Application of generative artificial intelligence in clinical teaching of colorectal surgery: an exploratory cluster-randomized controlled study

To evaluate a faculty-supervised, ChatGPT-supported blended teaching model during a 4-week colorectal surgery internship. Thirty interns in four pre-existing internship groups participated between June and September 2025. An online draw assigned two groups (15 students) to conventional teaching and two (15 students) to ChatGPT-supported blended teaching. The primary educational outcome was the overall course score: 50% final theoretical examination and 50% practical examination. The practical score comprised 20% laparoscopic procedural assessment, 40% suturing, and 40% case analysis. The main analysis compared the 30 students by allocated teaching arm. A cluster-level analysis was performed as a sensitivity analysis. Mean overall scores were 76.77 (SD 3.80) in the control arm and 86.85 (SD 2.78) in the ChatGPT-supported arm. The student-level mean difference was 10.08 points (95% CI 7.58 to 12.58; P < 0.001). Holm-adjusted P values for the seven secondary outcomes ranged from < 0.001 to 0.003. In the cluster-level sensitivity analysis, the mean difference was 10.13 points (95% CI −0.19 to 20.45; exact P = 0.333). The ChatGPT-supported arm had higher short-term educational scores in the student-level analysis. Because allocation occurred in only four internship groups, the precision of this analysis may be overstated and the findings remain exploratory.