Google's AI now answers the health question before the patient reaches your website. Ahrefs analysed 146 million search results pages and found an AI Overview on 44.1% of medical queries, more than double the 20.5% rate across all searches. Local searches are the opposite: only 7.9% trigger one. So Google answers "does dry needling help" itself, and sends "physio near me" straight to the map. Your clinic is absent from the exact stage where the patient decides who to trust. You cannot edit that answer. You can only become the source worth quoting.
Key takeaways
- 44.1% of medical searches return an AI Overview, against a 20.5% baseline across all searches (Ahrefs, 146,122,391 SERPs).
- Only 7.9% of local searches show one. Google answers the health question and hands the booking search to the map.
- In a study of 50,807 health prompts, 65.55% of the sources cited were commercial or uncertified pages. Government and academic sources made up about 1%.
- You cannot edit an AI Overview, and the same search can return a different answer twice in a row.
- An AI answer you did not write is not your advertising. The moment you screenshot it and publish it, it is.
- Section 133 pushes you toward plain, checkable, specific writing. That is the writing answer engines quote.
How often does Google's AI answer a health question?
Far more often than for anything else people search.
Ahrefs studied 146,122,391 search results pages and sorted them by topic. Medical queries returned an AI Overview 44.1% of the time, which is 5,924,866 AI answers across 13,430,263 medical searches. Across every search in the sample, the rate was 20.5%.
Health is the category Google is most willing to answer for you. That is a strange choice, given Google's own quality guidance says these topics deserve extra care. But the data is the data.
Why is the gap between 44% and 8% the real problem?
Most coverage of AI Overviews stops at "clicks are down". The more useful number is the gap between those two bars.
Think about how a patient actually moves. They start with a problem, not a provider. "Why does my jaw click." "Does physio help sciatica." "Is a root canal painful." Then, once they trust a course of action, they search for someone nearby to do it.
I mapped that journey in detail in what patients actually Google before booking. Here is what AI search does to it.
The first half of the journey is now mostly answered on Google's page. Nearly half of those medical questions get a summary at the top, and the patient may never click through. The second half, the "near me" search, is barely touched at 7.9%. That still goes to the map pack and the local results.
So the funnel splits. You are still visible at the booking stage, where the patient has already made up their mind about the treatment. You are far less visible at the question stage, where they decide what to believe and who sounds credible.
That is a real loss for clinics, because the education stage is the only stage where a compliant clinic gets to build trust. You are not allowed to win the booking stage with testimonials or promises. Education was your lever. AI just moved it.
What is Google's AI actually reading?
Not what you would hope.
SE Ranking ran 50,807 health prompts through Google in December 2025 and logged 465,823 citations. Only 34.45% of those citations came from sources they classed as trustworthy medical publishers. The other 65.55% were commercial sites, uncertified health portals and general blogs.
Academic and government sources together made up roughly 1%. Medical journals were 0.48%. National health institutions were under 0.4%. The single most cited domain was YouTube, at 4.43% of all citations.
One caveat, and it matters: that study ran in Germany, on German prompts. Treat it as directional rather than as an Australian measurement. Australian data at that scale has not been published.
Even as a directional signal, it reframes the opportunity. If two thirds of the sources filling health answers are commercial pages and general blogs, then a well-written clinic page is not competing with the Mayo Clinic for that slot. It is competing with a content-farm article about knee pain. That is a fight a real practitioner can win.
Is the answer about your treatment even accurate?
Sometimes it is not, and you will not be told.
In January 2026, a Guardian investigation reported by Search Engine Journal tested a set of health searches and asked clinicians and health charities to review the AI summaries. Reviewers flagged advice they called incorrect or harmful across several conditions, including guidance that could lead someone to delay care. Google disputed the findings and said the vast majority of AI Overviews are factual and helpful.
The finding that matters most for clinic marketing is quieter. The investigation noted that repeating the same search produced different summaries drawn from different sources.
That is the part clinics keep missing. An AI Overview is not a ranking you hold. It is generated fresh each time. You can check it on Monday, see a fair summary of your treatment area, and get something different on Thursday. There is no version to approve and no complaint form that edits it.
Does AHPRA hold you responsible for what the AI says?
This is the question I get asked most, and the answer has a clean line through it.
The AHPRA advertising requirements apply to advertising a regulated health service. That means content you create, control, or endorse. A Google AI Overview you did not write, cannot edit and did not commission is not your advertising.
The risk starts the moment you touch it. Here is where clinics cross the line.
| What the clinic does | Is it your advertising? |
|---|---|
| An AI Overview summarises your treatment area, favourably or not | No. You did not create or control it. |
| You screenshot a flattering AI answer and post it to Instagram | Yes. You published it, so you own every claim in it. |
| You quote an AI answer on your service page as support for a treatment | Yes. And an AI summary is not acceptable evidence for a therapeutic claim. |
| You used AI to draft your website copy | Yes, entirely. Who typed it makes no difference to your obligations. |
| An AI answer repeats a patient review about your clinic | No, until you republish it. Then it is a testimonial you have adopted. |
That last row catches good clinics. A five-star quote you did not solicit sitting on a third-party platform is one thing. The same quote pasted onto your own homepage is a testimonial in advertising, and testimonials about clinical care are banned under section 133 of the National Law. An AI summary does not launder it.
Why are compliant clinics better placed to get cited?
Here is the part nobody says out loud. The rules that feel like a marketing handicap are an advantage in AI search.
Answer engines are built to extract claims they can attribute. They favour writing that is specific, plain, self-contained and checkable. They struggle with vague persuasion, because there is nothing in it to quote.
Now look at what section 133 forbids: testimonials, superlatives, promised outcomes, unsubstantiated claims, offers without terms. Every one of those is a persuasion device with no extractable fact inside it. "The best clinic in Melbourne" gives a model nothing to work with. "Our initial physiotherapy assessment runs for 45 minutes, includes a movement screen, and is charged at a single set fee" gives it something to cite.
A clinic writing to the AHPRA standard is forced to be concrete. Concrete is what gets quoted. I made the wider version of this argument in how to rank in AI search when you are a regulated brand, and the difference between the two search surfaces in GEO versus SEO.
What should your clinic publish instead?
Stop trying to influence the AI answer. Start being the page it has to read. That is the whole practical shape of AI search in healthcare: you cannot edit the answer, so you make yourself the best available source for it.
- Write a page per real question, not per service. Your service list answers "what do you offer". The AI is answering "does this help my problem". Those are different pages.
- Answer in the first two sentences. Put the direct, factual answer at the top, before the background. Extraction happens early on a page.
- Name the boring specifics. Qualifications, years in practice, session length, what the appointment involves, what it costs, who a treatment may not suit. Specifics are citable. Adjectives are not.
- Use question headings. Match the words a patient would type, then answer directly underneath.
- Keep the local layer separate and accurate. Only 7.9% of local searches get an AI answer, so your Google Business Profile, hours, address and location pages still do heavy lifting. Fix them.
- Add Article and FAQPage schema. It will not buy you a citation on its own, but it makes your entity and authorship unambiguous.
Then run the compliance pass, because the same page has to survive both readers. My hub on SEO for healthcare covers the structural side in more depth.
One honest note on effort. This is slower than publishing a service page and hoping. It is also the only version that works twice, once for the patient reading and once for the model quoting.
This is general information for clinics and their marketing teams. It is not legal advice. The AHPRA advertising guidelines and the National Law are updated from time to time. Check the current versions on the AHPRA website, and get formal advice for high-stakes decisions.
