Rhetorical distinctions: Comparing metadiscourse in essays by ChatGPT and students

This study investigates the use of metadiscourse in argumentative essays generated by ChatGPT compared to those written by British university students. Using Hyland’s (2005) framework, we analysed interactive and interactional metadiscourse to uncover rhetorical and linguistic distinctions. The findings reveal that ChatGPT essays, though structurally coherent and logically organised, exhibit a significantly lower frequency of interactional metadiscourse, such as hedges, boosters, and attitude markers, leading to a more impersonal and expository tone. Conversely, student essays demonstrate higher rhetorical engagement, employing nuanced stance markers and personalised expressions to foster reader interaction. ChatGPT prioritises clarity and structural coherence through transitions and endophoric markers, reflecting its algorithmic nature and training. The variability in student writing highlights the influence of individual style and instructional practices. These differences underscore the complementary roles of AI and human authorship in academic writing, with implications for pedagogy. This research advances our understanding of the rhetorical strategies employed by large language models and their potential in academic contexts