Exploring the effects of AI-generated feedback on EFL writing fluency and quality: a process and product-based perspective

Abstract

As artificial intelligence (AI) tools become increasingly embedded in educational contexts, their role in supporting EFL writing development warrants closer empirical examination. Grounded in Flower and Hayes' Cognitive Process Model and informed by psycholinguistic research using keystroke logging, this exploratory pilot study explores the potential effects of AI-generated feedback on EFL learners' writing quality and writing processes from both product- and process-oriented perspectives. Using a between-subjects experimental design, fourteen EFL undergraduates were randomly assigned to either a ChatGPT-assisted revision group or a self-revision control group. Participants produced an initial draft and subsequently revised their texts under their assigned condition. Writing processes were examined through Inputlog keystroke data capturing pausing behavior and revision patterns, while writing quality was assessed using an analytic rubric with established inter-rater reliability. The findings indicate that both revision conditions improved writing quality; however, participants who received AI-generated feedback achieved significantly higher post-revision writing quality scores than those who relied on self-revision. AI-assisted revision was also associated with reductions in several pausing- and revision-related measures, suggesting changes in observable writing behavior during revision. These findings highlight the potential of AI-generated feedback to support both the product and process dimensions of EFL writing.