Revising with Intelligence: ChatGPT Feedback and Its Impact on EFL Students’ Revision and Self-Efficacy

Abstract

While prior research on AI-assisted writing has primarily focused on surface-level gains in accuracy or anxiety reduction, this study investigates how ChatGPT-supported feedback affects EFL learners’ writing development across behavioral, cognitive, and affective dimensions. Employing a quasi-experimental mixed-methods design, 82 universitylevel EFL students were assigned to either a ChatGPT-supported or teacher-supported writing group. Both groups completed three writing tasks over ten weeks. The ChatGPT group received only ChatGPT feedback, while the teacher group received conventional teacher comments. Pre- and post-intervention measures included writing self-efficacy questionnaires, writing samples, and semi-structured interviews. Quantitative results showed that ChatGPT-supported feedback significantly increased revision productivity and led to more macro-level and content-based revisions. Self-efficacy gains in the ChatGPT group were also significantly higher across all three measured dimensions: substantive revision, discourse organization, and writing conventions. Qualitative findings revealed that students in the ChatGPT group perceived the feedback as clearer, more actionable, and more dialogic, which promoted revision ownership, strategic engagement, and confidence. These findings suggest that ChatGPT can function as a cognitive and emotional scaffold, enabling deeper learner engagement with feedback and supporting self-regulated revision. In contrast, teacher feedback, while valued for its credibility, was perceived as less interactive and less helpful for content-level improvement. This study highlights the pedagogical potential of integrating generative AI into process-based writing instruction. By fostering greater revision productivity, deeper feedback engagement, and enhanced writing self-efficacy, ChatGPT can serve as a catalyst for developing more autonomous and reflective EFL writers.