Quality of Artificial Intelligence-Generated Versus Conventional Informed Consent Documents in Shoulder Surgery: A Comparative Study

Background/Objectives: Informed consent remains a cornerstone in orthopaedic surgery, with important ethical and legal implications. The complexity of shoulder procedures and patient expectations require clear and accurate information. Artificial intelligence (AI) and large language models (LLMs) offer new possibilities for generating medical texts in shoulder surgery. This study aimed to perform a comparative evaluation of AI-generated and formal informed consent documents. Methods: Consent documents for arthroscopic rotator cuff repair, stabilization, capsular release, and reverse shoulder arthroplasty were evaluated. AI-generated documents were created using ChatGPT. Conventional documents were obtained from the Orthoconsent website supported by the British Orthopaedic Association. Twenty independent shoulder surgeons assessed the documents blindly, using Likert scales for four clinically relevant domains. Descriptive statistics, paired Wilcoxon signed-rank tests with Holm adjustment, and an exploratory linear mixed-effects analysis were performed. Results: Consent documents of both groups achieved high scores across all domains. No statistically significant difference was identified for completeness (p = 0.689). AI-generated documents scored significantly higher for clinical accuracy (p = 0.010) and suitability for clinical use (p = 0.010), whereas conventional documents scored significantly higher for risk communication (p = 0.010). Procedure-specific analysis demonstrated higher scores for conventional documents in arthroscopic rotator cuff repair (p = 0.044) and capsular release (p = 0.024), and higher scores for AI-generated documents in reverse shoulder arthroplasty (p = 0.010), with no significant difference in shoulder stabilization (p = 0.689). Mean differences were generally modest, with no consistent direction of superiority. Conclusions: Both AI-generated and conventional informed consent documents received generally high expert ratings, with modest domain- and procedure-specific differences. LLMs may be valuable tools in clinical practice, but potential risks remain with unsupervised use; therefore, they should complement rather than replace surgeons’ clinical judgment and responsibility.