Multi-agent debate in L2 argumentative writing: An exploratory study of Chinese EFL learners

Abstract

Argumentative writing poses significant challenges for second-language (L2) learners. Generative AI (GenAI) offers pedagogical support, but existing tools typically act as single assistants and fail to simulate the dynamic, multi-perspective dialogue essential for robust argumentation. We introduce a Multi-Agent Debate (MAD) framework in which multiple AI personas engage in structured debates to foster divergent thinking and expose learners to varied viewpoints. In a one-group pretest–posttest exploratory study (N = 22 Chinese EFL undergraduates), we investigated MAD’s impact on argumentative writing and learner perceptions. Preliminary results indicate that MAD enhanced argument structure and quality, promoted metacognitive awareness, cultivated critical AI skepticism, and reduced anxiety when addressing controversial topics. These findings suggest MAD is a promising approach for improving both L2 argumentative skills and GenAI literacy.