Int Dent J. 2026 May 20;76(4):109636. doi: 10.1016/j.identj.2026.109636. Online ahead of print.
ABSTRACT
PURPOSE: The integration of artificial intelligence (AI) into prosthodontics represents a paradigm shift in the design and fabrication of removable partial dentures (RPDs). This study evaluates the current AI technologies, including large language models such as ChatGPT, Copilot, Gemini, and DeepSeek, for designing RPDs.
METHODS: Standardized prompts were submitted to four AI-assisted systems (ChatGPT, Copilot, Gemini, and DeepSeek) to generate RPD designs for 25 partially edentulous clinical scenarios based on Kennedy classifications. The outputs were compared with reference models (AiDENTAL), a validated retention prediction model, and evaluations by six blinded experts. Statistical comparisons were performed using a mixed-effects ordinal regression model (p < 0.05).
RESULTS: Compared with the reference standards, Copilot and Gemini produced the most consistent and accurate RPD designs, whereas DeepSeek showed the greatest deviation, despite achieving the highest predicted retention scores. Expert evaluations confirmed that AiDENTAL achieved the highest overall design quality, followed by Copilot, whereas ChatGPT and DeepSeek demonstrated lower and more variable performances.
CONCLUSIONS: The design of RPDs using different large language models, including ChatGPT, Copilot, Gemini, and DeepSeek, resulted in notable variations in the design output. AiDENTAL achieved the highest overall RPD design quality, whereas Copilot and Gemini produced the most consistent and accurate designs across all the evaluated criteria.
PMID:42160812 | DOI:10.1016/j.identj.2026.109636
