1 Introduction

Self-regulation ability is crucial in human-machine collaboration and significantly influences the learning-enhancing effects of generative artificial intelligence (GAI) feedback. The dynamic and complex nature of GAI feedback requires learners to flexibly employ strategies to regulate cognition, behavior, and emotions to maintain active feedback engagement and achieve revision goals. Effective feedback uptake behaviors can promote the internalization of self-regulation and enhance agency in human-machine interaction. Although existing studies have revealed the correlation between feedback engagement and self-regulation strategies, how self-regulation ability influences learners' engagement with GAI feedback still requires further empirical exploration. This study aims to investigate the mechanisms by which self-regulation ability affects feedback engagement across behavioral, cognitive, and affective dimensions, uncovering the intrinsic mechanisms of GAI-empowered writing feedback to provide insights for intelligent writing instruction.

2 Literature Review

This section reviews research on AI-assisted writing and feedback engagement, noting that GAI-assisted writing feedback can improve writing quality and skill development, with the degree of feedback engagement being a key factor in AI's learning-enhancing effects. Feedback engagement exhibits diversity and dynamism, encompassing three core dimensions: cognition, behavior, and affect. Studies have explored the characteristics of learners' feedback engagement under different feedback modes and their influencing factors, revealing close relationships among the dimensions of feedback engagement but varying levels of engagement. Self-regulation ability is closely related to feedback engagement, as it constitutes a multidimensional and dynamic competency system strongly associated with academic achievement and autonomous learning. Self-regulated learning in human-machine collaborative environments has become a research focus. GAI-assisted writing feedback is characterized by complexity and dynamism, requiring learners to possess stronger self-regulation abilities to integrate information, flexibly adjust strategies, and efficiently process machine feedback. This study adopts a case study approach to deeply track the characteristics, changes, and causes of cognitive, behavioral, and affective engagement among foreign language learners with varying self-regulation abilities in GAI-assisted writing feedback.

3 Research Design

This chapter introduces the research design, including participants, data collection, and data analysis. The participants were six non-English major freshmen, divided into high and low self-regulation ability groups through purposive sampling. Multisource data collection methods were employed, including scales, human-machine interaction dialogues, writing revision texts, and retrospective interviews, to explore the impact of self-regulation ability on feedback engagement. Data analysis involved two rounds of data collection, capturing human-machine interaction data, writing revision texts, and retrospective interviews to construct a framework for analyzing writing feedback engagement across cognitive, behavioral, and affective dimensions. Content analysis was used for iterative coding and categorization, calculating the frequency of engagement across dimensions and their sub-indicators, with triangulation ensuring the reliability and validity of the findings.

4 Findings

The study found that students with high self-regulation ability exhibited higher cognitive, behavioral, and affective engagement than those with low self-regulation ability, with individual differences observed in both groups. In cognitive engagement, high self-regulation students more frequently employed higher-order cognitive strategies, such as comparing, analyzing, and elaborating on feedback information, while low self-regulation students tended to selectively process feedback. In cognitive monitoring, high self-regulation students adjusted their interaction instructions with AI based on revision goals, whereas low self-regulation students mechanically accepted feedback. In feedback comprehension, high self-regulation students could identify the limitations of AI feedback and focus on discourse coherence and sentence structure, while low self-regulation students primarily attended to vocabulary and grammar-level feedback. In behavioral engagement, high self-regulation students demonstrated more successful uptake behaviors and diversified revision strategies, while low self-regulation students relied on superficial strategies. In affective engagement, high self-regulation students held positive attitudes toward feedback content and quality, exhibiting deep cognitive processing behaviors, whereas low self-regulation students, while acknowledging the efficiency of feedback, limited their revisions to linguistic aspects. Both groups expressed minor negative attitudes toward feedback, primarily concerning insufficient feedback intelligence, monotonous content, and inconsistent scoring standards. Analysis of revision texts, human-machine interaction dialogues, and interview data across two writing tasks revealed an overall decline in students' cognitive and behavioral engagement, though the patterns of change differed by self-regulation ability. High self-regulation students showed significant declines in information organization strategy use and feedback uptake frequency but maintained stable revision behaviors and feedback comprehension, while low self-regulation students exhibited weaker feedback comprehension and fewer revision behaviors in the second round. In affective engagement, both groups demonstrated a positive shift in attitudes toward GAI feedback, with high self-regulation students recognizing its role in resolving language issues, reducing revision-related psychological stress, providing clear revision directions, and alleviating cognitive load. Low self-regulation students, as their proficiency improved, gradually found the revision process enjoyable, praised the AI's capabilities, and felt satisfied with improved text scores.

5 Discussion

Through multisource data analysis, this study found that in GAI-assisted writing contexts, self-regulation ability is a key factor influencing multidimensional feedback engagement. High self-regulation students performed better in cognitive and behavioral engagement, employing higher-order cognitive strategies and goal-oriented monitoring to effectively organize and comprehend feedback information. In contrast, low self-regulation students relied on superficial revision strategies, exhibiting rigid revision behaviors. High self-regulation students demonstrated diversified and proactive revision behaviors, reflecting deep cognitive processing, while low self-regulation students displayed signs of affective fatigue. Self-regulation ability modulates learners' feedback engagement patterns by influencing cognitive processing depth, behavioral flexibility, and affective experiences. The study also found that high self-regulation students maintained stable feedback comprehension and revision behaviors through active strategy use, whereas low self-regulation students showed significant declines in these areas. This study offers important implications for GAI-empowered writing instruction, recommending tiered teaching for students with varying self-regulation abilities, enhancing feedback personalization, human-machine interaction literacy, and metacognitive training, and adopting diversified interactive feedback modes to shift learners from passive acceptance to active negotiation, thereby improving human-machine collaborative co-construction abilities.

6 Conclusion

This chapter summarizes the impact of self-regulation ability on multidimensional feedback engagement among six students in GAI-assisted writing. The study found that high self-regulation students exhibited more active feedback engagement across cognitive, behavioral, and affective dimensions, while low self-regulation students showed cognitive fatigue and superficial behavioral engagement. Limitations include the small sample size and single analytical method. Future research could expand the sample scope and incorporate multimodal data analysis to further explore the neurocognitive mechanisms of feedback acceptance and revision behaviors.