Key points are not available for this paper at this time.
Introduction This study investigates the effectiveness of AI-generated feedback compared to teacher-generated feedback on the argumentative writing performance of English as a Foreign Language (EFL) learners at different proficiency levels. Methods Sixty undergraduate students from a writing-focused EFL course in Jordan participated in a quasi-experimental, pretest-posttest study. Participants were stratified into two ACTFL proficiency levels (Intermediate-Low and Advanced-Low) and assigned to either an AI feedback group or a teacher feedback group. Students completed an argumentative writing task, received feedback based on their group, and revised their essays accordingly. An analytic rubric was used to assess writing performance, and inter-rater reliability was established on a stratified 30% subsample to support the validity of the scoring process, with pre- and post-test scores analyzed for gains. Results Results showed significant improvement in writing performance across all groups, regardless of feedback source or proficiency level. Importantly, no statistically significant difference was found between the AI and teacher feedback groups, and the effect size for this comparison was small (Cohen’s d = 0.10). A two-way ANOVA revealed a significant main effect for proficiency level but no significant interaction between feedback type and proficiency. Intermediate-Low learners demonstrated the greatest within-group gains, suggesting that both feedback types were particularly impactful for lower-proficiency students. Discussion The findings underscore the potential of large language models (LLMs), when carefully scaffolded and ethically deployed, to support writing development in EFL contexts. AI-generated feedback may serve as a scalable complement to teacher feedback in large, mixed-proficiency classrooms, particularly when guided by well-developed prompts and pedagogical oversight.
Alnemrat et al. (Tue,) studied this question.