Can AI mirror the human voice? Stance in scholar-written and ChatGPT-generated medical abstracts

The Covid-19 pandemic has intensified the need for the swift dissemination of medical research, reinforcing the centrality of abstracts' as concise summaries of scientific findings. Concurrently, advances in Large Language Models (LLMs) such as ChatGPT have prompted growing interest in their potential to generate abstracts, raising questions about whether AI can replicate the rhetorical and interpersonal strategies characteristic of academic discourse. One such strategy is stance, defined as the expression of attitudes, evaluations, and authorial positioning, which remains underexplored in research on AI-generated medical writing. Drawing on Hyland’s (2005a) stance model, we compare stance features in scholar-written and ChatGPT-generated Covid-19 abstracts, with the scholar-authored abstracts collected from three high-impact medical journals. The analysis reveals that scholars employed stance markers to convey epistemic caution and construct a credible authorial voice, whereas ChatGPT employed a higher overall frequency of stance markers, especially boosters and attitude markers. These differences reveal distinct rhetorical orientations, contributing to debates on authorial voice in AI-generated academic writing and underscoring the pedagogical value of raising students' awareness of stance-taking in human and AI-generated texts.