Comparing ChatGPT and teacher feedback for EFL writing development

Abstract

This mixed-methods study compared ChatGPT-generated and teacher-written feedback conditions in university EFL writing and examined how learners perceived the revision value of each condition. A nonrandomized pretest-posttest design was combined with open-ended post-intervention responses. Participants were 46 upper-intermediate learners in a ChatGPT feedback condition (n = 25) or a teacher-written feedback condition (n = 21). Across four weeks, both conditions completed the same instructional sequence and two feedback-revision cycles. Writing was scored for task achievement, coherence and cohesion, lexical resource, grammatical accuracy, and an overall composite. Baseline-adjusted models showed a robust difference favoring the ChatGPT condition for overall writing, F(1, 43) = 16.86, p < .001, partial eta squared = 0.282. Unadjusted domain tests favored the ChatGPT condition for task achievement, coherence and cohesion, and lexical resource; after Holm adjustment across the four domains, only task achievement remained below 0.05. Grammatical accuracy did not differ between conditions. Learners described ChatGPT feedback as immediate and elaborated, particularly for content, organization, and vocabulary, while also reporting concerns about trust and contextual fit. Teacher-written feedback was perceived as selective, context-sensitive, and authoritative. The findings support a condition-level advantage in overall writing and suggest different perceived revision affordances, but they do not isolate feedback source from differences in feedback elaboration and immediacy.