Evaluation of artificial intelligence–based software programs applied to the radiographic and photographic identification of dental implant systems

Abstract

Purpose: The purpose of this evaluation study was to assess the performance of ChatGPT-5, Google Gemini (v2.5), and Microsoft Copilot in identifying five demonstration dental implant systems and four implants from real clinical cases based on clinical and radiographic images. Materials and Methods: Clinical photographs of the demonstration implant platforms and corresponding radiographs of NobelActive (NobelActive; Nobel Biocare), NobelParallel Conical Connection (NobelParallel CC; Nobel Biocare), NobelReplace Conical Connection (NobelReplace CC; Nobel Biocare), Straumann Bone Level Conical (Straumann BLC; Institut Straumann), and Straumann BLX (Straumann BLX; Institut Straumann), were entered into ChatGPT-5, Google Gemini (v2.5), and Microsoft Copilot with one standardized prompt. In addition, four previously placed implants from real patient cases (NobelActive, NobelReplace CC, Straumann Bone Level Tapered [BLT], and Straumann BLX) were analyzed using the same software programs and standardized prompt. All inputs were conducted in a clean session. All responses from ChatGPT-5, Gemini, and Copilot software programs were compared for accuracy and descriptive completeness. Results: ChatGPT-5 correctly identified two of the five demonstration implant systems, but did not definitively identify any of the four real clinical cases. Google Gemini failed to correctly identify any implant system in either the demonstration or clinical cases. Microsoft Copilot did not correctly identify the demonstration implants but correctly identified one of the four real clinical cases (Straumann BLX). Diagnostic overlap among tapered internal-connection systems was a common source of misclassification. Conclusions: None of the evaluated artificial intelligence (AI) software programs consistently identified dental implant systems across both demonstration and real clinical cases. Although ChatGPT-5 showed partial diagnostic capability in demonstration implants and Microsoft Copilot correctly identified one real clinical case, overall system-level identification remained unreliable.