AI-generated versus standard informed consent forms in otolaryngology: a blinded multirater pilot feasibility study.

Background

Informed consent documents often exceed recommended readability levels, potentially compromising patients' decision-making. AIMS/

Objectives

This pilot feasibility study evaluated two large language models (LLMs), ChatGPT-5.2 and Gemini 3 Pro, as tools for generating more comprehensible otolaryngology surgical consent forms, and estimated effect-size parameters for a future trial.

Materials And Methods

In a prospective, double-blind design, 123 Turkish consent documents (41 procedures × 3 sources: official forms, ChatGPT-5.2 and Gemini 3 Pro) were evaluated. Three independent otolaryngologists rated medical quality on five 5-point criteria; five blinded lay volunteers rated comprehensibility. Two validated Turkish readability indices (Ateşman; Bezirci-Yılmaz) provided objective measurement. Comparisons used linear mixed-effects models.

Results

Official forms showed the highest expert-rated quality (4.62 ± 0.35) versus ChatGPT5.2 (3.85 ± 0.48) and Gemini 3 Pro (3.99 ± 0.44; p < 0.001). Gemini forms achieved the highest comprehensibility (4.33 ± 0.42 vs 3.18 ± 0.51 for official forms; p < 0.001). Objective readability mirrored this, with the Ateşman score correlating with comprehensibility (r = 0.81).

Conclusions

AND

Significance

Publicly available LLMs, particularly Gemini 3 Pro, can substantially improve consent-form comprehensibility at some cost to medical-accuracy and ethical-legal adequacy; these findings motivate evaluation of a hybrid AI-plus-expert workflow in a future confirmatory trial.