Evaluating the Accuracy, Reliability, Consistency, and Readability of Different Large Language Models in Restorative...

This study aimed to evaluate the reliability, consistency, and readability of responses provided by various artificial intelligence (AI) programs to questions related to Restorative Dentistry.

Forty-five knowledge-based information and 20 questions (10 patient-related and 10 dentistry-specific) were posed to ChatGPT-3.5, ChatGPT-4, ChatGPT-4o, Chatsonic, Copilot, and Gemini Advanced chatbots. The DISCERN questionnaire was used to assess the reliability; Flesch Reading Ease and Flesch-Kincaid Grade Level scores were utilized to evaluate readability. Accuracy and consistency were determined based on the chatbots' responses to the knowledge-based questions.

ChatGPT-4, ChatGPT-4o, Chatsonic, and Copilot demonstrated "good" reliability, while ChatGPT-3.5 and Gemini Advanced showed "fair" reliability. Chatsonic exhibited the highest "DISCERN total score" for patient-related questions, while ChatGPT-4o performed best for dentistry-specific questions. No significant differences were found in readability among the chatbots (p > 0.05). ChatGPT-4o showed the highest accuracy (93.3%) for knowledge-based questions, while Copilot had the lowest (68.9%). ChatGPT-4 demonstrated the highest consistency between repetitions.

Performance of AIs varied in terms of accuracy, reliability, consistency, and readability when responding to Restorative Dentistry questions. ChatGPT-4o and Chatsonic showed promising results for academic and patient education applications. However, the readability of responses was generally above recommended levels for patient education materials.

The utilization of AI has an increasing impact on various aspects of dentistry. Moreover, if the responses to patient-related and dentistry-specific questions in restorative dentistry prove to be reliable and comprehensible, this may yield promising outcomes for the future.