Artificial intelligence tools such as ChatGPT and Gemini are increasingly used by trainee endoscopists for rapid access to guideline-based recommendations. During colonoscopy, endoscopists often encounter polyps requiring endoscopic resection. Rapid support from AI tools can be obtained either in voice mode (direct verbal interaction with vocal responses) or via audio-to-text (spoken input transcribed to text with written AI output), offering different levels of hands-free support.

The aim of our study is to evaluate the concordance of two commonly used AI tools with ESGE 2024 polypectomy recommendations, comparing audio-to-text and voice mode performance.

Ten clinical scenarios covering different polyp types were submitted to ChatGPT and Gemini in both modalities. Responses were compared to ESGE 2024 colorectal polypectomy and endoscopic mucosal resection guidelines. For discordant responses, follow-up prompts (“Are you sure?”) assessed each model's ability to self-correct. Descriptive analyses were performed using Jamovi.

For ChatGPT, audio-to-text responses achieved a concordance of 100% (10/10) with ESGE guidelines, and minor transcription inaccuracies (40%) did not affect the correctness of the answers. In voice mode, initial concordance was 50% (5/10), and follow-up prompts corrected 2 of the discordant answers, yielding a final concordance of 70%.

For Gemini, audio-to-text responses showed 60% concordance, 20% mixed answers (two techniques suggested, only one matching the guidelines), and 20% discordance. Gemini’s live voice mode showed a concordance of 70%, which remained unchanged after prompting.

AI tools can assist in selecting appropriate polypectomy techniques according to ESGE recommendations; however, their performance varies significantly between models and usage modes. ChatGPT provided consistently accurate responses via audio-to-text, offering an almost hands-free support tool during endoscopy.