Aligning AI and Human Feedback in EFL Writing: A Mixed-Methods Comparison of ChatGPT, Peer, and Teacher Feedback

Abstract

This mixed-methods study compared three feedback modalities—ChatGPT-only feedback, ChatGPT plus peer feedback with checklist and peer writing conferences, and ChatGPT plus teacher oral feedback during writing conferences—in an EFL writing course. Twenty-five second-year university English majors participated in a semester-long intervention, completing three writing assignments and receiving feedback according to their group. Quantitative analyses of the 22 participants with complete pretest and posttest data indicated significant overall improvement, with statistically significant gains observed only in the ChatGPT plus peer feedback with checklist and peer writing conferences group; between-group comparisons showed no significant differences. Qualitative analysis of 22 complete open-ended surveys revealed that each feedback method offered distinct advantages and challenges across four dimensions: satisfaction, motivation, engagement, and perceived effectiveness. ChatGPT feedback was valued for its immediacy and technical accuracy, but was seen as impersonal and less supportive of higher-order skills. Peer feedback was perceived as supporting collaboration and motivation, although inconsistencies in quality were noted. Teacher-supported feedback was valued for promoting clearer understanding by contextualizing ChatGPT suggestions, although motivational responses were mixed and conference time was limited. Findings suggest that a blended feedback approach integrating ChatGPT and human feedback may support writing improvement and engagement. Pedagogical recommendations include training students to interpret ChatGPT feedback, improving peer review practices, and increasing structured teacher-student interactions to complement technological affordances.