Self-Regulated Learning in Generative AI-Assisted Academic Writing: A Systematic Review of ESL/EFL University Students

Abstract

The growing use of generative AI writing assistants, such as ChatGPT, in ESL/EFL academic contexts has sparked debate about their influence on self-regulated learning (SRL). This systematic review examines 41 empirical studies published between 2021 and March 2026 to evaluate how university students engage with AI tools in relation to SRL. Guided by PRISMA 2020 standards, the review synthesises quantitative, qualitative, and mixed methods research through narrative analysis, drawing on Zimmerman’s (2000) three phase SRL model and metacognitive regulation theory. Three central patterns emerge. First, SRL phases are unevenly addressed: the performance phase dominates (92.7% of studies), while forethought (58.5%) and self-reflection (51.2%) receive less attention, suggesting an incomplete regulatory cycle. Second, AI demonstrates a dual role in metacognitive regulation, facilitating monitoring, evaluation, and revision (68.3%), yet also contributing to cognitive offloading and reduced learner agency (14.6%), with 17.1% of studies reporting mixed outcomes. Third, pedagogical design is critical: structured approaches that integrate AI literacy, guided feedback, and teacher mediation consistently enhance SRL, whereas unguided use yields weaker results. The evidence base remains limited by its concentration in Asian contexts (73.2%), reliance on self-report data, and absence of longitudinal designs, restricting claims about sustained competence. Overall, the review concludes that generative AI can strengthen self-regulated academic writing when embedded in intentional pedagogical frameworks that activate all SRL phases; without such alignment, AI may improve immediate performance but hinder the development of lasting regulatory skills.