BACKGROUND: Currently, there is a growing body of research examining the role of generative Artificial Intelligence (GenAI) in medical undergraduate education. However, the findings of these studies exhibit considerable variability. Therefore, this review was designed to systematically evaluate the effectiveness of GenAI tools in improving basic knowledge and clinical skills among undergraduate medical students compared to conventional pedagogy. METHOD: Eleven online databases were systematically searched (e.g., Medline, Embase, CINAHL, APA PsycArticle, APA PsycInfo, ERIC, Scopus, CNKI, WanFang Database, VIP, and SinoMed), from inception to 30 Oct, 2025, and updated to April 24, 2026. Risk assessment was conducted using the Cochrane Risk of Bias tool (ROB 2) and the Medical Education Research Study Quality Instrument (MERSQI). Data were analyzed using RevMan 5.3 and Stata 15.0. Overall effects were estimated using either the fixed effects model or the random effects model. The quality of evidence was evaluated using the Grading of Recommendations, Assessment, Development, and Evaluation (GRADE) framework. RESULTS: This systematic review and meta-analysis included 31 studies involving 2615 medical undergraduates. It found no significant difference in clinical medicine students' knowledge exam score between GenAI and control groups with a small effect size (n = 1592, P = 0.22, I2 = 92%, SMD = 0.22, 95%CI: - 0. 13 to 0.56, 14 studies, random-effect model). However, nursing students showed a significant improvement with GenAI (moderate effect size, n = 205, P = 0.0002, I2 = 76%, SMD = 1.11, 95%CI: 0.54 to 1.69, random-effect model). Compared with traditional teaching, GenAI did not improve the clinical skills for clinical medicine students with a small effect size (n = 726, P = 0.18, I2 = 88%, SMD = 0.27, 95%CI: -0.13 to 0.66, random-effect model). Leave-one-out sensitivity analysis found that one study on fine motor clinical skills had a significant impact on the results. That is, when this study was excluded, the results became statistically significant (P = 0.03). Owing to the poor methodology quality and inconsistency of the results, the evidence quality for the primary outcomes (basic knowledge and clinical skills among clinical medicine students) was downgraded to a low level, while the quality of other evidence was rated as very low. CONCLUSION: GenAI's educational benefits are context- and domain-dependent. Its efficacy is concentrated in cognitive clinical skills, academic writing, and nursing education, while its advantage over high-fidelity video instruction for fine procedural skills remains uncertain. These findings support a differentiated integration strategy rather than uniform adoption. Future high-quality, multi-center, large-sample, and long-term RCTs are still required to robustly confirm its effects on medical undergraduates' knowledge acquisition, clinical skills, and academic writing performance. TRIAL REGISTRATION: The protocol was registered on the International Platform of Registered Systematic Review and Meta-analysis Protocols (INPLASY) and the registration number is INPLASY2025120075.
Source
Xie, Guanli; Liao, Jianglong; Tang, Xiaoxia; Fu, Han; Liu, Duo; Deng, Li; Wang, Tao; Hong, Xin. BMC medical education, 2026. DOI: 10.1186/s12909-026-09458-3