Language-Dependent Accuracy of Large Language Models in Dental Avulsion Scenarios: A Comparative Study

Purpose: To evaluate whether the accuracy of responses generated by large language model–based systems to dental avulsion–related questions differs according to language.
Methods: Dental avulsion questions based on International Association of Dental Traumatology guidelines were administered to ChatGPT (version 5.1) and Meta AI in Turkish and English. Queries were submitted using two independent user accounts over seven consecutive days. All responses were restricted to a true–false format and assessed for guideline compliance. Paired comparisons were performed using the McNemar test.
Results: A total of 2,800 responses were analyzed. ChatGPT showed comparable accuracy in Turkish and English. In contrast, Meta AI demonstrated higher accuracy in English than in Turkish. When compared under identical conditions, ChatGPT exhibited higher overall accuracy, primarily due to differences in Turkish responses.
Conclusion: Language may influence the performance of large language model–based systems in dental avulsion scenarios in a model-dependent manner, underscoring the need to consider linguistic factors when evaluating AI-generated clinical information.

Keywords: Artificial ıntelligence, dental trauma, endodontics, tooth avulsion