Can artificial intelligence (AI)-based software programs generate accurate clinical dictation?

Clinical dictation in dental practice is time-consuming and prone to quality variability. Although generative artificial intelligence (AI) systems such as Microsoft Copilot, OpenAI ChatGPT, Google Gemini, and OpenEvidence offer potential alternatives for automating and enhancing this process, their ability to produce clinically reliable and contextually appropriate dictations has not been evaluated.

The purpose of this evaluation study was to assess the performance of Copilot, 2 versions of ChatGPT (v4 and 5), Gemini (v2.5), and OpenEvidence in generating accurate, complete, and clinically relevant dental dictations based on varying prompt structures.

Two different representative clinical scenarios were generated. One described the placement of a single mandibular implant with the dictation focused on the surgical steps, while the second described a prosthodontic procedure and the steps required to place a crown on a single maxillary implant. Each scenario was entered into Copilot, 2 different versions of ChatGPT, Gemini, and OpenEvidence using a structured prompt format (SP). All inputs were conducted in a clean session. All responses from Copilot, both included versions of ChatGPT, Gemini, and OpenEvidence software programs, were compared for both scenarios. Minimum specific prompts were also generated for the same clinical scenarios and tested to determine the least amount of clinician input required to produce acceptable dictations. Overall responses were compared with the more detailed structured prompts to assess differences in output quality.

Across both surgical and restorative procedures, all 5 AI software programs (Copilot, ChatGPT-4, ChatGPT-5, Gemini, and OpenEvidence) produced clinically accurate dictations when using the SP format. For the surgical procedure, core steps-including preoperative assessment, anesthesia, incision design, osteotomy preparation, implant placement, closure, and postoperative instructions-were consistent. For the restorative procedure, all software programs documented atraumatic healing abutment removal, healthy peri-implant mucosa, correct crown fit, occlusion, shade matching, and standardized tightening (35 Ncm). No substantial differences in procedural accuracy were identified; discrepancies were primarily stylistic or related to the level of descriptive detail. With minimum specific prompts, the dictations were shorter and less descriptive but still captured the essential procedural steps for both clinical scenarios and were considered clinically acceptable in all included AI software programs.

Microsoft Copilot, both versions of OpenAI ChatGPT (GPT-4 and GPT-5), Google Gemini, and OpenEvidence can effectively assist clinicians in generating clinical dental procedure dictations, particularly when guided with precise and structured prompts. Minimum specific prompts produced shorter but clinically acceptable outputs that may be particularly useful in busy clinical settings. While not a substitute for clinician oversight, the AI technology promises to improve efficiency and documentation consistency in dental practice.