JMIR Publications

ABSTRACT

Background:

Artificial intelligence (AI) is becoming increasingly embedded in healthcare, yet evidence on how medical students view and experience AI remains fragmented. Prior reviews were limited by mixed health professions populations, smaller evidence bases, and incomplete assessment of heterogeneity and evidence certainty.

Objective:

This review aimed to estimate the prevalence of 9 pre-specified attitudinal, perceptual, and self-reported familiarity outcomes among medical students worldwide and to identify moderators of between-study heterogeneity.

Methods:

We searched 6 databases (PubMed/MEDLINE, Embase, Web of Science, Scopus, PsycINFO, and Cochrane CENTRAL) from inception to April 1, 2026, supplemented by citation searching. Eligible studies reported quantitative outcomes among students enrolled in medical education programs (MD, MBBS, MBChB, or DO). Two reviewers independently screened records and assessed risk of bias using the Joanna Briggs Institute checklist. Proportion outcomes were pooled using Freeman-Tukey double arcsine transformation with DerSimonian-Laird random-effects models and Hartung-Knapp-Sidik-Jonkman adjusted confidence intervals. Prediction intervals, pre-specified subgroup analyses, sensitivity analyses, and Grading of Recommendations, Assessment, Development and Evaluations (GRADE) assessments were performed.

Results:

Of 13,008 records, 98 cross-sectional studies published between 2019 and 2026 were included, representing more than 40,000 medical students from 37 countries. Approximately three-quarters of students reported positive attitudes toward AI (76.9%, 95% CI 72.4%-81.2%; k=46), perceived AI as beneficial to their future career (77.9%, 95% CI 69.5%-85.3%; k=17), and supported curricular integration (76.3%, 95% CI 71.6%-80.7%; k=40). Concern about physician replacement was reported by 40.4% (95% CI 34.4%-46.6%; k=34). Self-reported familiarity was 63.9% (95% CI 56.6%-70.9%; k=53), with extreme heterogeneity (I²=99.5%) and a very wide prediction interval (8.2%-100.0%). Familiarity and prior AI tool use were higher in studies published from 2023 onward (both P<.001). Replacement concern was lower in high-income settings and in studies evaluating domain-specific AI (both P<.01). Sensitivity analyses were robust, whereas GRADE certainty was very low for most outcomes.

Conclusions:

Medical students are generally receptive to AI and strongly support its inclusion in medical education, but familiarity remains uneven and pooled estimates vary substantially across settings. By focusing specifically on medical students and incorporating prediction intervals and certainty appraisal, this review provides a more cautious and educationally relevant synthesis than earlier mixed-population reviews. The findings support locally adapted AI literacy curricula that emphasize clinically grounded applications, critical appraisal of AI outputs, and ethical reasoning. Clinical Trial: Not applicable.