1　Introduction

This section elaborates on the critical role of self-regulation ability in human-AI collaboration, pointing out that the dynamic complexity of GAI feedback requires learners to flexibly regulate cognition, behavior, and emotion to maintain active engagement, while feedback uptake behavior can also facilitate the internalization of self-regulation; on the basis of acknowledging existing research revealing the correlation between the two, it proposes that how self-regulation ability affects multidimensional GAI feedback engagement remains to be empirically explored, thereby clarifying that this study focuses on the mechanisms of behavioral, cognitive, and emotional feedback engagement to reveal the internal mechanisms of GAI-empowered writing feedback and provide implications for intelligent writing instruction.

2　Literature Review

This section reviews the current research status of AI-assisted writing feedback and feedback engagement, pointing out that GAI feedback has the advantages of immediacy and personalization, feedback engagement covers three dimensions of cognition, behavior, and emotion, and existing research has revealed characteristic differences and influencing factors of engagement under different feedback modes, but in-depth exploration of GAI feedback engagement remains insufficient. Meanwhile, it explains the multidimensional dynamic characteristics of self-regulation ability and its close relationship with academic engagement and feedback processing, emphasizing that in human-AI collaborative environments, learners need stronger self-regulation ability to cope with the complexity of GAI feedback, and then proposes tracking the multidimensional engagement characteristics, changes, and causes of learners with different self-regulation abilities in two rounds of GAI feedback through case studies.

3　Research Design

The study employed purposive sampling to select 6 non-English major freshmen for case tracking, with participants covering high and low self-regulation ability groups, who needed to meet conditions including voluntary participation and provision of complete data from two rounds of GAI-assisted writing revision and human-AI interaction. Data collection adopted multi-source methods, including the Second Language Writing Self-Regulation Questionnaire, human-AI interactive dialogues, writing revision texts, and retrospective interviews; feedback training was conducted before the experiment, and students needed to independently complete two argumentative essay drafts and interact with Wenxin Yiyan for optimization. Data analysis employed content analysis, conducting cyclic coding on 12 human-AI interaction data, 12 revision texts, and 6 interviews, constructing an analytical framework covering three dimensions of cognitive, behavioral, and emotional engagement, and ensuring research reliability and validity through text re-reading, expert discussion, and triangulation.

4　Research Findings

This study examined the multidimensional feedback engagement characteristics and dynamic changes of students with different self-regulation abilities in AI writing feedback mode. Findings revealed that students with high self-regulation ability overall outperformed those with low ability in cognitive, behavioral, and emotional engagement, with significant individual differences within groups. In cognitive engagement, both types of students mainly focused on information organization with minimal cognitive monitoring, but high-level students were better at employing higher-order strategies such as comparison, analysis, and elaboration, and could identify AI feedback limitations and attend to discourse structure; low-level students showed selective processing characteristics, mainly focusing on lexical and grammatical levels, and tended to mechanically accept feedback. In behavioral engagement, both groups showed similar feedback uptake performance, but high-level students employed more diversified revision strategies, often using higher-order strategies such as reorganization and rewriting, while low-level students mostly relied on shallow correction or even direct copying; only individual students used external tools for verification or sought social support. In emotional engagement, both groups showed predominantly positive emotions, but high-level students, due to expectations for deep structural feedback, easily generated disappointment and irritability when unmet, while low-level students showed higher satisfaction with language feedback. Longitudinal analysis showed that students' cognitive and behavioral engagement overall declined across two rounds of tasks, but with different patterns: high-level students showed obvious declines in information organization strategies and feedback uptake frequency, but maintained stable feedback understanding and revision behavior, demonstrating stronger critical awareness and metacognitive regulation; low-level students showed larger declines in understanding, uptake, and revision behavior, influenced by factors such as feedback uniformity and diminished interest. Regarding emotional attitudes, both groups showed positive transformation, with high-level students shifting from skepticism and resistance to recognizing AI's value in language support and reducing cognitive load, while low-level students shifted from irritability due to operational unfamiliarity to satisfaction from improved proficiency and higher scores.

5　Discussion

This section reveals that self-regulation ability is a key factor influencing multidimensional feedback engagement in GAI-assisted writing contexts. Students with high self-regulation ability demonstrated advantages in cognitive and behavioral engagement dimensions, being adept at employing higher-order cognitive strategies, possessing goal-directed monitoring capabilities, and showing diversified and proactive revision behaviors; students with low self-regulation ability relied on shallow revision strategies, showed behavioral rigidity, and were affected by the directness and patternization problems of GAI feedback, with their reflection space being suppressed. The study found interactive relationships among cognitive, behavioral, and emotional engagement dimensions, with successful feedback uptake behavior promoting the internalization of self-regulation ability and enhancing positive emotional experiences. Tracking two rounds of writing tasks showed that students with high self-regulation ability maintained stable feedback understanding and revision behavior, while those with low self-regulation ability declined significantly, revealing the dynamically constructed nature of feedback engagement. Based on these findings, the study proposes a hierarchical teaching pathway of "strategy internalization—cognitive offloading—emotional empowerment," designing differentiated programs for students with different abilities, and recommends optimizing instruction through three-dimensional integration, developing a multi-interactive feedback mode of "AI initial evaluation—peer evaluation—teacher final evaluation," and promoting the transformation of writing instruction toward a "digital-intelligence adaptation" paradigm.

6　Conclusion

This section reveals the regulatory mechanism of self-regulation ability on multidimensional feedback engagement, with high self-regulation ability individuals performing better in cognitive, behavioral, and emotional dimensions and forming positive cycles, while low self-regulation ability individuals experienced cognitive burnout and behavioral superficialization; it also points out limitations including small sample size and frequency analysis methods, suggesting future research to expand scope and employ multimodal data to deeply explore neurocognitive mechanisms.