Abstract
Background: Physiotherapy education faces mounting pressure to prepare graduates for an AI-augmented clinical environment. Despite rapid acceleration in AI publications since early 2024, the systematic evidence base for AI applications specifically within physiotherapy education remains sparse, methodologically immature, and critically lagging behind medicine and dentistry. No comprehensive synthesis incorporating the most recent empirical studies including the first randomised controlled trials (RCTs) in this domain existed prior to this review. Objectives: To (1) identify and critically evaluate all study designs examining AI integration in physiotherapy education up to July 2025; (2) characterise AI modalities and their educational applications; (3) synthesise learning outcomes including clinical reasoning, knowledge retention, practical competence, and student perceptions; and (4) identify methodological gaps and priorities for future research.
Methods: A PRISMA 2020 compliant systematic search across PubMed, CINAHL, Scopus, Web of Science, ERIC, and JMIR databases used a PICOS-informed strategy with an upper cutoff of July 2025. Inclusion required studies to explicitly address AI application in physiotherapy or allied health education. Quality was assessed using design-appropriate tools (RoB 2, Newcastle-Ottawa Scale, JBI checklist, AMSTAR-2). Narrative synthesis was performed due to methodological heterogeneity.
Results: Eight studies met full-inclusion criteria: one prospective parallel-group RCT, one pilot RCT, one qualitative focus-group study, one evaluative cross-sectional study, one mixed-methods study, one systematic review, one narrative review, and one bibliometric analysis. AI modalities included large language models (LLMs; ChatGPT/GPT-4), generative AI, VR with AI feedback, and ML-based wearable sensors. Evidence supported AI as beneficial for theoretical knowledge acquisition, clinical reasoning preparation, and student engagement. However, evidence for practical competency transfer and real-world performance improvement remained insufficient, with no Mini-CEX advantage demonstrated in the only RCT.
Conclusions: AI holds genuine transformative potential for physiotherapy education, but the evidence base remains nascent, heterogeneous, and weighted towards cognitive rather than practical learning outcomes. The first RCTs represent critical progress but expose deep implementation barriers including low student AI literacy and engagement. LLMs show the most immediate educational utility, while concerns regarding accuracy, over-reliance, and critical thinking erosion demand structured faculty oversight and explicit AI literacy curricula. Equity considerations and the digital divide require urgent research attention.