Accuracy and completeness of ChatGPT-4o in the management of non-carious cervical lesions

This study aimed to assess the accuracy and completeness of Chat Generative Pre-trained Transformer (ChatGPT) in controlling noncarious cervical lesions (NCCLs) associated with gingival recession. Twelve case scenarios were created using the clinical and radiographic examination data of young adults who came to the periodontology clinic for the first time because of dentin hypersensitivity or esthetic concerns. Gingival recession and NCCL classifications, as well as two reviews, were taken into consideration when developing case scenarios. ChatGPT was asked to offer answers in the fields of diagnosis, clinical management, and surgical management, as well as all bibliographic references used to develop those replies. All replies received during this procedure were examined by four independent reviewers. The reviewers used a 6-point Likert scale to assess the accuracy of each domain response, a 3-point Likert scale for completeness, and the modified global quality scale to assess existing references. The agreement among the reviewers ranged from 0.629 to 1.000. A statistically significant association was found between the accuracy and completeness ratings in the diagnosis ( p ​= ​0.005) and clinical management ( p ​= ​0.010) domains. However, no statistically significant association was discovered between surgical management accuracy and completeness. The majority of the references were of a moderate level, with three showing good quality. This study found that a full solution that addressed all parts of the cases was rarely possible. At this point, using ChatGPT as a supplemental tool for clinical decision-making in complex NCCL case scenarios does not appear practical.