Abstract
BACKGROUND: Due to their ability to mimic human responses, anthropomorphic entities such as ChatGPT have a higher likelihood of gaining people's trust. This study aimed to evaluate the quality of information generated by ChatGPT-4, as an artificial intelligence (AI) chatbot, on periodontal disease (PD) using the DISCERN instrument. METHODS: Using Google Bard, the topics related to PD that had the highest search volume according to Google Trends were identified. An interactive dialogue was created by placing the topics in the standard question pattern. As a patient with PD, detailed information was requested from ChatGPT-4 regarding the relevant topics. The 'regenerate response' feature was not employed, and the initial response generated by ChatGPT-4 was carefully considered for each topic as new prompts in the form of questions were entered. The response to each question was independently assessed and rated by two experienced raters using the DISCERN instrument. RESULTS: Based on the total DISCERN scores, the qualities of the responses generated by ChatGPT-4 were 'good', except for the two responses that rater-2 scored as 'fair'. It was also observed that the 'treatment choices' section of both raters had significantly fewer scores than the other sections. In both weighted kappa and Krippendorff alpha measures, the strength of agreement varied from 'substantial' to 'almost-perfect', and the correlation between values was statistically significant. CONCLUSION: Despite some limitations in providing complete treatment choice information according to the DISCERN instrument, it is considered valuable for PD patients seeking information, as it consistently offered accurate guidance in the majority of responses.