AI Is Writing Your Competitor's Content. Here's What Google Actually Does About It.

Google does not penalise AI content in healthcare for being AI. It says so in writing. Its published guidance is that it focuses on "the quality of content, rather than how content is produced", and that "Appropriate use of AI or automation is not against our guidelines". What Google does penalise is scaled content abuse, which is defined by volume and value, not by who typed it. Meanwhile AI detectors are a poor way to prove anything. In one study, seven detectors falsely flagged an average of 61.22% of essays written by non-native English speakers. For an Australian clinic the real exposure sits somewhere else entirely, in section 133 of the National Law.

Can Google tell if your competitor used AI?

Probably. But it is the wrong question, and Google has said so for years.

Google's guidance on AI-generated content is public. The line that matters is this one: "Our focus on the quality of content, rather than how content is produced, is a useful guide that has helped us deliver reliable, high quality results to users for years."

The FAQ on the same page is blunter. "Appropriate use of AI or automation is not against our guidelines."

And in the body: "it's important to recognize that not all use of automation, including AI generation, is spam".

Google also explains why it took that position. It points back to the era of mass-produced human writing. Its words: "No one would have thought it reasonable for us to declare a ban on all human-generated content in response."

So a clinic waiting for Google to punish the practice down the road is waiting for something Google has ruled out in writing. Worth knowing before you spend money on the wrong fix.

What does Google actually penalise?

Scaled content abuse. It is a named policy in Google's web spam policies, and the definition is short.

"Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users."

Then the sentence that does the real work: "This abusive practice is typically focused on creating large amounts of unoriginal content that provides little to no value to users, no matter how it's created."

Read the last five words again. No matter how it's created. The policy is deliberately neutral about the tool. A person can trigger it with a keyboard. So can a model.

The policy then lists examples. The first one is the one clinics walk into: "Using generative AI tools or other similar tools to generate many pages without adding value for users".

In practice, for a healthcare site, that almost always looks like one thing. Suburb pages. Physiotherapy in Carlton. Physiotherapy in Fitzroy. Physiotherapy in Brunswick. Thirty pages where the suburb name is the only thing that changes.

That pattern was a risk before anyone had heard of ChatGPT. What AI changed is the cost of making thirty of them. It did not change the policy.

Does Google treat health content differently?

Yes, and it says so on the same page.

Google names health as one of three areas where information quality is critically important, alongside civic and financial information. For those areas, it says, "our systems place an even greater emphasis on signals of reliability".

That is one sentence, and it is the whole reason a clinic cannot treat content the way a homewares shop can. The bar is higher on your topic by Google's own account.

It also lines up with what actually gets a clinic page cited in AI answers, which we covered in why ranking on Google is not enough. Reliability signals are practitioner names, registration numbers, real qualifications, dates, and sources. Those are exactly the things a model does not know about your practice and will not invent correctly.

Can anyone prove a competitor used AI?

Not reliably. And the attempt has a specific failure mode that matters in Australian healthcare.

Researchers tested seven widely used GPT detectors against 91 essays written by humans sitting the TOEFL exam, so people writing English as an additional language. They also tested 88 essays by US 8th-grade students.

The detectors were, in the study's words, "near-perfect" on the US 8th-grade essays. On the TOEFL essays they fell apart.

So the tool people reach for to prove a competitor cheated is a tool that wrongly accuses human writers, and it does it selectively. It misfires on people who write English as an additional language.

That matters here more than in most industries. In 2024/25 alone, 26,703 internationally qualified practitioners gained registration to practise in Australia, according to the Ahpra and the National Boards annual report for that year.

Being internationally qualified does not make someone a non-native English writer. Plenty qualify in the UK, Ireland, New Zealand and Canada, and the report does not break the group down by language. So treat this as an overlap, not an equation. It is still a large enough overlap that pointing a detector at an Australian clinic's team bios is a bad idea. The likeliest result is that you accuse a real practitioner of not writing their own page.

Who is liable when AI writes a clinic's advertising?

This is the part that has nothing to do with Google, and it is the part that can cost you $60,000.

Section 133 of the Health Practitioner Regulation National Law is short. It says a person must not advertise a regulated health service in a way that does any of five things:

• Is false, misleading or deceptive.
• Offers a gift, discount or other inducement without stating its terms.
• "Uses testimonials or purported testimonials about the service or business".
• "Creates an unreasonable expectation of beneficial treatment".
• Encourages the indiscriminate or unnecessary use of regulated health services.

The maximum penalty is $60,000 for an individual and $120,000 for a company.

Now read subsection (2), which almost nobody quotes:

That is a defence, and look who it is for. It protects the party that produces or publishes the advertisement for someone else. The printer. The platform. Arguably the agency.

There is no matching defence for the practice the advertisement is for. The law already decided, years before generative AI existed, which side of that line carries the risk.

So "the AI wrote it" is not a defence. Neither is "the agency wrote it", which we went through in detail in who is liable when an agency writes your copy.

And the failure is not hypothetical. When we asked ChatGPT to write compliant copy for a physiotherapy clinic, it produced three separate breaches, and the copy read beautifully. That test is written up in full in what happened when we asked ChatGPT for AHPRA-compliant content.

The reason is structural. A model writes what marketing copy usually looks like. Most marketing copy in its training data comes from countries where patient reviews, outcome promises and "best in the area" are all perfectly legal. In Australia they are not.

What does a clinic page need to survive both tests?

There are two graders, and they want different things. A page has to pass both.

Google wants a reason for the page to exist, plus reliability signals on a health topic. AHPRA does not care about your rankings at all. It cares whether the words on the page break section 133.

A page passes both when it carries things a model cannot generate about you:

• A named practitioner, with the registration number and the actual qualification.
• What the appointment involves, in specifics, including what it does not cover.
• Real prices, or a real reason there are none.
• Dates, and a source for any clinical claim.
• No patient quotes, no outcome promises, no "leading" or "best".

Notice that every item on that list is also a reliability signal. That is the useful part. The work that keeps you compliant is largely the same work that makes the page worth ranking. Compliance and quality point the same direction here, which is not true in every industry.

So what should you actually do?

Stop watching the competitor. Their AI content is not your problem, and you cannot prove it anyway.

Do these four things instead.

One. Count your own thin pages. Look for suburb or service pages that differ only by a word. That is the scaled content abuse pattern, and it is yours, not theirs. Merge them into one page that is actually good.

Two. Read your own site against section 133 before a regulator does. AHPRA assessed 775 advertising complaints in 2024/25 and has trialled AI scanning of websites, which we covered in what 775 advertising complaints actually led to. Enforcement no longer waits for a complaint.

Three. Keep using AI, and change where it sits. Use it for structure, for first drafts, for turning a practitioner's spoken answer into clean prose. Do not let it supply facts about your practice or the clinical claims. That division survives both tests.

Four. Put the human evidence back in. Names, numbers, dates, sources. It is the thing Google says it weighs more heavily on health topics, and it is the thing a model cannot fake about you.