Comparing AI and Teacher Corrective Feedback on Iranian EFL Learners’ Essay Writing Skills
Résumé
Aims: This study aimed to compare the effectiveness of teacher-generated versus AI-generated corrective feedback on the essay writing skills of Iranian intermediate EFL learners. Study Design: Quasi-experimental design. Place and Duration of Study: Department of TEFL, Islamic Azad University, North Tehran Branch, conducted over a 16-week period in 2025. Methodology: A total of 80 Iranian intermediate EFL learners were selected through convenience sampling and divided into two equal groups: Teacher-Generated Corrective Feedback Group (TGFG, n=40) and AI-Generated Corrective Feedback Group (AGFG, n=40). Both groups participated in 16 instructional sessions, receiving feedback on their essays either from a human instructor or an AI-based application (ChatGPT). Writing performance was assessed using IELTS-based pretests and posttests, evaluated based on content, organization, vocabulary, language use, and mechanics. The 6+1 Traits Writing Rubric was used for scoring, and inter-rater reliability was established. Results: While both groups showed improvement in essay writing performance, the AI-generated feedback group (M = 18.08, SD = 1.32) significantly outperformed the teacher-generated group (M = 16.93, SD = 1.29) on the posttest. An independent-samples t-test indicated a significant difference between the two groups (t(78) = 3.92, P = .000, Cohen’s d = .892), favoring the AI feedback group. Conclusion: The findings suggest that AI-generated corrective feedback is more effective than teacher-generated feedback in improving the essay writing performance of EFL learners. These findings have important implications for future teaching practices, particularly in enhancing learner autonomy and reducing teacher workload in large classrooms. AI tools can serve as a reliable and efficient alternative in writing instruction, particularly in contexts with limited instructional resources. However, their effectiveness may vary depending on learners’ proficiency levels, task complexity, and their familiarity with digital tools.