Educational potential of large language models in endodontics: A comparative analysis of responses to patient and exam-based questions

Aim: The aim of this study was to evaluate the performance of four large language models (LLMs) – ChatGPT (GPT-4o), Google Gemini (Gemini 1.5 Flash), Microsoft Copilot (GPT-4), and DeepSeek-V3 – in providing clinically and academically relevant information in the field of endodontics. Methods: Ten patient-generated questions were identified using Google Trends based on the keywords “root canal treatment” and “endodontics,” and ten multiple-choice questions were obtained from the official Turkish Dental Specialization Examination (DSE) archives. Each question was independently presented to all LLMs under identical conditions. The responses were evaluated by two experienced endodontists using the Modified Global Quality Scale (MGQS), with disagreements resolved by consensus. Statistical analyses included Kruskal–Wallis, Mann–Whitney U, and Pearson correlation analysis, with a significance level set at α =0.05. Results: No statistically significant differences were found among the models in MGQS scores for either patient-oriented questions ( P = 0.674) or DSE questions ( P = 0.089). However, DeepSeek demonstrated significantly higher performance on academic questions compared to patient-generated queries ( P = 0.030). Google Gemini achieved the highest mean score for patient-oriented responses, although this difference was not statistically significant. Significant positive correlations were observed among Gemini, Copilot, and DeepSeek in patient-level responses. Conclusion: Within the limitations of this study, DeepSeek appears to be more reliable for delivering structured academic content, whereas Google Gemini demonstrates comparatively stronger performance in patient-oriented educational responses.