Performance and potential of large language models in restorative dentistry, endodontic diagnosis, and education: A systematic review

Introduction: Large language models (LLMs) such as ChatGPT, Gemini, and Claude have emerged as promising tools in health care, with potential applications in dental diagnosis and education. However, their performance in specialized endodontic contexts remains unclear. This systematic review aimed to analyze the diagnostic and educational performance of LLMs in endodontics and restorative dentistry, focusing on accuracy, consistency, and clinical applicability. Materials and Methods: A comprehensive search of PubMed, Scopus, Google Scholar, and ProQuest was conducted until April 2025 using structured Boolean strategies. Peer-reviewed studies published from 2023 to 2025 evaluating LLM diagnostic accuracy, educational efficacy, or reasoning consistency in endodontics and restorative dentistry were included in the study. Studies focusing solely on image-based AI or lacking empirical evaluation were excluded from the study. Risk of bias (ROB) was assessed using the modified Cochrane ROB 2.0 tool. Due to methodological heterogeneity, narrative synthesis was performed. The protocol was registered in PROSPERO (CRD420251045162). Results: Eleven studies met the inclusion criteria from 20 identified records. ChatGPT-4o demonstrated the highest diagnostic accuracy (72%–99%), particularly for clinical endodontic diagnosis. ChatGPT-4 showed consistent reliability in glossary tasks and multiple-choice questions. ChatGPT-3.5, Gemini, and Bard exhibited inferior and inconsistent performance. Evaluation methods varied, including binary assessments, Likert scales, and expert consensus. Common limitations included small sample sizes, single-institution studies, lack of multimodal inputs, and inconsistent prompting strategies. Conclusions: Advanced LLMs, particularly ChatGPT-4o and ChatGPT-4, demonstrate considerable promise in restorative dentistry, endodontic diagnosis, and education. However, clinical integration requires human supervision, standardized evaluation methods, and validation through multimodal data before widespread implementation in dental practice.