The APA's Statement on AI in Physiotherapy: What It Means for Your Practice

A note before I start. I'm a member of the APA's inaugural AI Advisory Group, which helped shape this statement. What follows is my own commentary, not the APA's position or the group's.

The statement is short, and every physiotherapist should read it. Its central position fits in one sentence: AI must not replace professional judgement, clinical reasoning, diagnosis or advice, and a registered physiotherapist oversees its use and stays responsible for the care.

What the statement is, and isn't

It's a statement of principles, not a how-to guide. It doesn't tell you which tools to buy. That's deliberate: tools change every few months, and principles last longer.

It aligns with Ahpra's guidance on AI, Australia's AI Ethics Principles and the World Health Organization's six ethical principles for AI in health. Where it differs from Ahpra's guidance, Ahpra's takes precedence. The APA says it will review the statement as regulation and the technology mature.

The seven guiding principles, in plain English

Accountability. Someone identifiable is responsible at every stage, and a human is always in the loop. In your clinic, that's you.

Understanding AI. Keep learning about AI and use it ethically. You don't need to be an engineer, but you do need to know what your tools can and can't do.

Transparency. Be open about AI use, and don't present AI-generated content as entirely your own work.

Informed consent. Tell patients when AI is part of their care, explain what it is and its limits, let them agree freely, and document it.

Ethical and legal obligations. Protect privacy and confidentiality under the Privacy Act and the National Law, and follow your organisation's rules.

Inclusiveness and equity. Consider whether data or algorithms could be biased, particularly for Aboriginal and Torres Strait Islander peoples, and use tools only where they're appropriate.

Reliability and safety. Use validated tools for their intended purpose. If outputs become unreliable, stop relying on them. When in doubt, your clinical judgement wins.

What it changes on Monday morning

For most physiotherapists, the statement formalises good practice rather than inventing new rules. Four things deserve attention.

Consent becomes a documented step. Build it into intake, so it doesn't depend on memory. I cover the practicalities in what Ahpra expects when you use AI.

Transparency reaches beyond notes. Reports, GP letters, patient handouts and even your clinic's social posts can be AI-assisted. You can use AI to draft. You can't pretend it didn't help.

Intended use is a real limit. A general-purpose chatbot isn't a validated clinical tool. Using one to interpret imaging or reach a diagnosis sits outside what it was built and tested for. And if a tool starts getting things wrong, stop relying on it and tell the vendor.

Practice owners carry extra weight. Know which AI tools your team actually uses, including free ones on personal phones, and write a short policy: approved tools, approved uses, who reviews what and how consent is recorded.

What I'd like to see next

Principles are the right foundation. Clinicians will now want practical help: worked consent conversations, guidance on AI-assisted reports for funders and insurers, and support for assessing tools. Clinicians, owners, educators and builders can all contribute to that, not just the APA.

The principle I think about most is understanding AI. It's hard to judge a tool's limits if you've never seen one built. That's why I started CliniAcademy, where clinicians and practice teams build working prototypes and work through the privacy, Ahpra and safety questions that come before real-world use.

Why this matters

Patients trust physiotherapists, and AI used carelessly could spend that trust quickly. A clear statement that AI supports judgement protects it. As someone who builds AI for allied health, I see the statement as the bar the tools should meet, not a barrier to them.

---

For a real case where these principles were tested, see lessons from an NDIS tribunal decision on AI-written reports.