Large language model-driven Socratic questioning for endodontic case analysis: A randomised controlled trial

Endodontic education requires structured clinical reasoning practise, but resource constraints often limit the delivery of case-based instruction. This study developed a large language model (LLM)-based Socratic questioning system for endodontic case-analysis training and evaluated its feasibility, short-term educational outcomes, learner experience, and direct faculty teaching time requirements compared with case-based learning (CBL).

Seventy-nine third-year dental students who had completed endodontic theory courses were randomised to either an AI-guided group (n = 40) or a CBL group (n = 39). The AI system was built using the FastGPT framework and included three modules: a Socratic questioning engine, a multidimensional case-analysis scoring system, and an adaptive exercise recommendation module. The CBL group received instructor-facilitated small-group teaching from experienced endodontic instructors. Both groups completed a standardised one-hour supplementary intervention before independent simulated case assessment. The primary outcome was the short-term total simulated case-analysis score. Secondary outcomes included domain-specific scores, learner experience, and direct synchronous faculty teaching time.

Seventy-seven students completed the study (completion rate 97.47%). Both groups demonstrated significant within-group improvements from baseline (AI: 78.02 ± 4.93 to 83.97 ± 6.69, p < 0.01; CBL: 78.54 ± 5.32 to 85.13 ± 5.90, p < 0.01). No statistically significant between-group difference was detected in post-training total scores (mean difference: 1.16 points; 95% CI: -1.68 to 4.00; p = 0.422) or score improvement (mean difference: 0.64 points; 95% CI: -1.74 to 3.02; p = 0.593). Domain-specific and learner experience scores showed no statistically significant between-group differences after correction for multiple comparisons. After deployment, the AI-guided intervention required substantially less direct synchronous faculty teaching time per student than CBL (0.026 vs. 0.45 h).

This single-center randomised educational trial indicates that an LLM-based Socratic questioning system was feasible for standardised endodontic simulated case-analysis training. After a one-hour intervention, students' structured case-analysis performance did not differ significantly from that of the faculty-led CBL group. The system reduced direct synchronous faculty teaching time after deployment, serving as a valuable supplementary tool to address resource constraints in endodontic education. Future multicentre longitudinal studies are needed to evaluate long-term knowledge retention and transferability to authentic clinical care.