Generative Artificial Intelligence Use Among Medical Students: Learner Perspectives for Clinical Educators.
TL;DR
In this multi-institutional US convenience sample, students reported frequent AI tool use, although guidance needs were common, and educators should not assume AI is absent from informal learning and may need to clarify expectations, privacy boundaries, disclosure norms and AI-output appraisal strategies.
Abstract
Background
Generative artificial intelligence (AI) is readily available to medical students, yet educators may not know how students use these tools or what guidance students perceive as missing. We examined student-reported AI use, confidence, concerns and institutional support to inform practical guidance.
Methods
We conducted an online survey of medical students recruited primarily through US allopathic medical school peer networks from 15 July to 1 September 2025. We used non-probability sampling; because dissemination relied on voluntary forwarding, we could not calculate a response rate. Outcomes included AI use, frequency, usefulness, confidence, concerns and institutional support. We used descriptive statistics and Wilson 95% confidence intervals.
Results
Of 552 students from at least 37 institutions who answered at least one item, 517 answered the AI-use and frequency items. Across item-specific denominators, 87.0% used text-based generative AI, 74.7% used AI tools at least weekly, and 67.8% rated AI tools moderately or extremely useful. Only 50.3% were moderately or extremely confident evaluating AI-generated content, and 12.2% reported use of an institution-sponsored generative AI tool. Overall, 64.2% wanted more training or clearer guidance, and 78.7% reported academic integrity and/or ethical concerns.
Conclusions
In this multi-institutional US convenience sample, students reported frequent AI tool use, although guidance needs were common. These findings are not national prevalence estimates. These learner perspectives suggest educators should not assume AI is absent from informal learning and may need to clarify expectations, privacy boundaries, disclosure norms and AI-output appraisal strategies.
