Why AI doesn't recommend your practice

If AI engines don't name your practice, it's almost never a judgment about your surgery. It's that the engines couldn't verify enough about you to quote you with confidence, so they named someone they could, or gave generic advice instead. That's a records problem, and records problems are fixable.

The useful part is that you don't have to take anyone's word for it. You can check what the engines say about your practice tonight, in one sitting, and the result puts you in one of three situations. This guide covers how to run that check properly, what each outcome means, and which causes are yours to fix.

How do you check what AI engines say about your practice?

Open ChatGPT, Perplexity, and Google's AI results, and ask each one the question a patient would type. Not your practice name. The patient doesn't know your practice name yet, which is the whole point of the exercise.

Ask it three ways:

• The procedure and the place. "Who's the best knee replacement surgeon in Tucson?" Swap in your procedure and your metro.
• The symptom description. Patients don't always know the name of what they need. "Who treats bone-on-bone knee arthritis in Tucson?"
• The vetting question. Ask by name: "What is Dr. [your surname] at [your practice] known for?" This is the question a patient asks after a friend or a physician gives them your name, and it's the easiest one to forget, because it doesn't feel like a visibility question.

Two rules make the check honest. Use a logged-out or private window, because a logged-in session can feed your own history back to you. And run each question more than once. These systems don't return the same answer every time, so run it three times and read the pattern rather than the first answer.

Write down what comes back: who got named, what the engine said about them, and where it says it got that from. That list is your baseline, and it will be more specific than anything a vendor tells you about your visibility.

No engine names you

The engines couldn't confirm you well enough to put your name in an answer. Your baseline is "Not yet."

In our own July 2026 measurement, we put fifteen patient- and practice-side questions to two AI models with web search switched off. None of the thirty answers named an individual surgical practice. They named directories over and over. We publish that in full, including the limits of what it can tell you, in what AI models know about finding a surgeon.

That tells you what the models carry before they search: directory names, not practice names. It doesn't tell you what a live engine says about you today, which is exactly why you run the check yourself. What we can say is that the practices which do get named got there by being verifiable, not by being better at surgery.

Absence usually traces to one of four things, in the order worth checking:

• Your site doesn't answer the question in the words patients use. Engines can only quote a page that already contains the answer. A practice site that lists services as a menu, with no page explaining what a partial knee replacement is or who's a candidate, has given the engine nothing to quote.
• The machine-readable labels behind your pages are missing or wrong. Structured data is how a page tells an engine "this is a medical practice, at this address, with these clinicians". Without it, an engine reading your prose is guessing, and it tends not to quote its guesses.
• Your details disagree across the internet. One suite number on your site, a different one on an insurance directory, a former partner still listed on a hospital page. Contradictions cost you the citation, because an engine that cannot reconcile two records tends to use neither.
• Nothing independent corroborates you. Engines weigh what other sources say about you, and your own site can't supply that.

The engine named a competitor instead

Another practice holds that answer slot, and you now know which one and on what basis. This is more useful than absence, because the engine usually tells you what it cited.

Look at what it pointed to. If the engine is quoting the named practice from its own procedure pages, the slot was won with content you could also produce. A hospital system's directory as the source means that practice is borrowing authority from a domain it doesn't own, which is a weaker position than it looks and a reminder to build on ground you control. Reviews and third-party profiles as the source mean the corroboration layer is doing the work.

None of that means the slot is closed. The engine named that practice because it could verify it, not because the practice is entrenched, and verifiability is a thing you can build.

The answer names you and gets it wrong

This is the result worth acting on fastest. A wrong answer is worse than absence, because it's being read.

Three versions of it are worth testing for specifically: a surgeon credited with a procedure the practice no longer offers, an address belonging to a suite the practice has moved out of, and a clinician still attached to a group they've left. Each one traces to a stale record somewhere public, and the engine is faithfully repeating what it read. Fix the source, and the answer follows on the engines' schedule rather than yours.

Which of these causes can you actually fix?

Most of them. But the split matters, because a vendor who promises to fix all of it is telling you something about the vendor.

Yours to fix: the pages on your own site, the structured data behind them, the consistency of your name, address, credentials, and procedure list everywhere it appears, your review profiles, and your professional listings. Every one of those sits on a surface you own or control.

Not yours to control: what any individual engine decides to say on a given day. Nobody outside those companies decides what an answer says, and the models and the retrieval behind them change without notice. Anyone guaranteeing you a citation is guaranteeing something they don't own.

What you should expect from anyone doing this work is a number that moves. Run the check monthly and watch whether your name starts appearing. Keep the questions identical each time; if the test moves, the number means nothing.

Most of the fixable half is documented publicly. Google's structured data guidelines cover the machine-readable labels behind your pages, and its local business structured data reference covers the name, address, and credential fields engines cross-check. The wider view of how engines choose sits on our AI search visibility page, and the shared foundation underneath both classic search and AI answers is in our guide to SEO for doctors.

If you'd rather not run it yourself, our free visibility audit does the same three questions across the same engines and hands you the answers written down. We review every request, and qualifying practices get it back within 24 hours, at no charge.

Either way, run the check before you buy anything from anyone. You'll get more out of a vendor conversation once you know what the engines say.

Common questions

How do you check if ChatGPT recommends your practice?

Ask it the way a patient would, in a logged-out or private window, without using your practice name: "who's the best [procedure] surgeon in [your city]". Run it two or three times, since answers vary between runs, and repeat the check in Perplexity and Google's AI results. The third question is the one using your name, because that's what a patient does after someone refers you. Record who gets named and what the engine cites.

How often should you check?

Monthly, with the same questions every time. The value is in the trend. Changing the questions each month produces a number that moves for reasons that have nothing to do with your practice.

The engine named your practice but the details were wrong. What now?

Find the public source it's repeating and correct it there. Stale addresses, former partners, and dropped procedures usually live in directory listings, insurance panels, hospital affiliation pages, or an old profile nobody has logged into in years. Correct the record at the source, keep every listing identical, and the answers follow when the engines next refresh. Fixing your own website alone rarely resolves it, because the contradiction is what caused the problem.

Is there a way to submit a practice to ChatGPT?

No. Nothing you can fill in puts you in an answer, which is why a vendor selling placement is selling access it doesn't have. The engines assemble answers from sources they can already reach, so the only lever is what those sources say about you: your pages, your structured data, your listings, and what independent sources say about you.

What should you write down when you run the check?

Three things per answer: which practices got named, what the engine said about each, and what source it pointed to. The source matters most, because it tells you where the engine's confidence came from and therefore which surface to fix. Keep the list, because next month's version of it is the only way to tell whether anything you did worked.

Does it matter which engine you check first?

Not much, and checking only one is the real mistake. They retrieve differently and cite differently, so a practice named in one can be absent from another. Run the same questions across at least three, and treat a name that appears in only one as a weaker signal than a name that appears in all of them.