was then submitted to Kimi AI Chat, a free Chinese AI platform, which was instructed to assess the essay and provide feedback based on the HSK rubric. The prompt was “Please provide feedback for the essay based on the HSK rubric attached below”. The feedback points provided by Kimi AI Chat can be found in Appendix B. 3.3 Feedback perception questionnaire Students’ feedback perceptions were measured using a composite questionnaire consisting of seven dimensions: coverage (5 items), accuracy (3 items), elaboration (4 items), interest (4 items), utility (3 items), cost (4 items), and intention (3 items). The items were adapted from validated instruments used in prior feedback studies (Gu, 2025 ; Papi et al., 2020 ; Van der Kleij & Lipnevich, 2020 ) and revised to suit the context of this experiment. Responses were recorded on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree). The questionnaire demonstrated excellent internal consistency, with an overall Cronbach’s alpha of .899. All subscales exceeded the recommended reliability threshold of .80, indicating satisfactory measurement quality: coverage (α = .813), accuracy (α = .825), elaboration (α = .832), interest (α = .823), utility (α = .877), cost (α = .832), and intention (α = .824). 3.4 Data collection and analysis The experiment was administered in a computer lab. Participants were first presented with the selected essay and the accompanying AI-generated feedback. Afterwards, they were asked to answer a questionnaire regarding their perceptions of the feedback. Fifteen minutes were given to read the essay and its accompanying feedback. The questionnaire took approximately five minutes to complete. Afterward, participants were asked to revise the essay based on the feedback they had received. The revision was completed using MS Word with the Track Changes feature enabled (see Figs. 1 and 2 ). All participants completed the revision within forty-five minutes. The students’ revised essays were imported into MaxQDA, a qualitative analysis software used for coding. Textual changes were identified and coded based on meaning rather than on a clause-by-clause basis. A unit of analysis could be as short as a single Chinese character or as long as a full sentence. Once all units of analysis were identified, they were coded according to the type of revision: replacement, deletion, addition, or reordering. One author coded all the units of analysis twice, with a two-month interval between sessions. Intra-coder reliability was assessed to ensure consistency in coding. Another author reviewed each coding, and any disagreements were resolved through iterative discussions. 4 Results The normality of the data was first assessed. No severe violation of the normality assumption was detected. An independent samples t -test was performed to compare the feedback perceptions between Group 1 and Group 2 across several measures. No significant differences were found between the two groups on any of the measures (see Table 1 ). However, the mean values of feedback perceptions revealed some interesting patterns. Specifically, participants tended to perceive feedback as more comprehensive (coverage), accurate (accuracy), elaborate (elaboration), useful (utility), and interesting (interest value) when they were told it was from a teacher. Additionally, feedback was perceived as less costly when attributed to an experienced teacher. While students generally rated feedback more positively when it was labeled as teacher feedback, they were more willing to receive additional feedback when they believed it was generated by AI. Table 1 Comparison of feedback perceptions Measure Group 1 M ( SD ) Group 2 M ( SD ) p 95% CI Coverage 4.12 (0.58) 4.18 (0.66) .716 [-0.415, 0.287] Accuracy 3.84 (0.75) 4.17 (0.69) .107 [-0.741, 0.075] Elaboration 3.93 (0.73) 3.96 (0.78) .889 [-0.399, 0.459] Utility 4.29 (0.74) 4.39 (0.63) .633 [-0.484, 0.297] Cost 2.51 (1.00) 2.41 (1.04) .747 [-0.485, 0.672] Interest 3.54 (0.96) 3.82 (0.82) .460 [ -0.785, 0.225] Intention 4.44 (0.70) 4.28 (0.82) .271 [-0.272, 0.592] Note . Group 1: AI group; Group 2: Teacher group. An independent samples t -test was used to compare the revisions made by Group 1 (AI group) and Group 2 (Teacher group) across several measures(see Table 2 ). For replacement, a significant difference was found between the groups, p = .017, 95% CI [0.656, 6.113]. Group 1 reported a higher mean number of replacements ( M = 11.00, SD = 6.35) compared to Group 2 ( M = 7.62, SD = 2.53). However, no significant differences were observed for deletion, p = .223, 95% CI [-0.459, 1.920], addition, p = .296, 95% CI [-1.595, 5.133], or re-ordering, p = .887, 95% CI [-0.504, 0.581]. Still, students enacted more replacements, deletion, and addition when the feedback was labeled as AI-generated feedback than when it was labeled as teacher feedback. Overall, students implemented a larger number of revisions when they believed they were working with AI-generated feedback. Table 2 Comparison of revisions Measure Group 1 M ( SD ) Group 2 M ( SD ) p 95% CI Replacement 11.00 (6.35) 7.62 (2.53) .017 [0.656, 6.113] Deletion 2.42 (2.28) 1.69 (1.98) .223 [-0.459, 1.920] Addition 8.88 (7.10) 7.12 (4.74) .296 [-1.595, 5.133] Re-ordering 0.69 (0.65) 0.79 (1.13) .887 [-0.504, 0.581] Note . Group 1: AI group; Group 2: Teacher group. 5 Discussion This experimental study investigated the presence of bias in the perception and use of AI-generated feedback among CSL learners. The findings reveal a psychologically significant dissociation between cognitive appraisal and behavioral engagement: while students perceived identical feedback more favorably when attributed to a teacher, they made significantly more revisions when the same feedback was labeled as AI-generated. This study demonstrated that students had a less positive perception of identical feedback across all dimensions when they were informed that the feedback was provided by an AI system. This result aligns with a study by Lipnevich and Smith in 2014 on the perception of traditional automated feedback system. They found that students tended to perceive teacher feedback as more accurate and helpful than computer-generated feedback, even though the feedback was identical in content. It has thus become clear that such subjective bias persists to this day, even as AI has become more powerful than traditional AWE systems (Henderson et al., 2025 ). This study also aligns with Ruwe and Kuklick ( 2025 ), which showed that an AI label for feedback reduced the perceived trustworthiness of the feedback source. The present study, however, has offered more insights by revealing that an AI label can reduce the positive perceptions across both affective and cognitive dimensions, ranging from interest to perceived accuracy, elaboration, and utility. The results of this study have also advanced the strand of research on the preferences for teacher feedback (Henderson et al., 2025 ). Specifically, prior research comparing perceptions of AI feedback with those of teacher feedback has attributed differences in perceptions such as trust to the distinct characteristics of feedback from these sources (e.g., Henderson et al., 2025 ; Ranalli, 2021 ; Tossell et al., 2024 ; Zhan & Yan, 2025 ). This study, however, shows that with a different label, students could have different perceptions of the same feedback. Overall, the differing perceptions caused by a labelled source points to the existence of subjective bias toward feedback sources. Interestingly, the more positive perceptions elicited by the teacher feedback label did not lead to a higher level of engagement with the feedback. This finding challenges the intuitive assumption that positive cognitive appraisal directly translates into greater behavioral engagement—a phenomenon with important psychological implications. It contradicts assumptions that positive feedback perceptions regarding trustworthiness and feedback quality are associated with greater feedback engagement (Henderson et al., 2025 ; Ruwe & Kuklick, 2025 ). Instead, this study demonstrated that students responded more actively to feedback when informed that it was provided by an AI system, despite expressing less favorable perceptions of the feedback. Specifically, the AI group enacted a larger number of revisions than the teacher group, although the difference was not statistically significant. Contrary to earlier findings (e.g., Liu et al., 2024 ; Zou et al., 2025 ), which indicated stronger learner engagement with teacher feedback, this study revealed that AI-generated responses can elicit greater learner interaction in certain CALL contexts. The discrepancy between more positive feedback perceptions and a lower level of feedback engagement suggests alternative explanations other than those proposed in prior research. For instance, Liu et al. ( 2024 ) found that secondary school students who received AWE feedback on their Chinese writing made fewer revisions compared to peers who received feedback from teachers. The relatively fewer revisions when interacting with AWE feedback could be attributed to the overload caused by feedback quantity and its focus on surface issues. Zou et al. ( 2025 ) found that students showed a higher uptake of teacher feedback on their English writing compared to feedback generated by ChatGPT. They attributed this higher rate of uptake to greater trust in the accuracy of teacher feedback. These explanations do not apply to the present study, in which both groups acted on the same feedback, yet students rated the feedback more positively when they were told it was from a teacher. In other words, trust in teacher feedback is not the sole reason for the higher level of engagement with teacher feedback. Alternative explanations are needed. This paper accounts for the discrepancies between feedback perceptions and feedback engagement from a socio-affective psychological perspective (Henderson et al., 2025 ; Ruwe & Kuklick, 2025 ). Teacher feedback involves teacher-student relationships (Gu, 2025 ; Guo & Xu, 2025 ). From a social psychology standpoint, the authority inherent in teacher feedback activates different motivational and emotional responses compared to AI feedback. On the one hand, the authority of teacher feedback could decrease the likelihood of uninvited revisions, that is, revisions that are neither requested nor suggested in teacher feedback. In this experiment, most of the participants were descendants of overseas Chinese nurtured in Chinese culture. They tend to hold high regard for the authority of their teachers. This cultural orientation amplifies the psychological salience of authority and social evaluation. For them, the primary goal of revision behavior is to meet the teacher’s requirements and avoid negative evaluations. Due to their goal-oriented and risk-averse nature, they made only minimal corrections to the issues explicitly identified by the teacher. The assumption is that the absence of negative feedback indicates no serious issues in the writing. Unsolicited revisions may imply that teachers are neglecting their responsibilities and have overlooked important points. The threat to teachers’ authority is particularly alarming when unsolicited revisions can considerably enhance the quality of the writing. Thus, students tend to suppress “unnecessary” revisions to avoid causing dissonance between teacher feedback and the number of revisions. This behavior reflects impression management concerns and social-evaluative threat—well-documented psychological phenomena in feedback contexts. On the other hand, responding to teacher feedback may expose students’ shortcomings and thus incur self-presentation costs (Gu, 2025 ). The embarrassment is particularly pronounced when unsolicited revisions fail to improve the writing or, even worse, degrade its quality. Such instances can reveal their deficiencies. Consequently, students were more cautious when revising text in response to teacher feedback than to AI feedback. For segments without teacher feedback, participants assumed the text had no serious issues and therefore made fewer revisions, possibly to avoid unnecessary changes and protect their self-esteem. In contrast, the socio-affective psychological factors are less salient in interactions with AI feedback. There is minimal risk of revealing personal deficiencies to AI systems (Henderson et al., 2025 ). From a self-determination theory perspective, AI feedback may better satisfy learners’ basic psychological needs for autonomy and competence. AI feedback is generally perceived as inferior to teacher feedback. When interacting with AI feedback, students are confident that they can outperform it. Consequently, they make more attempts to revise their texts. Due to its impersonal nature and lack of direct association with external rewards, punishments, or interpersonal evaluations, AI feedback can satisfy learners’ autonomy needs, thereby fostering autonomous motivation (Zong & Yang, 2025 ). Under this motivation, students revise their essays in response to AI-generated feedback with an intrinsic desire to prove their own abilities. Students demonstrate their language skills as superior to the AI tool through text optimization, achieving self-affirmation of human agency. It is no wonder that students addressing computer-generated feedback often display a higher level of self-confidence and a lower level of anxiety (Liu et al., 2024 ; Zhang et al., 2025 ). This psychological state—characterized by reduced evaluative threat and enhanced intrinsic motivation—explains why revision behaviors extend beyond the scope of the feedback, exhibiting both high frequency and a broad scope when learners believe they are interacting with AI. 6 Implications This study offers important implications for research on AI-generated feedback. Future comparisons between AI and teacher feedback should account for potential subjective biases that learners may hold toward AI systems. It is crucial to differentiate between variations in learners’ engagement and subsequent learning outcomes that arise from biased perceptions of the feedback source from those that stem from the intrinsic qualities of the feedback itself. The bias is not constrained to trustworthiness or source credibility; it also relates to other aspects of feedback perception, such as perceived utility and cost (Gu, 2025 ; Gu & Man, 2025 ). Such differentiation will enable a more precise understanding of how students perceive and enact AI-mediated feedback and help inform the equitable incorporation of AI into language learning contexts. The study also carries implications for the design and delivery of feedback in language teaching. First, the findings provide empirical evidence of potential biases in students’ perceptions of AI-generated feedback, enriching our understanding of the conditions under which different feedback sources should be selected. Given that teacher feedback is generally preferred over AI feedback, AI-generated feedback should be positioned as supplementary support rather than a replacement for teacher feedback input (Henderson et al., 2025 ; Zhang et al., 2025 ). In contexts where teachers face heavy feedback workloads but students still value teacher feedback, AI systems can be strategically integrated into the feedback process. For instance, students may initially rely on AI to refine and revise their drafts before submitting them for teacher assessment. Teachers may also invite students to compose rebuttal letters in response to AI-generated comments to encourage deeper engagement in L2 writing. When clearly framed as a supportive and preparatory tool, AI feedback may enhance students’ confidence and willingness to revise. Second, the findings highlight that students’ biased perceptions of AI-generated feedback can shape how they interpret and respond to such feedback. Effective feedback design must therefore take into account learners’ subjective attitudes toward AI tools. Negative or skeptical perceptions may discourage students from engaging with AI feedback, especially when the feedback volume is high and cognitively overwhelming. To counter such biases, teachers can help learners develop an informed understanding of the affordances and limitations of AI systems, fostering strategic and critical use of AI feedback. Additionally, as excessive or overly critical comments may activate learners’ self-defensive responses and heighten attention to negative aspects of performance (Brummernhenrich et al., 2025 ; Huang et al., 2018 ), the quantity and tone of AI feedback should be carefully mediated to create an emotionally supportive learning climate. 7 Conclusion This experimental study provides empirical evidence of subjective bias in students’ responses to AI-generated feedback among a group of international students learning Chinese as a second language. The findings reveal that students tend to perceive identical feedback more favorably when it is attributed to an experienced teacher. Conversely, when informed that the feedback originates from an intelligent AI system, they are more likely to act upon it, making a greater number of revisions. These results underscore the social-affective dimension of feedback reception and offer new insights into why students may demonstrate heightened engagement when responding to AI feedback. Overall, the study advances understanding of how source cues shape students’ feedback perceptions and subsequent revisions. The findings of the study point to the need for exploration into the mechanisms behind feedback perceptions, uptake, and engagement in human-AI feedback contexts. Concrete implications for designing and integrating AI feedback into language learning are also discussed. This study has several limitations. First, the study was carried out in a controlled laboratory environment, which may constrain the applicability of its findings to authentic, real-world educational settings. In authentic classroom settings, students’ engagement with feedback can be shaped by multiple factors, including the importance of the writing task, the nature of the teacher-student relationship, and students’ perceptions of the potential costs and benefits of seeking feedback. Whether subjective bias exists in authentic settings and to what extent such bias influences textual revisions merit replication studies. Second, this experimental study did not control for the number of feedback points provided. AI systems, such as ChatGPT, can generate a large volume of feedback, potentially leading to cognitive overload for students. An excessive amount of feedback may discourage student’s feedback engagement. The interaction between biased perceptions and feedback quantity on feedback engagement also merits scholarly attention in future inquiries. Declarations Human Ethics and Consent to Participate declarations This study was approved by the Research Ethics Committee of College of Chinese Language and Culture，Jinan University (Approval No: 2024-11, Date: 03/09/2024). All procedures performed in this study involving human participants were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki declaration and its later amendments. Informed consent was obtained electronically from all participants prior to data collection. Consent for publication Not applicable. Availability of data and materials The datasets analyzed during the current study are available from the corresponding author on reasonable request. Funding Not applicable Acknowledgments Not applicable Competing interests The author declares no competing interests Authorship contributions H.Y is responsible for all work for this manuscript. References Allen TJ, Mizumoto A. ChatGPT over my friends: Japanese English-as-a-foreign-language learners’ preferences for editing and proofreading strategies. RELC J. 2024. https://doi.org/10.1177/00336882241262533 . Almusharraf N, Alotaibi H. An error-analysis study from an EFL writing context: Human and automated essay scoring approaches. Technol Knowl Learn. 2022;28(3):1015–31. https://doi.org/10.1007/s10758-022-09592-z . Bai L, Hu G. In the face of fallible AWE feedback: How do students respond? Educational Psychol. 2016;37(1):67–81. https://doi.org/10.1080/01443410.2016.1223275 . Barrot JS. Using automated written corrective feedback in the writing classrooms: effects on L2 writing accuracy. Comput Assist Lang Learn. 2021;36(4):584–607. https://doi.org/10.1080/09588221.2021.1936071 . Brummernhenrich B, Paulus CL, Jucks R. Applying social cognition to feedback chatbots: Enhancing trustworthiness through politeness. Br J Edu Technol. 2025. https://doi.org/10.1111/bjet.13569 . Chen Z, Zhu X, Lu Q, Wei W. L2 students’ barriers in engaging with form and content-focused AI-generated feedback in revising their compositions. Comput Assist Lang Learn. 2024;1–21. https://doi.org/10.1080/09588221.2024.2422478 . Cohen J. Statistical power analysis for the behavioral sciences. 2nd ed. ed.). Routledge; 2013. Ding L, Zou D. Automated writing evaluation systems: A systematic review of Grammarly, Pigai, and Criterion with a perspective on future directions in the age of generative artificial intelligence. Educ Inform Technol. 2024;29(11):14151–203. https://doi.org/10.1007/s10639-023-12402-3 . Escalante J, Pack A, Barrett A. AI-generated feedback on writing: Insights into efficacy and ENL student preference. Int J Educational Technol High Educ. 2023;20(1). https://doi.org/10.1186/s41239-023-00425-2 . Fokides E, Peristeraki E. Comparing ChatGPT's correction and feedback comments with that of educators in the context of primary students' short essays written in English and Greek. Educ Inform Technol. 2024. https://doi.org/10.1007/s10639-024-12912-8 . Fu Q-K, Zou D, Xie H, Cheng G. A review of AWE feedback: Types, learning outcomes, and implications. Comput Assist Lang Learn. 2024;37(1–2):179–221. https://doi.org/10.1080/09588221.2022.2033787 . Gu H. Exploring factors contributing to feedback-seeking strategies in L2 writing: An extended cost-value framework. Stud Educational Evaluation. 2025;85. https://doi.org/10.1016/j.stueduc.2025.101453 . Gu H, Man D. Understanding the motivations behind feedback-seeking behavior in L2 writing: A situated expectancy-value perspective. Int J Appl Linguistics. 2025. https://doi.org/10.1111/ijal.12842 . Guo Q, Feng R, Hua Y. How effectively can EFL students use automated written corrective feedback (AWCF) in research writing? Comput Assist Lang Learn. 2022;35(9):2312–31. https://doi.org/10.1080/09588221.2021.1879161 . Guo Q, Xu Y. The effect of achievement goals on students’ perceptions of feedback: The mediating roles of teacher-student relationships and academic self-concept. Asia-Pacific Educ Researcher. 2025. https://doi.org/10.1007/s40299-025-00996-w . Han T, Sari E. An investigation on the use of automated feedback in Turkish EFL students’ writing classes. Comput Assist Lang Learn. 2022;37(4):961–85. https://doi.org/10.1080/09588221.2022.2067179 . Hassanzadeh M, Fotoohnejad S. Implementing an automated feedback program for a foreign language writing course: A learner-centric study. J Comput Assist Learn. 2021;37(5):1494–507. https://doi.org/10.1111/jcal.12587 . Hattie J, Timperley H. The power of feedback. Rev Educ Res. 2007;77(1):81–112. https://doi.org/10.3102/003465430298487 . Henderson M, Bearman M, Chung J, Fawns T, Buckingham Shum S, Matthews KE, de Mello Heredia J. Comparing Generative AI and teacher feedback: Student perceptions of usefulness and trustworthiness. Assess Evaluation High Educ. 2025;1–16. https://doi.org/10.1080/02602938.2025.2502582 . Huang P-S, Liu C-H, Chen H-C, Sommers S. Attentional bias of students toward negative feedback in bad outcome situations: The mechanism of self-defense. Soc Psychol Educ. 2018;21(3):565–83. https://doi.org/10.1007/s11218-018-9429-y . Kim R. Effects of learner uptake following automatic corrective recast from artificial intelligence chatbots on the learning of English caused-motion construction. Lang Learn Technol. 2024;28(2):109–33. https://doi.org/https://hdl.handle.net/10125/73574 . Koltovskaia S. (2020). Student engagement with automated written corrective feedback (AWCF) provided by Grammarly: A multiple case study. Assessing Writing , 44 . https://doi.org/10.1016/j.asw.2020.100450 Koltovskaia S. Postsecondary L2 writing teachers’ use and perceptions of Grammarly as a complement to their feedback. ReCALL. 2022;35(3):290–304. https://doi.org/10.1017/s0958344022000179 . Link S, Mehrzad M, Rahimi M. Impact of automated writing evaluation on teacher feedback, student revision, and writing improvement. Comput Assist Lang Learn. 2022;35(4):605–34. https://doi.org/10.1080/09588221.2020.1743323 . Lipnevich AA, Smith JK. Response to assessment feedback: the effects of grades, praise, and source of information. ETS Res Rep Ser. 2014;2008(1). https://doi.org/10.1002/j.2333-8504.2008.tb02116.x . Liu S, Boudouaia A, Chen X, Li Y. Comparative study about the impacts of Chinese automated writing evaluation (AWE) feedback and teacher feedback on middle school students’ writing practice. Asia-Pacific Educ Researcher. 2024. https://doi.org/10.1007/s40299-024-00903-9 . Luo L, Yusuf A. Bridging AI and pedagogy: How AI-adaptive feedback shapes Chinese EFL students’ writing engagement, metacognitive writing strategies, and writing performance. Assess Evaluation High Educ. 2025;1–24. https://doi.org/10.1080/02602938.2025.2548919 . Man D. The predictive effects of user perceptions on the willingness to continue using machine translation. Translation Interpreting Stud. 2026. https://doi.org/10.1075/tis.25014.man . Mizumoto A, Shintani N, Sasaki M, Teng MF. Testing the viability of ChatGPT as a companion in L2 writing accuracy assessment. Res Methods Appl Linguistics. 2024;3(2). https://doi.org/10.1016/j.rmal.2024.100116 . Oneill R, Russell A. Stop! Grammar time: University students’ perceptions of the automated feedback program Grammarly. Australasian J Educational Technol. 2019;35(1). https://doi.org/10.14742/ajet.3795 . Papi M, Bondarenko AV, Wawire B, Jiang C, Zhou S. Feedback–seeking behavior in second language writing: Motivational mechanisms. Read Writ. 2020;33:485–505. https://doi.org/10.1007/s11145-019-09971-6 . Ranalli J. L2 student engagement with automated feedback on writing: Potential for learning and issues of trust. J Second Lang Writ. 2021;52. https://doi.org/10.1016/j.jslw.2021.100816 . Rong M, Yao Y, Li Q, Chen X. Exploring student engagement with artificial intelligence-guided chatbot feedback in EFL writing: Interactions and revisions. Comput Assist Lang Learn. 2025;1–30. https://doi.org/10.1080/09588221.2025.2539979 . Ruwe T, Kuklick L. Quality counts? Examining the role of feedback provider and feedback quality on students' feedback perceptions. Br J Edu Technol. 2025. https://doi.org/10.1111/bjet.70011 . Shadiev R, Feng Y. Using automated corrective feedback tools in language learning: a review study. Interact Learn Environ. 2024;32(6):2538–66. https://doi.org/10.1080/10494820.2022.2153145 . Shi H, Aryadoust V. A systematic review of AI-based automated written feedback research. ReCALL. 2024;36(2):187–209. https://doi.org/10.1017/s0958344023000265 . Steiss J, Tate T, Graham S, Cruz J, Hebert M, Wang J, Moon Y, Tseng W, Warschauer M, Olson CB. Comparing the quality of human and ChatGPT feedback of students’ writing. Learn Instruction. 2024;91. https://doi.org/10.1016/j.learninstruc.2024.101894 . Sun B, Fan T. The effects of an AWE-aided assessment approach on business English writing performance and writing anxiety: A contextual consideration. Stud Educational Evaluation. 2022;72. https://doi.org/10.1016/j.stueduc.2021.101123 . Thi NK, Nikolov M. How teacher and Grammarly feedback complement one another in myanmar EFL students’ writing. Asia-Pacific Educ Researcher. 2021;31(6):767–79. https://doi.org/10.1007/s40299-021-00625-2 . Tian L, Zhou Y. Learner engagement with automated feedback, peer feedback and teacher feedback in an online EFL writing context. System. 2020. https://doi.org/10.1016/j.system.2020.102247 . Tossell CC, Tenhundfeld NL, Momen A, Cooley K, de Visser EJ. Student perceptions of ChatGPT use in a college essay assignment: Implications for learning, grading, and trust in artificial intelligence. IEEE Trans Learn Technol. 2024;17:1069–81. https://doi.org/10.1109/tlt.2024.3355015 . van de Ridder JMM, Berk FCJ, Stokking KM, Ten Cate OTJ. Feedback providers' credibility impacts students' satisfaction with feedback and delayed performance. Med Teach. 2015;37(8):767–74. https://doi.org/10.3109/0142159X.2014.970617 . Van der Kleij FM, Lipnevich AA. Student perceptions of assessment feedback: a critical scoping review and call for research. Educational Assess Evaluation Account. 2020;33(2):345–73. https://doi.org/10.1007/s11092-020-09331-x . Wang Z, Han F. The effects of teacher feedback and automated feedback on cognitive and psychological aspects of foreign language writing: A mixed-methods research. Front Psychol. 2022;13:909802. https://doi.org/10.3389/fpsyg.2022.909802 . Winstone NE, Hepper EG, Nash RA. Individual differences in self-reported use of assessment feedback: The mediating role of feedback beliefs. Educational Psychol. 2021;41(7):844–62. https://doi.org/10.1080/01443410.2019.1693510 . Woo DJ, Wang D, Guo K, Susanto H. Teaching EFL students to write with ChatGPT: Students' motivation to learn, cognitive load, and satisfaction with the learning process. Educ Inform Technol. 2024;29(18):24963–90. https://doi.org/10.1007/s10639-024-12819-4 . Yan D, Tian F, Li H, Gao Y, Li J. (2026). A GenAI-supported dialogic model to enhance the functions of peer feedback in EFL writing: A mixed-method interventional study. System , 136 . https://doi.org/10.1016/j.system.2025.103895 Yang L, Li R. ChatGPT for L2 learning: Current status and implications. System 124. 2024. https://doi.org/10.1016/j.system.2024.103351 . Zhan Y, Boud D, Dawson P, Yan Z. Generative artificial intelligence as an enabler of student feedback engagement: a framework. High Educ Res Dev. 2025;44(5):1289–304. https://doi.org/10.1080/07294360.2025.2476513 . Zhan Y, Yan Z. Students’ engagement with ChatGPT feedback: Implications for student feedback literacy in the context of generative artificial intelligence. Assess Evaluation High Educ. 2025;1–14. https://doi.org/10.1080/02602938.2025.2471821 . Zhang Z, Aubrey S, Huang X, Chiu TKF. The role of generative AI and hybrid feedback in improving L2 writing skills: A comparative study. Innov Lang Learn Teach. 2025;1–19. https://doi.org/10.1080/17501229.2025.2503890 . Zong Y, Yang L. How AI-enhanced social-emotional learning framework transforms EFL students' engagement and emotional well-being. Eur J Educ. 2025;60(1). https://doi.org/10.1111/ejed.12925 . Zou S, Guo K, Wang J, Liu Y. Investigating students’ uptake of teacher- and ChatGPT-generated feedback in EFL writing: A comparison study. Comput Assist Lang Learn. 2025;1–30. https://doi.org/10.1080/09588221.2024.2447279 . Additional Declarations No competing interests reported. Supplementary Files FeedbackPerceptionQuestionnaire.docx Appendices.docx Cite Share Download PDF Status: Published Journal Publication published 22 Apr, 2026 Read the published version in BMC Psychology → Version 1 posted Editorial decision: Revision requested 11 Mar, 2026 Reviews received at journal 09 Mar, 2026 Reviews received at journal 09 Mar, 2026 Reviewers agreed at journal 08 Mar, 2026 Reviewers agreed at journal 06 Mar, 2026 Reviewers invited by journal 05 Mar, 2026 Editor assigned by journal 04 Mar, 2026 Editor invited by journal 04 Mar, 2026 Submission checks completed at journal 04 Mar, 2026 First submitted to journal 27 Feb, 2026 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-8942475","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":602944712,"identity":"d98c4a89-4a83-4e1f-8fb7-ff65789fc13b","order_by":0,"name":"Haiyan YU","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA6ElEQVRIiWNgGAWjYFACHgaGBCDFxszY/uGDgY0c8Vr42ZnbGGcUpBkTpwUEJPvZ25h5PhxOJKjB4EbuwQ8P2+zyDA4ztj22MWBOYGA/fHQDfi15yRKJbcnFQC3txjkGbHkMPGlpN/BryTEAamFO3HCYsUE6x4CnmEGCx4yQFuMfiW31EC0WQO0NRGgxA9pyOHFmM2ObNIOBAWEtkmfemFkknDue2M/M2GzYY5BgzEbIL3zHc4xv/iirTmzjP/7wwY8//+X42Q8fw6tF4QCQYGRDEmHDoRIO5BtA5B9CykbBKBgFo2BEAwC6Y01HmAsKhQAAAABJRU5ErkJggg==","orcid":"","institution":"College of Chinese Language and Culture，Jinan University","correspondingAuthor":true,"prefix":"","firstName":"Haiyan","middleName":"","lastName":"YU","suffix":""}],"badges":[],"createdAt":"2026-02-23 03:23:25","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8942475/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8942475/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1186/s40359-026-04568-5","type":"published","date":"2026-04-22T15:56:56+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":104510886,"identity":"24c601af-1d5f-43b6-951b-5105da79df03","added_by":"auto","created_at":"2026-03-12 15:58:18","extension":"jpg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":328400,"visible":true,"origin":"","legend":"\u003cp\u003eRevisions made by a student receiving feedback with the AI label\u003c/p\u003e","description":"","filename":"Picture1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8942475/v1/ff96c0a0a651e94b01e3e9d5.jpg"},{"id":104510889,"identity":"0479c8d0-75f5-4bd3-b15a-611529673dd6","added_by":"auto","created_at":"2026-03-12 15:58:18","extension":"jpg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":161310,"visible":true,"origin":"","legend":"\u003cp\u003eRevisions made by a student receiving feedback with the label of teacher feedback\u003c/p\u003e","description":"","filename":"Picture2.jpg","url":"https://assets-eu.researchsquare.com/files/rs-8942475/v1/b8b4a5027efc305429e2a3d3.jpg"},{"id":107927665,"identity":"9cb20f00-1d09-467b-a6ce-ae2a02792ae5","added_by":"auto","created_at":"2026-04-27 16:00:48","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":828296,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8942475/v1/f94602fa-ddb1-4501-8f4d-a582229dfee1.pdf"},{"id":104510887,"identity":"ffe8300c-3430-4eeb-a189-ff485d6c1b32","added_by":"auto","created_at":"2026-03-12 15:58:18","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":20739,"visible":true,"origin":"","legend":"","description":"","filename":"FeedbackPerceptionQuestionnaire.docx","url":"https://assets-eu.researchsquare.com/files/rs-8942475/v1/8bb9a0feccc888467d0d8915.docx"},{"id":104510888,"identity":"4dc4c70e-79fd-4f39-8d81-e268ebcdd506","added_by":"auto","created_at":"2026-03-12 15:58:18","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":20279,"visible":true,"origin":"","legend":"","description":"","filename":"Appendices.docx","url":"https://assets-eu.researchsquare.com/files/rs-8942475/v1/9f5334d3c7dbede6d99b6491.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Students’ psychological biases towards teacher and AI-generated feedback: An experiment study","fulltext":[{"header":"1 Introduction","content":"\u003cp\u003eThe rise of generative artificial intelligence (GenAI) has opened up promising new possibilities for enhancing second language (L2) writing pedagogy (Luo \u0026amp; Yusuf, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Rong et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zhan et al., \u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zhan \u0026amp; Yan, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zou et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). AI-generated feedback is increasingly recognized for its potential to offer fast, consistent, and scalable support to L2 writers (Steiss et al., \u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Woo et al., \u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Yang \u0026amp; Li, \u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). In response to these affordances, scholars have started examining whether AI feedback can effectively supplement or even replace teacher feedback (Yan et al., \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). However, the output from AI has been viewed less than satisfactory by its end users (Man, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2026\u003c/span\u003e). Recent studies consistently show that students tend to value and engage more deeply with teacher feedback than with AI feedback, despite recognizing AI\u0026rsquo;s advantages such as accessibility, immediacy, and comprehensiveness (Henderson et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zou et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eFrom a psychological perspective, this preference pattern raises a fundamental question: Do students\u0026rsquo; differential responses to AI and teacher feedback reflect genuine differences in feedback quality, or do they stem from pre-existing psychological biases toward the feedback source? A key limitation in prior comparative studies lies in the lack of control over feedback content. It remains uncertain whether differences in students\u0026rsquo; preferences and subsequent revisions are driven by the actual content of the feedback or by subjective perceptions of its source. Understanding learners\u0026rsquo; perceptions of AI-generated feedback and their subsequent revisions is critical because feedback from automated systems, whether traditional tools like Grammarly or newer conversational agents such as ChatGPT, has often been perceived as focusing primarily on surface-level issues and lacking contextual specificity (Shi \u0026amp; Aryadoust, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Thi \u0026amp; Nikolov, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Such negative perceptions may be amplified when AI systems are used for languages other than English, where available linguistic resources are comparatively limited (Shadiev \u0026amp; Feng, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). For example, while ChatGPT has outperformed teachers in detecting errors and providing detailed, balanced feedback on English writing, it may underperform in other low-resource languages, where its comments were less balanced and occasionally inaccurate (Fokides \u0026amp; Peristeraki, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). These results underscore the need to investigate the performance of AI feedback and its perceptions beyond English. The suboptimal quality of AI-generated feedback may lead to persistent negative perceptions, to the extent that simply labeling feedback as AI-generated can bias students\u0026rsquo; evaluations of feedback message (Brummernhenrich et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Ruwe \u0026amp; Kuklick, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eDespite recent interest in the topic of subjective bias in feedback perceptions, the few relevant studies have focused on perceived trustworthiness, with other dimensions of feedback perceptions underexplored. The present paper addresses this research gap by exploring whether a psychological bias exists in students\u0026rsquo; perceptions of AI feedback within the context of learning Chinese as a second language, and whether such bias influences their subsequent behavioral engagement with the feedback. Two groups of learners receive identical feedback produced by an AI system but are informed of different feedback sources (AI vs. teacher). Specifically, this paper answers two questions:\u003c/p\u003e\n\u003ch3\u003e1)Does the label of the feedback source influence students’ perceptions of the feedback?\u003c/h3\u003e\n\u003cp\u003e \u003cdiv class=\"BlockQuote\"\u003e \u003cp\u003e2)Does the label of the feedback source influence students\u0026rsquo; textual revisions in response to the feedback?\u003c/p\u003e \u003c/div\u003e \u003c/p\u003e"},{"header":"2 Literature review","content":"\u003cdiv id=\"Sec4\" class=\"Section2\"\u003e \u003ch2\u003e2.1 The quality and functions of automated feedback\u003c/h2\u003e \u003cp\u003eFeedback has been identified as a key support for student learning (Hattie \u0026amp; Timperley, \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2007\u003c/span\u003e). However, in many low-resource educational contexts, quality teacher feedback is often unavailable or delayed. Automated writing evaluation (AWE) systems emerged to address this problem by providing scalable, consistent, and timely feedback to learners (Almusharraf \u0026amp; Alotaibi, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Shi \u0026amp; Aryadoust, \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Despite these affordances, AWE feedback has often been viewed as inferior to teacher feedback. Research on established AWE tools such as \u003cem\u003ePigai\u003c/em\u003e, \u003cem\u003eGrammarly\u003c/em\u003e, and \u003cem\u003eCriterion\u003c/em\u003e reveals persistent limitations in their ability to identify and explain errors accurately (Ding \u0026amp; Zou, \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). For instance, Bai and Hu (\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2016\u003c/span\u003e) found that Pigai produced correct suggestions in only 58.61% of grammar cases and 21.83% of collocation cases, confirming AWE\u0026rsquo;s tendency to emphasize surface-level issues over higher-order concerns such as idea, coherence, and style (Thi \u0026amp; Nikolov, \u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Zou et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eRecently, GenAI marks a turning point in AWE feedback. Tools such as ChatGPT are believed to move beyond rule-based error detection to emulate human-like evaluation of ideas, argumentation, and discourse structure (Mizumoto et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Comparative studies, however, show mixed results about the quality of GenAI feedback (Chen et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Steiss et al. (\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) compared feedback from human evaluators and ChatGPT 3.5 on a corpus of English essays and found that human feedback scored higher on most quality dimensions, including accuracy, prioritization, and supportive tone. They noted that ChatGPT, though more consistent in referencing assessment criteria, occasionally produced inaccurate or contradictory comments. Similarly, Fokides and Peristeraki (\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) observed that while ChatGPT surpassed teachers in error detection for English essays, it underperformed in Greek writing due to inaccurate flagging and less attention to mechanics. These findings suggest that GenAI\u0026rsquo;s capacity to generate high-quality feedback varies across languages, especially those with limited training resources.\u003c/p\u003e \u003cp\u003eOverall, this body of research reveals both the potential and the limitations of AI-mediated feedback. While AI can provide high-volume, rapid responses, it still faces challenges with linguistic nuance, contextual understanding, and cross-linguistic equity.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec5\" class=\"Section2\"\u003e \u003ch2\u003e2.2 Potential and uptake of automated feedback\u003c/h2\u003e \u003cp\u003eAWE has the potential to promote L2 writing. AWE systems can promote noticing, provide metalinguistic explanations, and encourage self-directed learning (Barrot, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Kim, \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). Empirical evidence shows that learners who receive AWE feedback often outperform those who do not, particularly in grammatical accuracy (e.g., Barrot, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Yet, findings comparing AWE and teacher feedback remain inconclusive. Some studies report stronger effects of AWE on writing quality (e.g., Hassanzadeh \u0026amp; Fotoohnejad, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2021\u003c/span\u003e), while others find no significant difference (e.g., Escalante et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) or highlight complementary effects when AWE is combined with teacher input (Han \u0026amp; Sari, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). These results suggest that AWE is most useful as a \u003cem\u003ecollaborative tool\u003c/em\u003e rather than a replacement for teachers (Henderson et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eHowever, feedback is not useful until it is properly utilized (Winstone et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). While AI can provide timely and scalable feedback, there is no guarantee that students can make good use of the feedback provided, especially given the large amount of feedback. Feedback can be easily interpreted as criticism even if the feedback focuses on the areas for improvement. Excessive critical feedback can induce anxiety and impede feedback uptake (Koltovskaia, \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Sun \u0026amp; Fan, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Studies consistently show low uptake rates of automated feedback: learners adopt only about half of the system\u0026rsquo;s suggestions (Bai \u0026amp; Hu, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Tian \u0026amp; Zhou, \u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). For example, Koltovskaia (\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) found that their participants corrected about 57% of the errors flagged by Grammarly. Low uptake has been attributed not only to technical limitations, such as inaccuracy and lack of metalinguistic explanation (Barrot, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Guo et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2022\u003c/span\u003e), but also to learners\u0026rsquo; \u003cem\u003eperceptions\u003c/em\u003e of the credibility and usefulness of AWE feedback (Link et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). For instance, Zou et al. (\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) found that students incorporated more revisions based on teacher feedback than on AI-generated feedback. These findings highlight that feedback uptake is not only a cognitive process but also a socially and affectively mediated one, dependent on learners\u0026rsquo; trust and perceived value of the feedback source (Ranalli, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec6\" class=\"Section2\"\u003e \u003ch2\u003e2.3 Feedback source and feedback perceptions\u003c/h2\u003e \u003cp\u003eResearch has shown that the source of feedback can impact students\u0026rsquo; perceptions of the feedback they receive (Van der Kleij \u0026amp; Lipnevich, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). This influence may arise from the characteristics of the feedback itself or from the credibility of its source (van de Ridder et al., \u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e2015\u003c/span\u003e). In the case of automated feedback, both students and teachers have reported mixed perceptions regarding its characteristics (Fu et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). On one hand, some students perceive automated feedback as more timely and detailed than teacher feedback (Oneill \u0026amp; Russell, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2019\u003c/span\u003e). Automated feedback helps students address surface-level issues such as typos and grammatical errors while providing guidance for improvement and textual revisions (Allen \u0026amp; Mizumoto, \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2024\u003c/span\u003e). On the other hand, students often perceive AWE feedback as inferior to teacher feedback because it may lack accuracy and explicitness and has limited capacity to offer in-depth feedback on content, coherence, style, and overall precision (Fu et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Sun \u0026amp; Fan, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Wang \u0026amp; Han, \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Students tend to rate teacher feedback as significantly more accurate and helpful than computer-generated feedback, even though the feedback was identical in content (Lipnevich \u0026amp; Smith, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2014\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe mixed perceptions associated with traditional automated feedback systems also apply to AI. Recent research has examined how students perceive feedback generated by AI in comparison to teacher feedback. For example, Escalante et al. (\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) examined the use of ChatGPT-4 among 48 tertiary EFL learners over six weeks. Their study suggested that students generally preferred teacher feedback over AI feedback. Similarly, Zou et al. (\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) investigated how Chinese EFL students responded to feedback on their English writing when it was provided by teachers versus generated by ChatGPT. Students perceived teacher feedback as more helpful than AI feedback; however, only the difference in perceived usefulness of feedback for language use was statistically significant. Overall, teacher feedback was preferred over AI feedback. Unlike these two experiments, Henderson et al. (\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) administered a survey of 6,960 students from four Australian universities. They found that half of the students sought feedback from GenAI, and that students perceived teacher feedback to be more useful and more trustworthy. The findings from these recent studies suggest that students may question the authority of AI tools because they are not human experts who are the real judges of student performance (Zhan \u0026amp; Yan, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eUser perceptions of feedback can play a key role in feedback utilization (Winstone et al., \u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e2021\u003c/span\u003e). Students evaluate the quality of feedback and weigh the costs and benefits of seeking and implementing it (Gu, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Their subjective perceptions of AWE feedback influence their subsequent feedback uptake. Although artificial intelligence has become more powerful than ever, and AI-generated feedback appears comparable to human-generated feedback (Fokides \u0026amp; Peristeraki, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2024\u003c/span\u003e), language learners may still harbor lingering biases regarding AWE feedback and its functions (Fu et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Sun \u0026amp; Fan, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Students might assume that AI-generated feedback is less accurate than teacher feedback and develop a sense of distrust toward it (Fu et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Sun \u0026amp; Fan, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). Without knowing the source of feedback, the experiment by Ruwe and Kuklick (\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) showed that participants rated feedback labeled as AI-generated as less trustworthy than feedback labeled as teacher-generated. Such stereotypes not only influence students\u0026rsquo; judgements regarding the authority and relevance of AI feedback but also potentially influence students\u0026rsquo; response to AI feedback. While the potential biased view of AI feedback has far-reaching implications, limited research has been conducted to confirm its existence. To address this gap, the present paper reports on an experiment that aims to determine whether CSL learners hold biased perceptions of AI-generated feedback and whether the designated source of the feedback affects their revisions.\u003c/p\u003e \u003c/div\u003e"},{"header":"3 Method","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Participants and the context\u003c/h2\u003e \u003cp\u003e A total of 52 CSL learners were invited to participate in the experiment. All participants were international students studying Chinese at a public university in China. They came from multiple countries: Indonesia (24), Myanmar (4), Thailand (4), Venezuela (3), the Dominican Republic (2), Mongolia (2), Peru (2), Japan (2), Vietnam (2), Belarus (1), Russia (1), France (1), Kyrgyzstan (1), Canada (1), Malaysia (1), and South Africa (1). Seventy-five percent of the participants (39) are descendants of overseas Chinese. All participants demonstrated advanced Chinese proficiency, as evidenced by a score of \u0026ge;\u0026thinsp;220 on the HSK, an international standardized test for Chinese language proficiency. They had normal or corrected-to-normal visual acuity. They were informed of the experimental procedures and gave written consent to participate. But they were na\u0026iuml;ve to the purpose of the experiment, i.e., to test for subjective bias in feedback perceptions. Each participant received a small gift as a token of appreciation upon completing the experiment.\u003c/p\u003e \u003cp\u003eThe sample size was determined based on guidelines proposed by Cohen (\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2013\u003c/span\u003e), which recommends a minimum statistical power of 0.8 and a corresponding effect size for robust results. Using G*Power software, we estimated that a total sample size of 50 participants was required for a one-factor, two-level between-subjects design. To account for potential attrition, we recruited 52 participants, comprising 18 males and 34 females, with an average age of 21.90 years (\u003cem\u003eSD\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2.22). Participants were randomly allocated to two groups. The first group (hereafter Group 1) was informed that the feedback they processed was generated by an AI system. The second group (hereafter Group 2) was told that the feedback they processed was composed by an experienced teacher.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.2 Materials\u003c/h2\u003e \u003cp\u003eThe materials include an essay and the feedback comments that an AI system generated for the essay. The essay was selected through a systematic procedure. First, three essays, representative of high, medium, and low proficiency levels, were randomly selected from the Dynamic Composition Corpus for HSK, developed by Beijing Language and Culture University. The three essays were evaluated by five senior CSL instructors, each with at least seven years of teaching experience, to assess their suitability for the experiment. The essay titled \u003cem\u003eThe Impact of Smoking on Personal Health and Public Welfare\u003c/em\u003e was chosen because it was typical of the proficiency level of the participating students. The selected essay was presented to two advanced Chinese learners and an experienced CSL teacher, who were invited to revise it. Their textual changes were used to ensure that the essay requires revisions.\u003c/p\u003e \u003cp\u003eThe essay was then submitted to Kimi AI Chat, a free Chinese AI platform, which was instructed to assess the essay and provide feedback based on the HSK rubric. The prompt was \u0026ldquo;Please provide feedback for the essay based on the HSK rubric attached below\u0026rdquo;. The feedback points provided by Kimi AI Chat can be found in Appendix B.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec10\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Feedback perception questionnaire\u003c/h2\u003e \u003cp\u003eStudents\u0026rsquo; feedback perceptions were measured using a composite questionnaire consisting of seven dimensions: coverage (5 items), accuracy (3 items), elaboration (4 items), interest (4 items), utility (3 items), cost (4 items), and intention (3 items). The items were adapted from validated instruments used in prior feedback studies (Gu, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Papi et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Van der Kleij \u0026amp; Lipnevich, \u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e2020\u003c/span\u003e) and revised to suit the context of this experiment. Responses were recorded on a 5-point Likert scale ranging from 1 (strongly disagree) to 5 (strongly agree).\u003c/p\u003e \u003cp\u003eThe questionnaire demonstrated excellent internal consistency, with an overall Cronbach\u0026rsquo;s alpha of .899. All subscales exceeded the recommended reliability threshold of .80, indicating satisfactory measurement quality: coverage (α\u0026thinsp;=\u0026thinsp;.813), accuracy (α\u0026thinsp;=\u0026thinsp;.825), elaboration (α\u0026thinsp;=\u0026thinsp;.832), interest (α\u0026thinsp;=\u0026thinsp;.823), utility (α\u0026thinsp;=\u0026thinsp;.877), cost (α\u0026thinsp;=\u0026thinsp;.832), and intention (α\u0026thinsp;=\u0026thinsp;.824).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.4 Data collection and analysis\u003c/h2\u003e \u003cp\u003eThe experiment was administered in a computer lab. Participants were first presented with the selected essay and the accompanying AI-generated feedback. Afterwards, they were asked to answer a questionnaire regarding their perceptions of the feedback. Fifteen minutes were given to read the essay and its accompanying feedback. The questionnaire took approximately five minutes to complete. Afterward, participants were asked to revise the essay based on the feedback they had received. The revision was completed using MS Word with the Track Changes feature enabled (see Figs.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e and \u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). All participants completed the revision within forty-five minutes.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe students\u0026rsquo; revised essays were imported into MaxQDA, a qualitative analysis software used for coding. Textual changes were identified and coded based on meaning rather than on a clause-by-clause basis. A unit of analysis could be as short as a single Chinese character or as long as a full sentence. Once all units of analysis were identified, they were coded according to the type of revision: replacement, deletion, addition, or reordering. One author coded all the units of analysis twice, with a two-month interval between sessions. Intra-coder reliability was assessed to ensure consistency in coding. Another author reviewed each coding, and any disagreements were resolved through iterative discussions.\u003c/p\u003e \u003c/div\u003e"},{"header":"4 Results","content":"\u003cp\u003eThe normality of the data was first assessed. No severe violation of the normality assumption was detected. An independent samples \u003cem\u003et\u003c/em\u003e-test was performed to compare the feedback perceptions between Group 1 and Group 2 across several measures. No significant differences were found between the two groups on any of the measures (see Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e). However, the mean values of feedback perceptions revealed some interesting patterns. Specifically, participants tended to perceive feedback as more comprehensive (coverage), accurate (accuracy), elaborate (elaboration), useful (utility), and interesting (interest value) when they were told it was from a teacher. Additionally, feedback was perceived as less costly when attributed to an experienced teacher. While students generally rated feedback more positively when it was labeled as teacher feedback, they were more willing to receive additional feedback when they believed it was generated by AI.\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\u003eComparison of feedback perceptions\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\"\u0026minus;\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMeasure\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGroup 1 \u003cem\u003eM\u003c/em\u003e (\u003cem\u003eSD\u003c/em\u003e)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGroup 2 \u003cem\u003eM\u003c/em\u003e (\u003cem\u003eSD\u003c/em\u003e)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e95% CI\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCoverage\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.12 (0.58)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4.18 (0.66)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.716\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e[-0.415, 0.287]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.84 (0.75)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4.17 (0.69)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.107\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e[-0.741, 0.075]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eElaboration\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.93 (0.73)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.96 (0.78)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.889\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e[-0.399, 0.459]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eUtility\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.29 (0.74)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4.39 (0.63)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.633\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e[-0.484, 0.297]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.51 (1.00)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e2.41 (1.04)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.747\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e[-0.485, 0.672]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eInterest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e3.54 (0.96)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e3.82 (0.82)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.460\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e[ -0.785, 0.225]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eIntention\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e4.44 (0.70)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e4.28 (0.82)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.271\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\"\u0026minus;\" colname=\"c5\"\u003e \u003cp\u003e[-0.272, 0.592]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003e\u003cem\u003eNote\u003c/em\u003e. Group 1: AI group; Group 2: Teacher group.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eAn independent samples \u003cem\u003et\u003c/em\u003e-test was used to compare the revisions made by Group 1 (AI group) and Group 2 (Teacher group) across several measures(see Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). For replacement, a significant difference was found between the groups, \u003cem\u003ep\u003c/em\u003e = .017, 95% CI [0.656, 6.113]. Group 1 reported a higher mean number of replacements (\u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;11.00, \u003cem\u003eSD\u003c/em\u003e\u0026thinsp;=\u0026thinsp;6.35) compared to Group 2 (\u003cem\u003eM\u003c/em\u003e\u0026thinsp;=\u0026thinsp;7.62, \u003cem\u003eSD\u003c/em\u003e\u0026thinsp;=\u0026thinsp;2.53). However, no significant differences were observed for deletion, \u003cem\u003ep\u003c/em\u003e = .223, 95% CI [-0.459, 1.920], addition, \u003cem\u003ep\u003c/em\u003e = .296, 95% CI [-1.595, 5.133], or re-ordering, \u003cem\u003ep\u003c/em\u003e = .887, 95% CI [-0.504, 0.581]. Still, students enacted more replacements, deletion, and addition when the feedback was labeled as AI-generated feedback than when it was labeled as teacher feedback. Overall, students implemented a larger number of revisions when they believed they were working with AI-generated feedback.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eComparison of revisions\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMeasure\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eGroup 1 \u003cem\u003eM\u003c/em\u003e (\u003cem\u003eSD\u003c/em\u003e)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eGroup 2 \u003cem\u003eM\u003c/em\u003e (\u003cem\u003eSD\u003c/em\u003e)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e95% CI\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eReplacement\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e11.00 (6.35)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e7.62 (2.53)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e[0.656, 6.113]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDeletion\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e2.42 (2.28)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e1.69 (1.98)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.223\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e[-0.459, 1.920]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAddition\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e8.88 (7.10)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e7.12 (4.74)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.296\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e[-1.595, 5.133]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRe-ordering\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.69 (0.65)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.79 (1.13)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e.887\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e[-0.504, 0.581]\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"5\"\u003e\u003cem\u003eNote\u003c/em\u003e. Group 1: AI group; Group 2: Teacher group.\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e"},{"header":"5 Discussion","content":"\u003cp\u003eThis experimental study investigated the presence of bias in the perception and use of AI-generated feedback among CSL learners. The findings reveal a psychologically significant dissociation between cognitive appraisal and behavioral engagement: while students perceived identical feedback more favorably when attributed to a teacher, they made significantly more revisions when the same feedback was labeled as AI-generated. This study demonstrated that students had a less positive perception of identical feedback across all dimensions when they were informed that the feedback was provided by an AI system. This result aligns with a study by Lipnevich and Smith in 2014 on the perception of traditional automated feedback system. They found that students tended to perceive teacher feedback as more accurate and helpful than computer-generated feedback, even though the feedback was identical in content. It has thus become clear that such subjective bias persists to this day, even as AI has become more powerful than traditional AWE systems (Henderson et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). This study also aligns with Ruwe and Kuklick (\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), which showed that an AI label for feedback reduced the perceived trustworthiness of the feedback source. The present study, however, has offered more insights by revealing that an AI label can reduce the positive perceptions across both affective and cognitive dimensions, ranging from interest to perceived accuracy, elaboration, and utility. The results of this study have also advanced the strand of research on the preferences for teacher feedback (Henderson et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Specifically, prior research comparing perceptions of AI feedback with those of teacher feedback has attributed differences in perceptions such as trust to the distinct characteristics of feedback from these sources (e.g., Henderson et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Ranalli, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Tossell et al., \u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Zhan \u0026amp; Yan, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). This study, however, shows that with a different label, students could have different perceptions of the same feedback. Overall, the differing perceptions caused by a labelled source points to the existence of subjective bias toward feedback sources.\u003c/p\u003e \u003cp\u003eInterestingly, the more positive perceptions elicited by the teacher feedback label did not lead to a higher level of engagement with the feedback. This finding challenges the intuitive assumption that positive cognitive appraisal directly translates into greater behavioral engagement\u0026mdash;a phenomenon with important psychological implications. It contradicts assumptions that positive feedback perceptions regarding trustworthiness and feedback quality are associated with greater feedback engagement (Henderson et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Ruwe \u0026amp; Kuklick, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Instead, this study demonstrated that students responded more actively to feedback when informed that it was provided by an AI system, despite expressing less favorable perceptions of the feedback. Specifically, the AI group enacted a larger number of revisions than the teacher group, although the difference was not statistically significant.\u003c/p\u003e \u003cp\u003eContrary to earlier findings (e.g., Liu et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Zou et al., \u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e), which indicated stronger learner engagement with teacher feedback, this study revealed that AI-generated responses can elicit greater learner interaction in certain CALL contexts. The discrepancy between more positive feedback perceptions and a lower level of feedback engagement suggests alternative explanations other than those proposed in prior research. For instance, Liu et al. (\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2024\u003c/span\u003e) found that secondary school students who received AWE feedback on their Chinese writing made fewer revisions compared to peers who received feedback from teachers. The relatively fewer revisions when interacting with AWE feedback could be attributed to the overload caused by feedback quantity and its focus on surface issues. Zou et al. (\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e2025\u003c/span\u003e) found that students showed a higher uptake of teacher feedback on their English writing compared to feedback generated by ChatGPT. They attributed this higher rate of uptake to greater trust in the accuracy of teacher feedback. These explanations do not apply to the present study, in which both groups acted on the same feedback, yet students rated the feedback more positively when they were told it was from a teacher. In other words, trust in teacher feedback is not the sole reason for the higher level of engagement with teacher feedback. Alternative explanations are needed.\u003c/p\u003e \u003cp\u003eThis paper accounts for the discrepancies between feedback perceptions and feedback engagement from a socio-affective psychological perspective (Henderson et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Ruwe \u0026amp; Kuklick, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Teacher feedback involves teacher-student relationships (Gu, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Guo \u0026amp; Xu, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). From a social psychology standpoint, the authority inherent in teacher feedback activates different motivational and emotional responses compared to AI feedback. On the one hand, the authority of teacher feedback could decrease the likelihood of uninvited revisions, that is, revisions that are neither requested nor suggested in teacher feedback. In this experiment, most of the participants were descendants of overseas Chinese nurtured in Chinese culture. They tend to hold high regard for the authority of their teachers. This cultural orientation amplifies the psychological salience of authority and social evaluation. For them, the primary goal of revision behavior is to meet the teacher\u0026rsquo;s requirements and avoid negative evaluations. Due to their goal-oriented and risk-averse nature, they made only minimal corrections to the issues explicitly identified by the teacher. The assumption is that the absence of negative feedback indicates no serious issues in the writing. Unsolicited revisions may imply that teachers are neglecting their responsibilities and have overlooked important points. The threat to teachers\u0026rsquo; authority is particularly alarming when unsolicited revisions can considerably enhance the quality of the writing. Thus, students tend to suppress \u0026ldquo;unnecessary\u0026rdquo; revisions to avoid causing dissonance between teacher feedback and the number of revisions. This behavior reflects impression management concerns and social-evaluative threat\u0026mdash;well-documented psychological phenomena in feedback contexts. On the other hand, responding to teacher feedback may expose students\u0026rsquo; shortcomings and thus incur self-presentation costs (Gu, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). The embarrassment is particularly pronounced when unsolicited revisions fail to improve the writing or, even worse, degrade its quality. Such instances can reveal their deficiencies. Consequently, students were more cautious when revising text in response to teacher feedback than to AI feedback. For segments without teacher feedback, participants assumed the text had no serious issues and therefore made fewer revisions, possibly to avoid unnecessary changes and protect their self-esteem.\u003c/p\u003e \u003cp\u003eIn contrast, the socio-affective psychological factors are less salient in interactions with AI feedback. There is minimal risk of revealing personal deficiencies to AI systems (Henderson et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). From a self-determination theory perspective, AI feedback may better satisfy learners\u0026rsquo; basic psychological needs for autonomy and competence. AI feedback is generally perceived as inferior to teacher feedback. When interacting with AI feedback, students are confident that they can outperform it. Consequently, they make more attempts to revise their texts. Due to its impersonal nature and lack of direct association with external rewards, punishments, or interpersonal evaluations, AI feedback can satisfy learners\u0026rsquo; autonomy needs, thereby fostering autonomous motivation (Zong \u0026amp; Yang, \u003cspan citationid=\"CR52\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Under this motivation, students revise their essays in response to AI-generated feedback with an intrinsic desire to prove their own abilities. Students demonstrate their language skills as superior to the AI tool through text optimization, achieving self-affirmation of human agency. It is no wonder that students addressing computer-generated feedback often display a higher level of self-confidence and a lower level of anxiety (Liu et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). This psychological state\u0026mdash;characterized by reduced evaluative threat and enhanced intrinsic motivation\u0026mdash;explains why revision behaviors extend beyond the scope of the feedback, exhibiting both high frequency and a broad scope when learners believe they are interacting with AI.\u003c/p\u003e"},{"header":"6 Implications","content":"\u003cp\u003eThis study offers important implications for research on AI-generated feedback. Future comparisons between AI and teacher feedback should account for potential subjective biases that learners may hold toward AI systems. It is crucial to differentiate between variations in learners\u0026rsquo; engagement and subsequent learning outcomes that arise from biased perceptions of the feedback source from those that stem from the intrinsic qualities of the feedback itself. The bias is not constrained to trustworthiness or source credibility; it also relates to other aspects of feedback perception, such as perceived utility and cost (Gu, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Gu \u0026amp; Man, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Such differentiation will enable a more precise understanding of how students perceive and enact AI-mediated feedback and help inform the equitable incorporation of AI into language learning contexts. The study also carries implications for the design and delivery of feedback in language teaching. First, the findings provide empirical evidence of potential biases in students\u0026rsquo; perceptions of AI-generated feedback, enriching our understanding of the conditions under which different feedback sources should be selected. Given that teacher feedback is generally preferred over AI feedback, AI-generated feedback should be positioned as supplementary support rather than a replacement for teacher feedback input (Henderson et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Zhang et al., \u003cspan citationid=\"CR51\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). In contexts where teachers face heavy feedback workloads but students still value teacher feedback, AI systems can be strategically integrated into the feedback process. For instance, students may initially rely on AI to refine and revise their drafts before submitting them for teacher assessment. Teachers may also invite students to compose rebuttal letters in response to AI-generated comments to encourage deeper engagement in L2 writing. When clearly framed as a supportive and preparatory tool, AI feedback may enhance students\u0026rsquo; confidence and willingness to revise.\u003c/p\u003e \u003cp\u003eSecond, the findings highlight that students\u0026rsquo; biased perceptions of AI-generated feedback can shape how they interpret and respond to such feedback. Effective feedback design must therefore take into account learners\u0026rsquo; subjective attitudes toward AI tools. Negative or skeptical perceptions may discourage students from engaging with AI feedback, especially when the feedback volume is high and cognitively overwhelming. To counter such biases, teachers can help learners develop an informed understanding of the affordances and limitations of AI systems, fostering strategic and critical use of AI feedback. Additionally, as excessive or overly critical comments may activate learners\u0026rsquo; self-defensive responses and heighten attention to negative aspects of performance (Brummernhenrich et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Huang et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2018\u003c/span\u003e), the quantity and tone of AI feedback should be carefully mediated to create an emotionally supportive learning climate.\u003c/p\u003e"},{"header":"7 Conclusion","content":"\u003cp\u003eThis experimental study provides empirical evidence of subjective bias in students\u0026rsquo; responses to AI-generated feedback among a group of international students learning Chinese as a second language. The findings reveal that students tend to perceive identical feedback more favorably when it is attributed to an experienced teacher. Conversely, when informed that the feedback originates from an intelligent AI system, they are more likely to act upon it, making a greater number of revisions. These results underscore the social-affective dimension of feedback reception and offer new insights into why students may demonstrate heightened engagement when responding to AI feedback. Overall, the study advances understanding of how source cues shape students\u0026rsquo; feedback perceptions and subsequent revisions. The findings of the study point to the need for exploration into the mechanisms behind feedback perceptions, uptake, and engagement in human-AI feedback contexts. Concrete implications for designing and integrating AI feedback into language learning are also discussed.\u003c/p\u003e \u003cp\u003eThis study has several limitations. First, the study was carried out in a controlled laboratory environment, which may constrain the applicability of its findings to authentic, real-world educational settings. In authentic classroom settings, students\u0026rsquo; engagement with feedback can be shaped by multiple factors, including the importance of the writing task, the nature of the teacher-student relationship, and students\u0026rsquo; perceptions of the potential costs and benefits of seeking feedback. Whether subjective bias exists in authentic settings and to what extent such bias influences textual revisions merit replication studies. Second, this experimental study did not control for the number of feedback points provided. AI systems, such as ChatGPT, can generate a large volume of feedback, potentially leading to cognitive overload for students. An excessive amount of feedback may discourage student\u0026rsquo;s feedback engagement. The interaction between biased perceptions and feedback quantity on feedback engagement also merits scholarly attention in future inquiries.\u003c/p\u003e "},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eHuman Ethics and Consent to Participate declarations\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis study was approved by the Research Ethics Committee of College of Chinese Language and Culture，Jinan University (Approval No: 2024-11, Date: 03/09/2024). All procedures performed in this study involving human participants were in accordance with the ethical standards of the institutional research committee and with the 1964 Helsinki declaration and its later amendments. Informed consent was obtained electronically from all participants prior to data collection.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets analyzed during the current study are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgments\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe author declares no competing interests\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthorship contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eH.Y is responsible for all work for this manuscript.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAllen TJ, Mizumoto A. ChatGPT over my friends: Japanese English-as-a-foreign-language learners\u0026rsquo; preferences for editing and proofreading strategies. RELC J. 2024. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1177/00336882241262533\u003c/span\u003e\u003cspan address=\"10.1177/00336882241262533\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAlmusharraf N, Alotaibi H. An error-analysis study from an EFL writing context: Human and automated essay scoring approaches. Technol Knowl Learn. 2022;28(3):1015\u0026ndash;31. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10758-022-09592-z\u003c/span\u003e\u003cspan address=\"10.1007/s10758-022-09592-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBai L, Hu G. In the face of fallible AWE feedback: How do students respond? Educational Psychol. 2016;37(1):67\u0026ndash;81. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/01443410.2016.1223275\u003c/span\u003e\u003cspan address=\"10.1080/01443410.2016.1223275\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBarrot JS. Using automated written corrective feedback in the writing classrooms: effects on L2 writing accuracy. Comput Assist Lang Learn. 2021;36(4):584\u0026ndash;607. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/09588221.2021.1936071\u003c/span\u003e\u003cspan address=\"10.1080/09588221.2021.1936071\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBrummernhenrich B, Paulus CL, Jucks R. Applying social cognition to feedback chatbots: Enhancing trustworthiness through politeness. Br J Edu Technol. 2025. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/bjet.13569\u003c/span\u003e\u003cspan address=\"10.1111/bjet.13569\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen Z, Zhu X, Lu Q, Wei W. L2 students\u0026rsquo; barriers in engaging with form and content-focused AI-generated feedback in revising their compositions. Comput Assist Lang Learn. 2024;1\u0026ndash;21. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/09588221.2024.2422478\u003c/span\u003e\u003cspan address=\"10.1080/09588221.2024.2422478\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCohen J. Statistical power analysis for the behavioral sciences. 2nd ed. ed.). Routledge; 2013.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDing L, Zou D. Automated writing evaluation systems: A systematic review of Grammarly, Pigai, and Criterion with a perspective on future directions in the age of generative artificial intelligence. Educ Inform Technol. 2024;29(11):14151\u0026ndash;203. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10639-023-12402-3\u003c/span\u003e\u003cspan address=\"10.1007/s10639-023-12402-3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eEscalante J, Pack A, Barrett A. AI-generated feedback on writing: Insights into efficacy and ENL student preference. Int J Educational Technol High Educ. 2023;20(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s41239-023-00425-2\u003c/span\u003e\u003cspan address=\"10.1186/s41239-023-00425-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFokides E, Peristeraki E. Comparing ChatGPT's correction and feedback comments with that of educators in the context of primary students' short essays written in English and Greek. Educ Inform Technol. 2024. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10639-024-12912-8\u003c/span\u003e\u003cspan address=\"10.1007/s10639-024-12912-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFu Q-K, Zou D, Xie H, Cheng G. A review of AWE feedback: Types, learning outcomes, and implications. Comput Assist Lang Learn. 2024;37(1\u0026ndash;2):179\u0026ndash;221. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/09588221.2022.2033787\u003c/span\u003e\u003cspan address=\"10.1080/09588221.2022.2033787\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGu H. Exploring factors contributing to feedback-seeking strategies in L2 writing: An extended cost-value framework. Stud Educational Evaluation. 2025;85. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.stueduc.2025.101453\u003c/span\u003e\u003cspan address=\"10.1016/j.stueduc.2025.101453\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGu H, Man D. Understanding the motivations behind feedback-seeking behavior in L2 writing: A situated expectancy-value perspective. Int J Appl Linguistics. 2025. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/ijal.12842\u003c/span\u003e\u003cspan address=\"10.1111/ijal.12842\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo Q, Feng R, Hua Y. How effectively can EFL students use automated written corrective feedback (AWCF) in research writing? Comput Assist Lang Learn. 2022;35(9):2312\u0026ndash;31. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/09588221.2021.1879161\u003c/span\u003e\u003cspan address=\"10.1080/09588221.2021.1879161\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuo Q, Xu Y. The effect of achievement goals on students\u0026rsquo; perceptions of feedback: The mediating roles of teacher-student relationships and academic self-concept. Asia-Pacific Educ Researcher. 2025. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s40299-025-00996-w\u003c/span\u003e\u003cspan address=\"10.1007/s40299-025-00996-w\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHan T, Sari E. An investigation on the use of automated feedback in Turkish EFL students\u0026rsquo; writing classes. Comput Assist Lang Learn. 2022;37(4):961\u0026ndash;85. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/09588221.2022.2067179\u003c/span\u003e\u003cspan address=\"10.1080/09588221.2022.2067179\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHassanzadeh M, Fotoohnejad S. Implementing an automated feedback program for a foreign language writing course: A learner-centric study. J Comput Assist Learn. 2021;37(5):1494\u0026ndash;507. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/jcal.12587\u003c/span\u003e\u003cspan address=\"10.1111/jcal.12587\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHattie J, Timperley H. The power of feedback. Rev Educ Res. 2007;77(1):81\u0026ndash;112. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3102/003465430298487\u003c/span\u003e\u003cspan address=\"10.3102/003465430298487\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHenderson M, Bearman M, Chung J, Fawns T, Buckingham Shum S, Matthews KE, de Mello Heredia J. Comparing Generative AI and teacher feedback: Student perceptions of usefulness and trustworthiness. Assess Evaluation High Educ. 2025;1\u0026ndash;16. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/02602938.2025.2502582\u003c/span\u003e\u003cspan address=\"10.1080/02602938.2025.2502582\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang P-S, Liu C-H, Chen H-C, Sommers S. Attentional bias of students toward negative feedback in bad outcome situations: The mechanism of self-defense. Soc Psychol Educ. 2018;21(3):565\u0026ndash;83. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11218-018-9429-y\u003c/span\u003e\u003cspan address=\"10.1007/s11218-018-9429-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim R. Effects of learner uptake following automatic corrective recast from artificial intelligence chatbots on the learning of English caused-motion construction. Lang Learn Technol. 2024;28(2):109\u0026ndash;33. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/https://hdl.handle.net/10125/73574\u003c/span\u003e\u003cspan address=\"https://doi.org/https://hdl.handle.net/10125/73574\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKoltovskaia S. (2020). Student engagement with automated written corrective feedback (AWCF) provided by Grammarly: A multiple case study. \u003cem\u003eAssessing Writing\u003c/em\u003e, \u003cem\u003e44\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.asw.2020.100450\u003c/span\u003e\u003cspan address=\"10.1016/j.asw.2020.100450\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKoltovskaia S. Postsecondary L2 writing teachers\u0026rsquo; use and perceptions of Grammarly as a complement to their feedback. ReCALL. 2022;35(3):290\u0026ndash;304. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1017/s0958344022000179\u003c/span\u003e\u003cspan address=\"10.1017/s0958344022000179\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLink S, Mehrzad M, Rahimi M. Impact of automated writing evaluation on teacher feedback, student revision, and writing improvement. Comput Assist Lang Learn. 2022;35(4):605\u0026ndash;34. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/09588221.2020.1743323\u003c/span\u003e\u003cspan address=\"10.1080/09588221.2020.1743323\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLipnevich AA, Smith JK. Response to assessment feedback: the effects of grades, praise, and source of information. ETS Res Rep Ser. 2014;2008(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/j.2333-8504.2008.tb02116.x\u003c/span\u003e\u003cspan address=\"10.1002/j.2333-8504.2008.tb02116.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu S, Boudouaia A, Chen X, Li Y. Comparative study about the impacts of Chinese automated writing evaluation (AWE) feedback and teacher feedback on middle school students\u0026rsquo; writing practice. Asia-Pacific Educ Researcher. 2024. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s40299-024-00903-9\u003c/span\u003e\u003cspan address=\"10.1007/s40299-024-00903-9\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLuo L, Yusuf A. Bridging AI and pedagogy: How AI-adaptive feedback shapes Chinese EFL students\u0026rsquo; writing engagement, metacognitive writing strategies, and writing performance. Assess Evaluation High Educ. 2025;1\u0026ndash;24. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/02602938.2025.2548919\u003c/span\u003e\u003cspan address=\"10.1080/02602938.2025.2548919\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMan D. The predictive effects of user perceptions on the willingness to continue using machine translation. Translation Interpreting Stud. 2026. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1075/tis.25014.man\u003c/span\u003e\u003cspan address=\"10.1075/tis.25014.man\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMizumoto A, Shintani N, Sasaki M, Teng MF. Testing the viability of ChatGPT as a companion in L2 writing accuracy assessment. Res Methods Appl Linguistics. 2024;3(2). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.rmal.2024.100116\u003c/span\u003e\u003cspan address=\"10.1016/j.rmal.2024.100116\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOneill R, Russell A. Stop! Grammar time: University students\u0026rsquo; perceptions of the automated feedback program Grammarly. Australasian J Educational Technol. 2019;35(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.14742/ajet.3795\u003c/span\u003e\u003cspan address=\"10.14742/ajet.3795\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePapi M, Bondarenko AV, Wawire B, Jiang C, Zhou S. Feedback\u0026ndash;seeking behavior in second language writing: Motivational mechanisms. Read Writ. 2020;33:485\u0026ndash;505. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11145-019-09971-6\u003c/span\u003e\u003cspan address=\"10.1007/s11145-019-09971-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRanalli J. L2 student engagement with automated feedback on writing: Potential for learning and issues of trust. J Second Lang Writ. 2021;52. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jslw.2021.100816\u003c/span\u003e\u003cspan address=\"10.1016/j.jslw.2021.100816\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRong M, Yao Y, Li Q, Chen X. Exploring student engagement with artificial intelligence-guided chatbot feedback in EFL writing: Interactions and revisions. Comput Assist Lang Learn. 2025;1\u0026ndash;30. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/09588221.2025.2539979\u003c/span\u003e\u003cspan address=\"10.1080/09588221.2025.2539979\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRuwe T, Kuklick L. Quality counts? Examining the role of feedback provider and feedback quality on students' feedback perceptions. Br J Edu Technol. 2025. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/bjet.70011\u003c/span\u003e\u003cspan address=\"10.1111/bjet.70011\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShadiev R, Feng Y. Using automated corrective feedback tools in language learning: a review study. Interact Learn Environ. 2024;32(6):2538\u0026ndash;66. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/10494820.2022.2153145\u003c/span\u003e\u003cspan address=\"10.1080/10494820.2022.2153145\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShi H, Aryadoust V. A systematic review of AI-based automated written feedback research. ReCALL. 2024;36(2):187\u0026ndash;209. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1017/s0958344023000265\u003c/span\u003e\u003cspan address=\"10.1017/s0958344023000265\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSteiss J, Tate T, Graham S, Cruz J, Hebert M, Wang J, Moon Y, Tseng W, Warschauer M, Olson CB. Comparing the quality of human and ChatGPT feedback of students\u0026rsquo; writing. Learn Instruction. 2024;91. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.learninstruc.2024.101894\u003c/span\u003e\u003cspan address=\"10.1016/j.learninstruc.2024.101894\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun B, Fan T. The effects of an AWE-aided assessment approach on business English writing performance and writing anxiety: A contextual consideration. Stud Educational Evaluation. 2022;72. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.stueduc.2021.101123\u003c/span\u003e\u003cspan address=\"10.1016/j.stueduc.2021.101123\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThi NK, Nikolov M. How teacher and Grammarly feedback complement one another in myanmar EFL students\u0026rsquo; writing. Asia-Pacific Educ Researcher. 2021;31(6):767\u0026ndash;79. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s40299-021-00625-2\u003c/span\u003e\u003cspan address=\"10.1007/s40299-021-00625-2\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTian L, Zhou Y. Learner engagement with automated feedback, peer feedback and teacher feedback in an online EFL writing context. System. 2020. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.system.2020.102247\u003c/span\u003e\u003cspan address=\"10.1016/j.system.2020.102247\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTossell CC, Tenhundfeld NL, Momen A, Cooley K, de Visser EJ. Student perceptions of ChatGPT use in a college essay assignment: Implications for learning, grading, and trust in artificial intelligence. IEEE Trans Learn Technol. 2024;17:1069\u0026ndash;81. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/tlt.2024.3355015\u003c/span\u003e\u003cspan address=\"10.1109/tlt.2024.3355015\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evan de Ridder JMM, Berk FCJ, Stokking KM, Ten Cate OTJ. Feedback providers' credibility impacts students' satisfaction with feedback and delayed performance. Med Teach. 2015;37(8):767\u0026ndash;74. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3109/0142159X.2014.970617\u003c/span\u003e\u003cspan address=\"10.3109/0142159X.2014.970617\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVan der Kleij FM, Lipnevich AA. Student perceptions of assessment feedback: a critical scoping review and call for research. Educational Assess Evaluation Account. 2020;33(2):345\u0026ndash;73. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s11092-020-09331-x\u003c/span\u003e\u003cspan address=\"10.1007/s11092-020-09331-x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Z, Han F. The effects of teacher feedback and automated feedback on cognitive and psychological aspects of foreign language writing: A mixed-methods research. Front Psychol. 2022;13:909802. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3389/fpsyg.2022.909802\u003c/span\u003e\u003cspan address=\"10.3389/fpsyg.2022.909802\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWinstone NE, Hepper EG, Nash RA. Individual differences in self-reported use of assessment feedback: The mediating role of feedback beliefs. Educational Psychol. 2021;41(7):844\u0026ndash;62. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/01443410.2019.1693510\u003c/span\u003e\u003cspan address=\"10.1080/01443410.2019.1693510\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWoo DJ, Wang D, Guo K, Susanto H. Teaching EFL students to write with ChatGPT: Students' motivation to learn, cognitive load, and satisfaction with the learning process. Educ Inform Technol. 2024;29(18):24963\u0026ndash;90. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s10639-024-12819-4\u003c/span\u003e\u003cspan address=\"10.1007/s10639-024-12819-4\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYan D, Tian F, Li H, Gao Y, Li J. (2026). A GenAI-supported dialogic model to enhance the functions of peer feedback in EFL writing: A mixed-method interventional study. \u003cem\u003eSystem\u003c/em\u003e, \u003cem\u003e136\u003c/em\u003e. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.system.2025.103895\u003c/span\u003e\u003cspan address=\"10.1016/j.system.2025.103895\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang L, Li R. ChatGPT for L2 learning: Current status and implications. System 124. 2024. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.system.2024.103351\u003c/span\u003e\u003cspan address=\"10.1016/j.system.2024.103351\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhan Y, Boud D, Dawson P, Yan Z. Generative artificial intelligence as an enabler of student feedback engagement: a framework. High Educ Res Dev. 2025;44(5):1289\u0026ndash;304. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/07294360.2025.2476513\u003c/span\u003e\u003cspan address=\"10.1080/07294360.2025.2476513\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhan Y, Yan Z. Students\u0026rsquo; engagement with ChatGPT feedback: Implications for student feedback literacy in the context of generative artificial intelligence. Assess Evaluation High Educ. 2025;1\u0026ndash;14. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/02602938.2025.2471821\u003c/span\u003e\u003cspan address=\"10.1080/02602938.2025.2471821\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang Z, Aubrey S, Huang X, Chiu TKF. The role of generative AI and hybrid feedback in improving L2 writing skills: A comparative study. Innov Lang Learn Teach. 2025;1\u0026ndash;19. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/17501229.2025.2503890\u003c/span\u003e\u003cspan address=\"10.1080/17501229.2025.2503890\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZong Y, Yang L. How AI-enhanced social-emotional learning framework transforms EFL students' engagement and emotional well-being. Eur J Educ. 2025;60(1). \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1111/ejed.12925\u003c/span\u003e\u003cspan address=\"10.1111/ejed.12925\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZou S, Guo K, Wang J, Liu Y. Investigating students\u0026rsquo; uptake of teacher- and ChatGPT-generated feedback in EFL writing: A comparison study. Comput Assist Lang Learn. 2025;1\u0026ndash;30. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1080/09588221.2024.2447279\u003c/span\u003e\u003cspan address=\"10.1080/09588221.2024.2447279\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-psychology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"psyo","sideBox":"Learn more about [BMC Psychology](http://bmcpsychology.biomedcentral.com/)","snPcode":"","submissionUrl":"","title":"BMC Psychology","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"subjective bias, feedback source, AI feedback, teacher feedback, international students, Chinese as a second language","lastPublishedDoi":"10.21203/rs.3.rs-8942475/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8942475/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eWhile AI-generated feedback has been promoted as a supplement to teacher feedback, recent empirical studies have consistently revealed that students prefer teacher feedback to AI feedback. It remains unclear whether such preferences reflect the inherent quality of the feedback or stem from psychological biases toward the feedback source.\u003c/p\u003e\u003ch2\u003eMethod\u003c/h2\u003e \u003cp\u003eThis study investigates the existence of psychological biases towards teacher and AI feedback. An experiment was conducted with 52 advanced learners of Chinese as a second language (CSL). The subjects were randomly divided into two groups: AI-label group and teacher-label group. All participants received identical AI-generated feedback on an essay, but were led to believe it came from either an AI system (AI-label group) or an experienced teacher (teacher-label group). Data included participants\u0026rsquo; perceptions of the feedback measured across seven psychological dimensions (coverage, accuracy, elaboration, utility, cost, interest, and intention), as well as their subsequent textual revisions coded by type and frequency.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eParticipants in the teacher-label group reported consistently more positive perceptions of the feedback across all dimensions, although these differences did not reach statistical significance. However, the AI-label group made significantly more textual revisions, particularly replacements, than the teacher-label group (\u003cem\u003ep\u003c/em\u003e = .017). This reveals a dissociation between cognitive appraisal of feedback and behavioral engagement with it.\u003c/p\u003e\u003ch2\u003eConclusion\u003c/h2\u003e \u003cp\u003eThese findings provide empirical evidence of a psychological bias against AI-generated feedback, wherein the same feedback is perceived less favorably when attributed to AI. However, this negative bias does not translate into reduced behavioral engagement; instead, learners interact more actively with AI-attributed feedback, potentially due to reduced social-evaluative concerns and enhanced autonomy. The study contributes to the psychology of human-AI interaction in educational contexts and highlights the need to consider both cognitive and socio-affective mechanisms in understanding feedback engagement.\u003c/p\u003e","manuscriptTitle":"Students’ psychological biases towards teacher and AI-generated feedback: An experiment study","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-03-12 15:58:14","doi":"10.21203/rs.3.rs-8942475/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2026-03-11T07:41:56+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-09T08:27:51+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-09T07:06:14+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"157877576586883200892203704702009625829","date":"2026-03-09T02:46:01+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"32185267482912267114505219451416691125","date":"2026-03-06T05:50:33+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-03-05T13:48:37+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-03-04T14:00:12+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2026-03-04T13:31:35+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-03-04T12:40:14+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Psychology","date":"2026-02-27T11:18:36+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-psychology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"psyo","sideBox":"Learn more about [BMC Psychology](http://bmcpsychology.biomedcentral.com/)","snPcode":"","submissionUrl":"","title":"BMC Psychology","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"5e6b6856-39e4-4951-a655-6d9d72bd0cd6","owner":[],"postedDate":"March 12th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2026-04-27T16:00:30+00:00","versionOfRecord":{"articleIdentity":"rs-8942475","link":"https://doi.org/10.1186/s40359-026-04568-5","journal":{"identity":"bmc-psychology","isVorOnly":false,"title":"BMC Psychology"},"publishedOn":"2026-04-22 15:56:56","publishedOnDateReadable":"April 22nd, 2026"},"versionCreatedAt":"2026-03-12 15:58:14","video":"","vorDoi":"10.1186/s40359-026-04568-5","vorDoiUrl":"https://doi.org/10.1186/s40359-026-04568-5","workflowStages":[]},"version":"v1","identity":"rs-8942475","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8942475","identity":"rs-8942475","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}