From AI Feedback to Language Learning: The Effects of Generative AI Feedback on EFL Learners’ Revision Quality and Delayed Writing Performance

Abstract

Introduction. Generative artificial intelligence is increasingly used as a source of written feedback in second and foreign language writing. The central pedagogical question is no longer whether an AI system can identify or reformulate language problems, but whether learners can convert AI feedback into durable learning that remains visible when the tool is removed.

Objective. This study is designed to examine whether structured generative-AI feedback improves EFL learners’ revision quality and whether such improvement transfers to a delayed writing task completed without AI support. It also examines how learners evaluate, accept, reject, and act on AI feedback.

Method. A mixed-methods, three-arm intervention is proposed for approximately 90 university EFL writers. Participants will complete a baseline writing task and will be allocated to AI feedback, teacher feedback, or hybrid teacher-plus-AI feedback. Across eight weeks, learners will complete four feedback–revision cycles. Immediate post-intervention writing and a two-week delayed writing task will be assessed using an analytic writing rubric. Revision changes, feedback uptake, and learner interviews will provide complementary evidence.

Results. No empirical results are claimed at this manuscript stage because no study dataset was supplied. The planned analyses will compare change over time and between groups, estimate effect sizes and confidence intervals, examine the persistence of learning at delayed post-test, and identify qualitative patterns in feedback use.

Conclusion. The proposed design distinguishes short-term improvement in a revised text from evidence of retained writing development. It is intended to provide a stronger test of whether generative-AI feedback functions as a learning resource rather than simply as an editing service.