Revising with Humans and AI: Effects of Teacher-, GenAI-, and Blended-Feedback on EFL Writing Outcomes
TL;DR
GenAI-feedback was associated with stronger performance during immediate revision, whereas teacher-feedback was associated with stronger performance on a subsequent writing task, suggesting that pedagogical uses of GenAI-feedback should consider how its efficiency and comprehensiveness can be integrated with teacher-guided interpretation, reflection, and interpersonal support.
Abstract
This mixed-methods, intact-class quasi-experimental study compared teacher-feedback, GenAI-feedback, and teacher–GenAI-blended-feedback in supporting Chinese EFL learners’ writing outcomes. A total of 109 first-year university students from three intact College English classes participated in a six-week intervention. Quantitative outcomes included performance on two revision tasks, performance on a new post-intervention narrative-writing task, writing motivation, and self-efficacy; open-ended questionnaires examined learners’ perceptions of teacher- and GenAI-feedback. Baseline-adjusted MANCOVA showed a significant multivariate effect of feedback type on writing performance, Pillai’s Trace = 0.281, F(6, 208) = 4.90, p < 0.001, partial η2 = 0.140. GenAI-feedback was associated with stronger immediate revision performance: on the first revision, the GenAI-feedback group outperformed the teacher-feedback group (d = 0.96, p < 0.001) and the blended-feedback group (d = 0.55, p = 0.046); on the second revision, both the GenAI-feedback and blended-feedback groups outperformed the teacher-feedback group (ds = 0.58 and 0.49, ps = 0.028 and 0.047, respectively). In contrast, on the new post-intervention writing task, the teacher-feedback group outperformed both the GenAI-feedback group (d = 0.59, p = 0.026) and the blended-feedback group (d = 0.81, p = 0.001). Feedback type did not significantly affect motivation or self-efficacy, Pillai’s Trace = 0.008, F(4, 208) = 0.21, p = 0.935, partial η2 = 0.004. Qualitatively, learners valued teacher-feedback for its clarity, personalization, and emotional support, whereas GenAI-feedback was valued for its efficiency, comprehensiveness, and consistency. These findings point to different short-term strengths of the two feedback approaches: GenAI-feedback was associated with stronger performance during immediate revision, whereas teacher-feedback was associated with stronger performance on a subsequent writing task. The findings suggest that pedagogical uses of GenAI-feedback should consider how its efficiency and comprehensiveness can be integrated with teacher-guided interpretation, reflection, and interpersonal support.