Generative AI could reshape physical therapy training, narrative review finds

Generative artificial intelligence has swept through higher education since ChatGPT appeared in November 2022, and physical therapy programs are now wrestling with what that means for training the next generation of clinicians. A narrative review published in the Bulletin of Faculty of Physical Therapy examines how generative AI, often abbreviated GenAI, might strengthen evidence-informed practice in Doctor of Physical Therapy curricula, with particular attention to first-year students. The review synthesizes literature from 2020 through 2025 and draws on fifteen peer-reviewed studies to map where AI literacy stands, what benefits educators report, and which obstacles remain unresolved.

The authors, Maritza Dominguez and Leah Lowe, searched Academic Search Complete, CINAHL Complete, MEDLINE Complete on the EBSCO platform, and ProQuest, including its Health and Medical Collection and Nursing and Allied Health sub-databases. They supplemented these traditional databases with AI-powered search engines: the paid version of Perplexity-Pro and the free versions of Elicit and Consensus. Search terms combined keywords for artificial intelligence and generative AI with phrases targeting physical therapy and physiotherapy education. The team restricted the search to peer-reviewed, full-text articles published within the past five years, and included studies addressing AI literacy and comfort among physical therapy professionals, educators, or students, as well as work covering benefits and challenges of AI in physical therapy education and practice.

The screening process was deliberately structured. After removing duplicates, two reviewers independently screened titles and abstracts at each stage, with a third reviewer available to resolve disagreements that consensus could not settle. No sorting software was used. From 28 identified responses, thirteen were excluded for failing inclusion criteria or falling outside scope, leaving fifteen articles. The final pool contained five narrative or review-based articles, three cross-sectional studies, two qualitative studies, one randomized controlled trial, one case report, one pre-post survey, one educational or descriptive study, and one mixed-methods study. One study of occupational therapy students was included because of the close overlap between occupational therapy and physical therapy education.

What did the evidence show about practicing clinicians? Across four studies in that category, a consistent theme emerged: licensed physical therapists recognize the potential benefits of AI in practice while simultaneously reporting limited formal education or structured training in its use. Aldhahi and colleagues reported that 62 percent of licensed physical therapists lacked formal AI education, a quantitative finding the review explicitly separates from qualitative explorations of practitioner attitudes. Most respondents in these studies acquired AI knowledge through self-directed learning or informal sources such as social media. Both Bills and Basheer and Abuzaid and colleagues identified younger professionals and those with higher AI literacy as more receptive to AI integration, suggesting that comfort with AI correlates with familiarity with digital tools.

Attitudes among clinicians were broadly optimistic but not uniform. Aldhahi and colleagues and Abuzaid and colleagues emphasized survey-based quantitative findings on percentages of clinicians trained, interested, or confident in AI use, and both reported optimism about AI’s role in enhancing productivity and patient outcomes. Shwali and colleagues took a qualitative approach instead, offering deeper insight into practitioners’ values and concerns, including ethical implications, privacy considerations, and preservation of the therapist-patient relationship. That qualitative work also highlighted uncertainty about workforce implications if AI displaces aspects of clinical judgment or patient interaction. Collectively, the review characterizes the profession as broadly optimistic toward AI, tempered by critical gaps in formal education, a need for guidelines, and ongoing concerns about ethics and professional identity.

In physical therapy education, the picture was similar but with more direct evidence of what works. The nine studies in this category consistently highlighted low baseline AI literacy and confidence among students and educators, and pointed to direct, hands-on exposure to AI tools as the most effective strategy for building competence and comfort. Severin and Gagnon found that 64 percent of academic physical therapy respondents reported moderate to high comfort with generative AI after direct exposure. Ferrer-Peña’s randomized controlled trial demonstrated a 23 percent improvement in digital competency scores among students engaged with AI-driven learning modules. Mansour and Wong, Reoli and colleagues, Zhang and Rapport, Thompson and Fecher, and Ferrer-Peña and colleagues all showed that purposeful integration of AI tools into coursework, whether through ChatGPT, problem-solving activities, research simulations, or clinical reasoning tasks, improved comfort, digital literacy, and application in practice. Lindbäck and colleagues found that physiotherapy students saw generative AI as a valuable study partner while raising concerns about reliability and over-reliance.

Method and emphasis varied across these education studies. Mansour and Wong and Ferrer-Peña and colleagues used experimental designs measuring concrete gains in comfort and digital skills, whereas Lindbäck and colleagues and Thompson and Fecher relied on qualitative findings to explore perceptions and critical thinking processes. Zhang and Rapport and Reoli and colleagues provided evidence that scaffolded integration into core courses such as evidence-based practice and clinical reasoning can yield measurable academic outcomes and higher self-efficacy. Ferrer-Peña and colleagues cautioned that student engagement with AI remains inconsistent, pointing to adoption challenges. On the faculty side, Severin and Gagnon emphasized that educators’ limited knowledge represents a major barrier to meaningful curricular adoption, a paradox in which faculty may teach content they do not feel prepared to use themselves. Rahman and colleagues underscored the limitations of AI in scholarship, stressing fabricated references and weak synthesis, which reinforces the call for critical use and continued human oversight.

The review also documents structural tools that educators are using to manage AI use. Many instructors have implemented an AI light system that defines when AI is prohibited, marked red; allowed with restrictions, marked yellow; or encouraged, marked green. More recently, Perkins and colleagues introduced and expanded the AI Assessment Scale, which refines these categories into five levels: No AI, where assessment is completed entirely without AI assistance in a controlled environment; AI-assisted idea generation and structuring, where AI may support brainstorming, outlining, and initial research; AI-assisted editing, where AI improves clarity or quality of student-created work but creates no new content; AI task completion with human evaluation, where AI completes certain task elements while students provide commentary and critical engagement with the output; and full AI, where AI acts as a co-pilot in a collaborative approach. This tiered system gives students and educators transparent, course-specific guidance for appropriate generative AI use.

Challenges loom as large as the opportunities. Two systematic reviews cited in the article, by Mohapatra and colleagues and Hafeez and colleagues, converge on data privacy and security as primary challenges given the sensitive nature of health data required by AI systems. Bias in datasets and algorithms is equally central: AI can perpetuate or amplify existing health disparities if training datasets lack diversity or are skewed by confounding variables, and models trained on non-diverse datasets may misclassify patients from underrepresented backgrounds as low risk, leading to inequitable care and potentially delayed interventions. Hafeez and colleagues quantified how frequently ethical issues appear in 60 percent of reviewed studies, professional issues in 50 percent, and technical issues in 75 percent, including difficulties adapting current workflows to integrate AI tools. Both reviews also flagged implementation costs spanning technology, ongoing maintenance, and training, along with resistance to change among faculty and practitioners driven by knowledge gaps, job security concerns, and ambiguity about accountability when AI-driven decisions are involved.

The authors conclude that purposeful, scaffolded integration of generative AI into Doctor of Physical Therapy curricula, guided by transparent policies and ethical frameworks, can enhance research and practice readiness while safeguarding critical thinking and patient-centered care. They point to the American Physical Therapy Association’s Vision for Excellence in Education and the World Health Organization’s 2021 guidance on ethics and governance of AI for health, which lays out six principles covering autonomy, human well-being and safety, transparency and explainability, responsibility and accountability, inclusiveness and equity, and responsive and sustainable AI. Ongoing educator development, clear guidelines, and adaptive multi-institutional strategies are described as vital to bridging gaps in AI literacy. The review carries important limitations: the included studies varied substantially in sample size and design, from small qualitative studies and case reports to larger cross-sectional surveys and review-level analyses, and the literature was concentrated primarily in Western and Gulf country settings, limiting transferability to other regions. The authors call for future studies from African, South Asian, and additional Arab physical therapy education contexts, and note that as a narrative review, the synthesis supplements gaps in systematic evidence rather than replacing it.

Subject of Research: Integration of generative artificial intelligence into Doctor of Physical Therapy education and evidence-informed practice

Article Title: From artificial to authentic: a narrative review of generative artificial intelligence’s potential to enhance evidence-informed practice in first-year physical therapy students

Article References: Dominguez, M., & Lowe, L. (2026). From artificial to authentic: a narrative review of generative artificial intelligence’s potential to enhance evidence-informed practice in first-year physical therapy students. Bulletin of Faculty of Physical Therapy, 31(1), Article 28. https://doi.org/10.1186/s43161-026-00361-y

Image Credits: AI Generated

DOI: 10.1186/s43161-026-00361-y

Keywords: generative AI, physical therapy education, AI literacy, evidence-informed practice, Doctor of Physical Therapy, ChatGPT, ethical frameworks, algorithmic bias, curriculum integration, narrative review, health professions education, AI Assessment Scale