From revision to learning transfer: Self- and socially shared-regulated pathways in AI-assisted feedback

Abstract

Background Generative artificial intelligence (GenAI) has transformed feedback provision in writing practices. However, little is known about the mechanisms through which AI-mediated engagement and revision contribute to longer-term learning transfer. Research aims Framed by theories of self-regulated learning (SRL), socially shared regulated learning (SSRL), and cognitive engagement, the study examined the relationship among engagement, immediate revision gains, and longer-term learning transfer in AI-assisted feedback. Methods This 16-week cluster-randomized experiment involving 108 undergraduates in China compared two feedback modes: AI-assisted self-feedback and AI-assisted peer feedback. Analyses drew on 1100 writing samples (including pretests, four assignments with two drafts each, and posttests), and 432 reflection forms on AI feedback. Findings Linear mixed models revealed a decoupling of learning and performance: while both conditions produced similarly strong second drafts, AI-assisted peer feedback demonstrated significantly greater learning transfer to subsequent tasks. In AI-assisted self-feedback, learning followed a performance-dependent SRL cycle in which AI engagement and revision improvements independently predicted learning. A threshold-like pattern further showed that benefits emerged only when students engaged constructively, measured by reflection forms, to internalize writing standards. Conversely, AI-assisted peer feedback was associated with stronger learning transfer that was less dependent on revision gains or engagement with AI, a pattern consistent with an SSRL perspective emphasizing interactive engagement and co-constructed understanding. Conclusion These findings highlight the importance of revision mechanisms and depth of engagement across feedback contexts. The study contributes to theories of regulation and cognitive engagement and offers pedagogical and design implications for fostering meaningful learning through AI-assisted feedback.

The authors' abstract, as published at the source. Learning and Instruction, 2026 · DOI ↗