Prompt-Constrained AI Case Review for Immediate Emergency Assessment in Novice Medical Students: A Randomized Study

Objective To determine whether prompt-constrained generative AI-mediated structured case review improves immediate emergency-assessment task performance and short-term knowledge acquisition among novice medical students. Methods In this single-center randomized educational study, 96 first-year clinical medicine students received a standardized 40-min emergency medicine session on 14 April 2026 and were randomized to AI-mediated structured case review or conventional PPT-based review. The AI group used Doubao 2.0 with fixed cases, a predefined knowledge boundary, a structured response template, and standardized feedback. Both groups had 15 min of review followed immediately by assessment. After six exclusions for missing data or incomplete records, 90 students were analyzed (46 AI; 44 conventional). The primary outcome was 5-point standardized initial emergency-assessment task performance; the secondary outcome was a 100-point written test. Results The AI group scored higher on emergency-assessment task performance (4.25±1.04 vs 3.15±0.93; mean difference, 1.10; 95% CI, 0.69 to 1.52; Hedges’ g=1.10; P<0.001) and written knowledge testing (89.28±8.33 vs 82.23±8.78; mean difference, 7.06; 95% CI, 3.47 to 10.64; Hedges’ g=0.82; P<0.001). Conclusion Prompt-constrained AI-mediated case review improved immediate performance on a closely aligned emergency-assessment task and short-term knowledge scores within a predefined instructional scope; findings should not be interpreted as evidence of independent AI effects, broad clinical competence, or real-world emergency performance.