The Effect of AI-assisted feedback vs Teacher corrective feedback on Iranian EFL Learners' Grammatical accuracy

Accurate feedback is essential for developing grammatical accuracy in writing, yet large classes and limited teacher time often constrain feedback quality in Iranian institutes. Building on recent advances in large-language-model (LLM) technology, this study compared the efficacy of AI-assisted corrective feedback, delivered through ChatGPT-4, with traditional teacher-provided feedback. Forty pre-intermediate Iranian EFL learners (age = 18–25; 20 male, 20 female) from Daneshvarane Bartar English Institute in Tabriz, East Azerbaijan, were first homogenized through a PET test and matched for age and class size, then randomly assigned to an AI-feedback group (20) or a teacher-feedback (control) group (20). Over seven 90-minute sessions (five weeks), each learner produced a 150–200-word argumentative essay per week. Both groups completed parallel pre- and post-test essays. Feedback was delivered within 48 hours of each writing submission, with AI feedback provided via Large Language Model (LLMs)’s standardized interface using ChatGPT and teacher feedback delivered by the trained instructor. Grammatical accuracy was operationalized as the number of errors per T-unit and scored independently by two trained raters. An ANCOVA controlling for pre-test performance revealed a significant main effect for feedback type. Post-test adjusted means showed that the teacher corrective feedback group outperformed the AI group. However, both groups from pre-test to post-test had an improvement in accuracy. However, this improvement in the group that received teacher corrective feedback was higher. As a result, in the Iranian context, teacher corrective feedback is more effective. Besides, these results advocate integrating AI tools into writing pedagogy to relieve teacher workload, provide immediate individualized input, and support data-driven instruction. Future research should test hybrid feedback models, track long-term retention, and explore learner perceptions across proficiency levels and institutional contexts.