Does Chatgpt argue through structure or stance? A metadiscourse analysis of English argumentative essays across topics

This study examines the use of interactive and interactional metadiscourse markers in English argumentative essays generated by ChatGPT across a range of social science topics. Drawing on Hyland’s (2005) Interpersonal Model of Metadiscourse, a corpus of seventy-five argumentative essays, comprising fifteen topics with five essays each, was analysed using both qualitative and quantitative methods. The log-likelihood analysis demonstrates a clear preference for interactional markers, particularly hedges, while transitions emerge as the most frequent interactive devices in the overall corpus. These patterns indicate that ChatGPT-generated essays foreground stance-taking and reader engagement rather than explicit structural guidance. At the topic level, interactional markers are especially prominent in essays on sports, gender studies, psychology, and economics, whereas topics such as food and nutrition, and history rely more heavily on interactive resources. Overall, these findings expand the understanding of AI discourse by highlighting both the strengths and limitations of AI-generated writing.

Keywords: Interactive markers, Interactional markers, Metadiscourse, Topic, ChatGPT-produced argumentative essays