Two searches, one suburb: what AI systems actually say when somebody looks for help with shoulder pain

On a Tuesday evening in August I ran two searches about the same suburb.

The first named a profession. Physiotherapist North Lakes.

The second described a problem instead. Who do I see for ongoing shoulder pain in North Lakes.

I put both through seven places: Google's ordinary results, Google Maps, Google's AI Overview, Google's AI Mode, Gemini, ChatGPT and Perplexity.

I expected the answers to differ a little. They differed a great deal — and one of the differences has a direct consequence for any practice that wants to be found.

Do the systems agree on who you should see?

No, and this is the part I did not expect.

Same question. Same suburb. Same evening.

Two of the systems said to start with a general practitioner. One put a GP first and framed everything else as what follows a diagnosis.

Two said to start with a physiotherapist, and one added that in Queensland you do not need a referral to book privately.

One offered both as equally reasonable first steps.

Every one of them was reasonable. Every one cited sources. And a person asking the same question twice, in two places, would be sent to two different professions.

That is worth sitting with if you run a practice. A patient with a sore shoulder is not choosing between your clinic and the one down the road. They are being told, by whichever system they happened to open, whether your entire profession is the right place to start.

You cannot control that. But you can make sure that when a system does decide your profession is the answer, it has something to read about you.

Does naming the condition change who appears?

Yes, completely — and this is the finding a practice can act on tomorrow.

Search one named a profession. The results were physiotherapy practices, near-identical across most surfaces. A handful of names, the same handful more or less everywhere.

Search two named a symptom. The field opened completely — and it crossed professional boundaries.

The results now included general practices, orthopaedic surgeons and a day hospital alongside chiropractors, osteopaths, a myotherapist, a Chinese medicine clinic, an occupational rehabilitation centre, a pain control clinic, a hand and upper limb clinic, and a general surgeon.

Most of those appeared for the symptom and not for the profession. They were invisible in the first search and prominent in the second.

This is why the point applies well beyond allied health. A general practice, a specialist consulting room and a physiotherapy clinic were all competing for the same question, because the person asking had not decided which of them they needed. Whenever a patient describes a problem rather than naming a profession, every profession that could plausibly treat it lands in the same result.

One example is worth describing properly. A clinic in the suburb that works only on the upper limb — shoulder, elbow, wrist, hand — did not appear at all when I searched for a physiotherapist. In the shoulder search it appeared in four of the six answering systems, was named in the AI answers with a description of what it does, and was also running an advertisement.

Nothing about that practice changed between the two searches. What changed was that the second search matched what their website actually says — and what it says is the body part and the conditions. They did not lead with the profession at all.

That is the opposite of how almost every practice website is written. Most say the profession, list the suburb, and describe services in professional language. That covers the search a patient makes once they already know what they need, and misses every search made before that.

Patients do not usually search for a profession. They search for what hurts.

Which practice appeared on every single surface?

Across both searches and all seven systems, exactly one practice appeared every time.

It has a page for its location in that suburb rather than one page covering the whole group. That page names shoulder, neck and back pain explicitly. It lists individual practitioners with their own areas of focus.

One of the AI systems named three of those practitioners individually, and said each listed a clinical focus in shoulder injuries.

That is the clearest demonstration I have seen of what these systems actually read. Not a homepage. Not a services list. A location page that names conditions, and practitioner pages that name what each person treats.

Two things follow from that.

• If you have more than one location, a single page covering all of them is doing half the work of separate pages.
• Practitioner biographies are not staff decoration. They are content a machine will read, attribute and quote.

Where did the AI answers get their information?

Every AI system cited its sources, and the sources are informative.

Alongside practice websites, they cited a health services directory run by a state health department, two entries from a national appointment booking platform, and a health provider directory.

That is worth knowing, because those listings are largely free and most practices have either never claimed them or filled them in once and forgotten.

Your website is not the only thing describing you. When a system builds an answer, it draws on whatever it can find. If a directory entry is thin or wrong, that is what gets read.

Is the AI answer at the top of the page?

No. On the symptom search the first thing on the page was neither the AI answer nor the ordinary results.

It was advertising. A block of sponsored listings, then more, then the AI Overview beneath them, and further sponsored listings after that.

So the AI answer was not the top of the page, and neither were the ordinary results. On a question with obvious commercial value, the space above both was bought.

One thing I noticed that I have no explanation for. Among the advertisers bidding on shoulder pain in this suburb was a national physiotherapy body's own consumer website. Whatever the reasoning, a practice competing for that space is competing with its own profession's peak organisation.

Do the systems tell you what they could not match?

Sometimes, and it is easily missed.

On two of the ordinary Google results, the listing carried a note saying a word from my query was missing from that page, with an option to force results containing it.

The word was "ongoing".

That is Google showing its working. Those pages matched shoulder, and pain, and the suburb. They did not match the word describing how long it had been going on — which is precisely the word that distinguishes a nagging problem from a new injury, and precisely the word a patient in that situation would use.

Chronic. Ongoing. For months. Will not go away. Those are patient words, they are searchable, and most clinical websites do not contain them.

What should a practice actually do about this?

What makes health practices different from every other business?

Everything above would apply to a plumber.

This next part applies to doctors, dentists, physiotherapists, psychologists and every other registered practitioner — and it does not apply to a plumber at all.

First, a distinction most writing on this subject gets wrong. Not every health profession is registered with AHPRA. Sixteen are, including medical practice, dental practice, nursing and midwifery, pharmacy and optometry, alongside physiotherapy, psychology, podiatry, chiropractic, osteopathy and occupational therapy.

Speech pathology, dietetics, exercise physiology, audiology, myotherapy and remedial massage are not. They are self-regulated by their professional associations, under different rules.

That matters in both directions. Applying AHPRA restrictions to a dietitian limits a practice unnecessarily. Ignoring them for a physiotherapist misses a legal obligation. If somebody offering you marketing help does not know which category you are in, that is worth noticing early.

Can an AHPRA-registered practice use patient reviews at all?

Some of them, and the rule is narrower than most people state it.

Section 133(1)(c) of the National Law prohibits advertising a regulated health service in a way that uses testimonials. You will read, constantly, that AHPRA practitioners cannot use testimonials. That is too broad.

AHPRA's own guidance defines a testimonial as a positive statement about the clinical aspects of a regulated health service, and sets out a three-part test. A clinical aspect exists if the comment expresses any one of:

• a symptom — the specific symptom, or the reason someone sought treatment
• a diagnosis or treatment — the specific diagnosis, or what the practitioner did
• an outcome — the specific result, or the practitioner's skills or experience, directly or by comparison

Comments that touch none of those are not testimonials and may be used. AHPRA's own examples are customer service and communication style.

So a review saying the reception staff were kind and the parking was easy is not caught. A review saying the treatment fixed a shoulder is.

Three things that catch practices out even so.

• Editing is its own breach. Publishing only the positive half of a mixed review, or tidying a negative one, is misleading regardless of whether the content is clinical. Complete and unedited, or not at all.
• Responsibility follows control — and that includes content produced by an agency on your behalf. If you can publish it, edit it or remove it, you are the advertiser.
• The rules reach far past paid advertising: websites, blogs, landing pages, social posts, reels, stories, email, brochures, signage, and Google Business Profile posts. And while you are not required to remove reviews from platforms you do not control, engaging with them can itself count as using them.

Outcome claims and before-and-after comparisons remain restricted separately.

There is a further consequence almost nobody discusses. Comparison and ranking content — the format these systems appear to cite most readily — is largely closed to registered practitioners. A practice cannot publish a page ranking the best physiotherapists in its own suburb. The most citable format in the field is unavailable to the practitioners who most need citing.

That is a real constraint and I am not going to pretend otherwise.

What remains available is more than it sounds, and everything recommended earlier in this article sits inside it. Explaining a condition. Naming the words patients use. Describing who you treat. Listing practitioners and their focus. None of that requires a claim, and all of it is what a machine needs in order to describe you accurately.

The harder question — what actually replaces a testimonial when a search engine is looking for signs of trust, and what the permitted non-clinical review is genuinely worth — deserves its own article, and will get one.

There is good compliance guidance available, some of it free and carefully done. What none of it addresses is what happens when a machine summarises a practice, because it was all written for a human reading an advertisement. A practitioner can control what they publish. They cannot directly control how a system describes them. That gap is where this work sits.

How this was done

I would rather set out the method than have you take the findings on trust.

Two searches about the same suburb, run within an hour of each other on 25 August 2026. Seven surfaces: Google's ordinary results, Google Maps, Google's AI Overview, Google's AI Mode, Gemini, ChatGPT and Perplexity. Private browsing throughout, with no account signed in anywhere. Results recorded as screenshots and full text rather than summarised from memory.

Three further systems — Claude, Copilot and Grok — were excluded deliberately. All three require an account before you can ask them anything, and a signed-in result carries that account's search history and personalisation with it. Mixing signed-in and signed-out results would produce a comparison that is not really a comparison. Everything reported here was run under identical conditions.

Two things worth knowing about what this can and cannot tell you.

It is one evening and one suburb. An observation, not a study. A different suburb, or a different week, could produce something else.

And these systems vary between sessions. Ask the same question twice and the answer moves. Which means every before-and-after screenshot in this industry is weaker evidence than it appears, including any I could produce for you. I would rather tell you that than sell you a screenshot.

What that argues for is not chasing a position on any single day. It is being described clearly and consistently everywhere a machine might look, so that you are a plausible answer more often.

What is the check, and how long does it take?

About two minutes. Do it before you pay anybody, including me.

On your phone, in a private browsing window.

• Search your practice name. You will almost certainly appear.
• Now search what you actually do, the way a stranger would say it. The service and the suburb.
• Now search the problem, the way a patient would say it. Not the clinical term. The complaint. Sore shoulder. Back pain that will not go away. Sore heel in the morning.
• Then ask ChatGPT the same three things.

Compare them.

Most practices find they appear for the first, sometimes for the second, and rarely for the third. That third search is the one where somebody has not yet decided who to see — which makes it the one that matters most and the one almost nobody has written for.

Being findable and being good are separate problems. Only the first one is discussed here, and no search engine has any way of assessing the second.

Searches run and recorded on 25 August 2026. Regulatory detail checked against the Australian Health Practitioner Regulation Agency's published advertising guidance and its guidance on testimonials on 25 August 2026. Platform behaviour and regulatory guidance both change. Verify before relying on any detail here. This article is not legal advice.