Bülent Kadir TARTUK
Journal of Dental Sciences and Education - 2026;4(3):84-90
Aims: This study compared the performance of the ChatGPT, Gemini, Copilot, and DeepSeek large language model-based chatbots in answering prosthodontic questions from the Turkish Dental Specialty Exam (DUS). Methods: A total of 126 prosthodontic questions from DUS examinations conducted between 2012 and 2021 were included in the study after excluding visually based and canceled questions. The questions were classified as knowledge-based and case-based. Additionally, they were categorized according to topic areas as fixed prosthodontics, removable prosthodontics and implant-supported prosthodontics. The data were analyzed using the generalized estimating equations model, which accounts for repeated measurements, and differences between models were evaluated using Bonferroni-corrected post hoc tests. Results: All chatbot models included in this study achieved an overall accuracy rate above 80% (ChatGPT-4o: 85.71%; Copilot: 83.33%; Gemini 3 Flash: 81.75%; DeepSeek-V3.2: 80.16%). No statistically significant difference was observed among the models in terms of overall performance (p=0.619). All models demonstrated significantly higher performance on knowledge-based questions than on case-based questions (p<0.05). Topic-based analysis showed numerically lower accuracy for implant-supported prosthodontic questions than for fixed and removable prosthodontic questions. However, this finding should be interpreted cautiously because the implant-supported subgroup included only four questions. No significant relationship was found between examination years and the accuracy rates of the models (chi²=7.1; p=0.62). Conclusion: The examined Artificial Intelligence models demonstrated a high level of accuracy in answering prosthodontic specialty questions. These systems have the potential to serve as supportive tools in dental education and examination preparation processes. However, owing to their limitations in questions requiring clinical reasoning, particularly case-based questions, their use should be approached cautiously and critically.
