The Effect of ChatGPT-Generated Reflective Feedback on the Coherence and Lexical Complexity of Intermediate EFL Learners’ Argumentative Essays at an Iranian Language Institute

Artificial intelligence has become a more prominent part of language education, offering new ways to give feedback on second language writing. Recently, large language models like ChatGPT have gained attention for their ability to create personalized responses to students' texts and assist in revising them. However, evidence on how effective these tools are in EFL writing contexts is still limited. This study looked at how ChatGPT-generated feedback affects the coherence and lexical complexity of argumentative essays written by intermediate EFL students at the Iran Language Institute. A quasi-experimental design with non-equivalent groups was used. Fifty learners took part in the study; one group received AI-generated feedback while another received traditional teacher feedback. Over sixteen weeks, participants completed pretest, posttest, and delayed posttest writing tasks. Their essays were evaluated using a rubric focused on coherence and lexical complexity. Analysis of covariance (ANCOVA) showed significant differences between the two groups after adjusting for pretest performance. Learners who received ChatGPT feedback earned higher scores on both writing quality dimensions and retained these improvements in the delayed posttest. Questionnaire results indicated that participants generally had a positive view of AI-assisted feedback, especially regarding revision support and vocabulary improvement. The findings suggest that reflective feedback from large language models can lead to enhancements in key areas of argumentative writing. This study also provides evidence from a less-explored EFL context, contributing to ongoing conversations about the role of AI in language learning and writing instruction.

کلیدواژه ها:

AI-generated feedback ، ChatGPT ، argumentative writing ، coherence ، lexical complexity ، learner attitudes