Eka Fadilah
Artificial intelligence (AI) is increasingly used in writing instruction, yet its role in collaborative writing remains underexplored. While prior research has focused on individual support and automated feedback, less attention has been given to how AI can be integrated into peer interaction to support learners’ reasoning. This study examines AI-mediated peer collaboration (using ChatGPT) in source- based argumentative writing in an EFL context. Using a pre–post design, 22 Indonesian EFL undergraduates completed individual writing tasks before and after three collaborative sessions in which AI use was constrained to support idea exploration. Learners’ texts were analyzed using Toulmin-based measures of argument structure, with Legitimation Code Theory (LCT) used to support the interpretation of meaning movement across levels of abstraction. Peer interaction data were used to examine how AI was taken up during collaboration. Results indicate limited change in surface features but clear gains in warrants, backing, and rebuttals, indicating stronger inferential reasoning. Interactional analysis further revealed a shift from form-focused work to reasoning-oriented dialogue, with learners evaluating and selectively using AI input. These findings suggest that higher-order reasoning developed through interaction shaped by pedagogically structured, inquiry-oriented AI use, in which AI functioned as a mediational resource.