Abstract
Background: For documentation, patient education, clinical reasoning support, and professional education, large language models (LLMs) such as ChatGPT, Gemini, and Claude are widely suggested. Physiotherapy-specific evidence is becoming more prevalent yet dispersed. The aim is to map the nature, extent, and gaps of the literature on LLM/generative-AI applications in physiotherapy clinical and clinical-adjacent workflows.
Methodology: Indexed literature (January 2022-June 2025, with practical yield from late 2022) was searched and screened; a scoping review was carried out in accordance with JBI and reported in accordance with PRISMA-ScR. Seven sources were charted and mapped by application domain.
Results: There are three main application clusters: (1) text generation, patient education and documentation, where models can create, simplify, and translate materials; (2) professional education, which includes clinical cases generated by LLM for students; and (3) clinical-reasoning benchmarking, which compares models to clinicians in particular rehabilitation domains. Although task-dependent and prone to errors, accuracy is promising, and there is a dearth of deployed clinical evaluations as opposed to instructional or benchmark use. Non-English ethical, governance, privacy, and equity issues are frequently brought up yet remain unresolved.
Conclusion: Supporting documentation and creating patient education materials are the most evidence-based applications; clinical reliance necessitates local validation and governance. The main deficiency is in the deployment of multilingual, physiotherapy-specific evidence.