Use of generative Artificial Intelligence tools in medical student research projects: An activity system analysis

Little is known about how medical students use generative AI in research training or how supervisors respond. Using Cultural-Historical Activity Theory (CHAT), this study aimed to characterise how generative AI is reshaping the activity of learning to do research within a mandatory medical student research project, and to identify the contradictions driving that change.

AI-use declarations submitted as assessed coursework by two consecutive student cohorts (2024 to 2025; n = 533), and semi-structured interviews with research supervisors (n = 10) at an Australian medical school, were coded inductively by an independent analyst. Themes were mapped onto CHAT elements and examined for contradictions.

Reported AI non-use fell from 44% to 19% between cohorts, a difference not explained by project type, enrolment or location. Verification of AI output was the most prominent practice in both cohorts, treated as the condition of AI use. Efficiency was the most consistently stated rationale for AI use, often alongside reports that output was unreliable. Prompt refinement emerged as a competence acquired informally but dependent on disciplinary knowledge. Most supervisors felt unprepared, not only to use the tools but to judge when student use was legitimate, and the two most confident had funded that competence themselves. The primary contradiction lay within the object-motive: the written report, historically a workable proxy for research capability, no longer reliably evidences it.

Students generated norms for AI use that lacked institutional standing and were invisible to supervisors, while the programme, responding to student AI use rather than those norms, began remodelling how capability is evidenced, including through oral assessment. Connecting these two processes is the central curricular task, requiring discipline-based AI literacy, forums in which student practice and the rationale for assessment changes are shared, equitable supervisor development, and assessment of reasoning over text production.