ChatGPT wrote confident, fluent copy for a physiotherapy clinic that contained three separate AHPRA violations: an implied patient testimonial, an unsubstantiated outcome claim, and a misleading qualification statement. Here's what went wrong and what it means for your content.

Key takeaways

The experiment setup

The brief was designed to be realistic. It's the kind of request a practice manager or small clinic owner might send on a Tuesday afternoon when the website desperately needs updating and there's no budget for a specialist.

The prompt: "Write a homepage for a physiotherapy clinic in Melbourne. The clinic offers sports injury treatment, post-surgical rehabilitation, and chronic pain management. Make it AHPRA compliant."

Specific. Reasonable. With an explicit compliance instruction.

ChatGPT returned a polished, well-structured homepage. Good headline. Clear service descriptions. A warm, approachable tone. It read like competent healthcare marketing copy from a working professional.

And it contained three AHPRA violations.

None of them were obvious. That's the problem. They were embedded in natural-sounding sentences that a practice manager with no compliance training would have published without a second thought. This is not a criticism of the clinic. It's a structural limitation of AI tools applied to regulated content.

Here's what they were.

Violation 1: The implied testimonial

The copy included this sentence in the "About our approach" section:

"Our patients consistently tell us they feel heard, supported, and significantly better after just a few sessions."

AHPRA's advertising guidelines prohibit testimonials. A testimonial is any statement, from a current or former patient, about the quality or outcome of a regulated health service. The prohibition applies whether the testimonial is direct (a named patient review) or indirect (a practitioner paraphrasing patient feedback).

"Our patients consistently tell us" is paraphrased patient feedback. It's an indirect testimonial. It also contains an outcome claim ("significantly better"), which is a separate violation in its own right.

The phrase reads like generic reassurance. It would not raise a red flag for someone unfamiliar with the guidelines. That's precisely what makes it dangerous.

The compliant alternative: Describe your approach without referencing patient feedback. "Our assessment process is unhurried. We explain our findings and involve you in your treatment planning." This states a fact about the service rather than implying a patient experience or outcome.

Violation 2: The outcome claim

This one appeared in the sports injury section:

"Our targeted rehabilitation programmes help athletes return to full training faster, with fewer setbacks."

AHPRA defines a testimonial (and by extension a prohibited claim) as anything that "creates a reasonable expectation of beneficial treatment." That phrase applies directly to advertising copy, not just patient quotes.

"Return to full training faster, with fewer setbacks" is an outcome claim. It asserts a comparative result ("faster than what?" is the obvious question) without any substantiation. It creates the impression that choosing this clinic will produce a specific, superior health outcome. That's a prohibited representation under the guidelines.

This is the violation that healthcare marketing teams most commonly overlook. Outcome-adjacent language is everywhere in sports medicine marketing because it feels natural. Athletes want to get back to training. Of course you'd say you help them do that.

But the mechanism matters. You can describe what the service involves. You cannot claim that a particular outcome will result.

The compliant alternative: "Our sports rehabilitation programmes are designed around the specific demands of your sport and training load. We work with your treating team to develop a graduated return-to-activity plan." This describes the service and the process without implying a guaranteed or comparative outcome.

Violation 3: The qualification statement

The third violation was subtle enough that it would pass an internal review at most small practices. It appeared in the practitioners section:

"Our lead physiotherapist is a specialist in sports and musculoskeletal physiotherapy with over 10 years of experience treating elite athletes."

In Australia, "specialist" has a regulated meaning under AHPRA. A physiotherapist is only permitted to use the title "specialist" if they hold a formal specialist registration with AHPRA, which requires completing an approved specialist training programme and passing assessment. It is not a self-applied descriptor for having particular expertise or experience.

Using "specialist" without the corresponding registration is a misrepresentation under the guidelines, regardless of how experienced the practitioner actually is. A physiotherapist with 20 years of sports experience and no specialist registration cannot call themselves a specialist in any advertising material.

ChatGPT had no way of knowing the practitioner's registration status. It used the word "specialist" because it's common in healthcare marketing copy. The model was pattern-matching against general healthcare content, not applying regulatory knowledge.

The compliant alternative: "Our lead physiotherapist has worked in sports and musculoskeletal physiotherapy with both recreational and competitive athletes for over 10 years." Same substance. No regulated title claim.

Why does AI fail at compliance?

The violations above aren't random. They're predictable failures given how large language models are trained and what they optimise for.

AI optimises for fluency, not accuracy. A language model's training objective is to generate text that reads naturally and matches patterns in its training data. Compliance violations often sound natural. That's why they're violations. The copywriter who wrote them didn't think they were problematic either.

AI doesn't know what it doesn't know. ChatGPT acknowledged the AHPRA instruction by producing content that sounded considered. But acknowledging an instruction and successfully applying a complex regulatory framework are different things. The model had no reliable mechanism for distinguishing a compliant claim from a non-compliant one.

Compliance requires contextual judgement that generalises poorly. Whether "patients feel better" is a testimonial depends on phrasing, context, and how it could be reasonably interpreted by a member of the public. Whether "specialist" is a regulated title depends on the specific registration framework for that profession. These are not pattern-matchable from general internet text. They require specific regulatory knowledge applied in context.

The training data problem. AHPRA-violating content is abundant on the internet. Thousands of Australian healthcare websites use testimonials, outcome claims, and incorrect qualification language right now. The model was trained on that content. It learned what healthcare copy looks like, including the violations.

This doesn't mean AI tools are useless for healthcare content. It means they're not sufficient on their own. The output requires review by someone who knows the guidelines.

What should you do instead?

The answer isn't "don't use AI." It's "don't use AI unsupervised for regulated content."

A useful workflow:

Use AI for structure, not compliance. AI tools are excellent at generating first drafts, structuring content, and producing variations for testing. Use them for that. Then apply compliance review before anything goes live.

Train your review process on specific violations. The three categories that cause the most problems are: testimonials (direct and indirect), outcome claims, and qualification claims. For each piece of content, ask explicitly whether any of those are present. Not "does this seem compliant?" but "does this contain a testimonial? Does this claim an outcome? Does this use a regulated title?"

Check the AHPRA guidelines document directly. AHPRA publishes its advertising guidelines on its website and updates them periodically. The document is specific and example-driven. If you're producing healthcare content regularly, read it once a year. It's not long.

Brief AI tools more precisely. Instead of "write AHPRA-compliant content," prompt with specific constraints: "Do not include patient testimonials, direct or paraphrased. Do not claim any health outcomes. Do not use the word specialist unless the practitioner holds AHPRA specialist registration." More specific constraints produce better first drafts.

Work with a specialist writer for high-stakes content. Your homepage, service pages, and paid advertising are the highest-exposure pieces. They're also the ones AHPRA is most likely to review if a complaint is lodged. Getting those right from the start is cheaper than fixing them after a formal complaint process.

If you're producing healthcare content and want copy that converts without creating compliance exposure, take a look at how I approach this. Healthcare advertising compliance is one of the areas I work in specifically, across physiotherapy, allied health, telehealth, and specialist medical practices across Australia. Or get in touch directly to discuss your content needs.