Researchers D. Coraci, Gianluca Regazzo, G. Santilli and colleagues published a systematic mapping review of generative AI in rehabilitation in Applied Sciences in 2026. They searched PubMed and Scopus for work on generative AI, large language models and rehabilitation, then screened records under PRISMA-ScR guidance. Of 1,313 unique records, 233 met the eligibility criteria; the publications appeared from 2023 to 2026.

The review coded 106/233 records (45%) as clinical or workflow support for rehabilitation, 64/233 (27%) as patient-level intervention or functional assistance, 56/233 (24%) as education for rehabilitation professionals, and 45/233 (19%) as education for patients or caregivers. Twenty-eight records (12%) covered more than one application. Model or output evaluation and technical development predominated, appearing in 155/233 records (67%); 13/233 (6%) were interventional or feasibility studies.

The authors conclude that rehabilitation GenAI research most often addresses clinician-facing support rather than direct patient intervention. Evidence is dominated by technical, simulated and observational evaluations, with limited clinical experimentation. They call for prospective patient-level evaluation, transparent model reporting, safety and human oversight.

Researchers D. Coraci, Gianluca Regazzo, G. Santilli and colleagues published a systematic mapping review of generative AI in rehabilitation in Applied Sciences in 2026. They searched PubMed and Scopus for work on generative AI, large language models and rehabilitation, then screened records under PRISMA-ScR guidance. Of 1,313 unique records, 233 met the eligibility criteria; the publications appeared from 2023 to 2026.

The review coded 106/233 records (45%) as clinical or workflow support for rehabilitation, 64/233 (27%) as patient-level intervention or functional assistance, 56/233 (24%) as education for rehabilitation professionals, and 45/233 (19%) as education for patients or caregivers. Twenty-eight records (12%) covered more than one application. Model or output evaluation and technical development predominated, appearing in 155/233 records (67%); 13/233 (6%) were interventional or feasibility studies.

The authors conclude that rehabilitation GenAI research most often addresses clinician-facing support rather than direct patient intervention. Evidence is dominated by technical, simulated and observational evaluations, with limited clinical experimentation. They call for prospective patient-level evaluation, transparent model reporting, safety and human oversight.