Comparative Assessment of Performance of Domain-Specific and Primed Versus Non-Primed Artificial Intelligence Chatbots for Clinical Decision-Making in Injuries to the Primary Dentition: An Exploratory Pilot Study.

BACKGROUND/AIMS: To compare the performance of a domain-specific dental trauma chatbot with general-purpose artificial intelligence chatbots under primed and non-primed conditions for clinical decision-making in injuries to the primary dentition.

METHODS

Fifteen standardized clinical case scenarios were developed for evaluating three chatbot systems: a domain-specific model (Dental Trauma Evo) and two general-purpose models (ChatGPT 5.2 and Perplexity Pro). General-purpose chatbots were evaluated under primed and non-primed conditions, where priming involved providing a summarized guideline document prior to scenario input. Chatbots were required to generate responses addressing diagnosis, immediate management, follow-up intervals, and radiographic recommendations. Two Pediatric dentists independently evaluated responses using a binary scoring system based on International Association of Dental Traumatology (IADT) guidelines. Statistical comparisons were performed using McNemar's test and agreement analysis with kappa statistics.

RESULTS

All chatbots demonstrated complete diagnostic accuracy across scenarios. The domain-specific chatbot achieved 100% accuracy in immediate management decisions, while minor inaccuracies were observed among general-purpose models. The most substantial differences were observed in follow-up interval and radiographic recommendations. Priming significantly improved the performance of general-purpose chatbots, with ChatGPT and Perplexity demonstrating marked gains in follow-up interval accuracy and radiographic recommendations. In these domains, primed general-purpose models demonstrated greater agreement with IADT guideline recommendations than the domain-specific chatbot.

CONCLUSIONS

Guideline-based priming was associated with improved performance of general-purpose AI chatbots in clinical decision-making for injuries to the primary dentition, particularly for follow-up scheduling and radiographic recommendations. While domain-specific models provide reliable management guidance, contextual exposure to clinical guidelines enables general large language models to demonstrate a high level of agreement with established recommendations in certain aspects of trauma care in primary dentition. These findings highlight the potential role of prompt-guided AI systems as supportive tools in dental traumatology decision-making.