Generative AI in academic writing: a comparison of human-authored and ChatGPT-generated research article titles

Abstract

Abstract This corpus-based study investigates the extent to which ChatGPT converges or diverges from human practices in the construction of research article (RA) titles in the field of medicine. Two parallel corpora were compiled: one consisting of 300 human-authored RA titles from high-impact medical journals ( The Lancet , JAMA , and The BMJ ), and another comprising 300 ChatGPT-generated titles based on the abstracts of those same articles. Titles were analysed quantitatively and qualitatively in terms of length, form, syntactic structure, and content focus. Descriptive statistics, independent samples t -tests, and chi -square analyses were used to compare means and frequencies of the observed features. Results revealed a close alignment between the two corpora. ChatGPT generated titles that closely mirrored human-authored titles in length, preference for multi-unit forms, and dominance of nominal constructions. At the content level, titles of both corpora were largely methods-focused, though ChatGPT produced proportionally more dataset- and results-oriented titles. These findings suggest that ChatGPT has the capacity to reproduce disciplinary conventions in medical RA titling with considerable accuracy. Meanwhile, subtle differences point to its tendency towards formulaicity and limited stylistic flexibility. The study contributes to ongoing discussions of GAI in academic writing, underscoring both the potential of GAI tools to model disciplinary conventions and the need for critical awareness when incorporating them into EAP pedagogy. The study also offers implications for EAP instruction, genre analysis, and future research on GAI-assisted academic writing.