Exploring learners’ self-regulated learning behaviors in generative AI-assisted academic writing: a cluster and sequential analysis

Despite the growing integration of generative artificial intelligence (GenAI) into academic writing, learners’ self-regulated learning (SRL) behavior patterns and their relationships with individual characteristics remain insufficiently understood. Drawing on 37 hours of screen-recorded GenAI-assisted writing data, comprising 3268 behavioral units from 30 graduate students, the present study examines the dynamic SRL behaviors of students across different learner characteristic clusters in GenAI-assisted academic writing. This study employs multiple learning analytics methods (clustering, descriptive statistics, non-parametric tests, and lag sequential analysis) and incorporates English proficiency and final writing performance as key learner characteristics. The results show that (a) three student clusters emerge based on English proficiency and final writing performance: the Higher-English-Proficiency–Lower-Writing-Performance group, the Higher-English-Proficiency–Higher-Writing-Performance group, and the Lower-English-Proficiency–Higher-Writing-Performance group; (b) higher-performing groups exhibit more cyclical and well-coordinated SRL behavioral patterns, whereas the lower-performing group demonstrates relatively less coordinated SRL behavioral patterns; and (c) among higher-performing students, those with lower English proficiency tend to engage more extensively in SRL behaviors and achieve comparable writing performance; however, this pattern also raises concerns about a possible overreliance on GenAI. Based on these findings, a series of theoretical and pedagogical implications are proposed to support the enhancement of students’ SRL in GenAI-assisted academic writing contexts.