Review of 9 studies finds AI chatbots can match experts on rehab plans, but they hallucinate

In musculoskeletal and cardiovascular rehabilitation, the chatbots produced plans aligned with clinical guidelines. The included studies also reported hallucinations, meaning the models generated incorrect content, which the authors say makes clinician supervision necessary.

The reviewers searched PubMed, Web of Science and Scopus up to February 2026. They conclude the tools show promise for adaptable, personalised planning, but say higher-quality studies are needed to test real-world effectiveness and define how to implement them.

The review asked whether large language models can generate appropriate rehabilitation plans that comply with expert clinicians or clinical practice guidelines. It compared them with expert opinion or other AI systems, and excluded studies of non-generative AI.

Across nine included studies, agreement with experts varied. The 98.75% figure came after an expert advisory board reviewed the AI-generated statements, not from unreviewed output alone.

Alignment with guidelines was reported in musculoskeletal and cardiovascular rehabilitation. Hallucinations appeared in the included studies, which the authors say makes supervision necessary.

The authors call the capabilities promising and say they may offer adaptability and personalisation. They add that high-quality studies are needed to validate real-world effectiveness and define implementation strategies.