Dr. CHAN Hon-Tung, ThomasThomasDr. CHAN Hon-TungKwan, Becky Siu ChuBecky Siu ChuKwan2025-08-012025-08-012024Chan, H. T., & Kwan, S. C. (11 Mar 2024). How rhetorical can AI-generated writing be? the case of research article abstracts in applied linguistics. 58 th RELC International Conference, RELC International Hotel, Singapore.http://hdl.handle.net/20.500.11861/24388Since its release by OpenAI in 2022, ChatGPT has rapidly attracted scholarly attention worldwide. While some scholars consider its use as a threat to academic integrity, some argue that it can serve as an aid to writing, as in the case of the research article (RA) abstract. Because of their rather predictable information structure and formulaic lexico-grammatical patterns, it has been claimed that RA abstracts can be generated using ChatGPT, which bear a resemblance to authentic ones that even some experienced researchers may not be able to distinguish. Testing of the authenticity of ChatGPT-generated RA abstracts, however, has so far been done based on human judgment, which may be subjective and not reliable. In this paper, we argue for robust benchmarking methods that involve systematic analyses of rhetorical and discursive features that characterize RA abstracts. One feature that has been found promising is the rhetorical move model. Drawing on Lorés’s (2004) study, which suggests that RA abstracts can follow the IMRD structure, the three-move patterns of the CARS model, or a combination of both, we compare the rhetorical structures of AI- and humangenerated RA abstracts sourced from journals in Applied Linguistics. Findings reveal that some rhetorical moves of the CARS model found in human-generated RA abstracts are absent in many of the AI-generated ones. Pedagogical implications for using AI-generated writing will be discussed.enConference Paper