Blog

What an AI assistant read before it named a clinic

Before it named a clinic, the assistant in these checks read a mix of clinic websites, dental and healthcare directories, review and booking pages, and "best of" style sources, then wrote an answer. It did not only read the clinic's own site. In a published dental check it pulled 608 pages to write 52 answers, and 46 of those answers drew on at least one directory, review or listing site. In a published physiotherapy check it treated one clinic as a sports injury clinic, cited competitors' condition and neighbourhood pages, and skipped the clinic for every other question. Both checks used Claude, with live web search, not ChatGPT. They are one run each, except for three repeated dental questions. The clinics are not named. AI results vary.

The figures below are the ones already published on this site, with their dates and limits. I have not added later runs or new businesses.

What did the dental check read?

On 16 October 2026 I asked Claude, with live web search, four patient questions about implants, Invisalign and teeth straightening, for 13 private dental practices in Greater Manchester and the West Midlands. The location was set to each practice's town. The practices were named in 21 of the 52 answers. Ten of the 13 were named at least once, one was named in all four, and three were not named in any. The implant question named the practice most often (7 of 13). Teeth straightening named it least (4 of 13). The full write-up is on the dentists page and in AI search for clinics.

Claude pulled 608 pages to write those 52 answers. Of the sources:

  • 46 of the 52 answers drew on at least one dental directory, review or listing site.
  • One directory, WhatClinic, appeared in 19 of the 52 answers.
  • 29 of the 52 answers quoted a star rating or a review count.
  • None of the 608 pages were Google pages (no Maps, Business Profile or other google. URL in the set that was returned).

That last point is easy to over-read. It means this run did not retrieve a Google URL as a source page. It does not mean Google Business Profile is irrelevant, and it does not tell us what, if anything, a listing provider might have copied from a profile. I still treat a complete, accurate Google Business Profile as part of the work.

In 6 answers, Claude's search returned the practice's own website and the answer still recommended other practices. Being found is not the same as being named. That distinction is in named or cited? What ChatGPT results actually mean. The counting rule is the same here: only a name in the answer text counts.

The anonymised Birmingham practice already on the sample report is one of those 13. It was named in 1 of 4 answers, for teeth straightening, and not named for implants, Invisalign or the cost question. Another practice in that same town's answers was named in all 4.

When three of the Manchester questions were each asked three times on the same day, from the same location, the practices named changed between runs for all three. Even the sources and the shortlist are not fixed.

What did the physiotherapy check read?

I also ran a Who-Gets-Named Check for a private physiotherapy clinic in a UK city: ten patient-style questions, Claude with live web search, one run, location set to that city. The clinic was named in 1 of 10 answers, even though it had a strong Google rating. Two rivals were each named in 6 of 10, and two more in 5 of 10. The questions and counts are on the physiotherapy page and in the sample report.

The clinic was named only for "best physio in [city] for a sports injury". It was not named for running injury, a private physiotherapist near the area, back pain, knee injury, shoulder pain, sports massage in the city centre, NHS or private self-referral, post-surgery rehab, or neck and posture. For the running-injury question, the assistant did not read the clinic's site and did not name it. For one other question, it read the clinic website and still recommended other clinics.

The sources help explain that pattern. The assistant only saw the clinic as a sports injury clinic, so it skipped the clinic for back, neck and post-surgery questions. Competitors had separate pages for each neighbourhood and each condition, and the assistant cited those pages directly. The clinic was missing from several of the directories the assistant read, such as healthcare listing and booking sites.

I do not have a published URL-level count for the physiotherapy check like the 608 dental pages, so I am not inventing one. The useful evidence is the question-by-question result plus the source types already in the sample: the clinic's own site in at least one answer where it was not named, competitor condition and area pages that were cited, and healthcare directories the clinic was missing from.

What does that mean if you run a clinic?

These snapshots are not a prediction of what an assistant will say about your practice. They are a record of what one assistant read, on one set of days, for these questions.

They do show a few things you can check for yourself:

  1. Open the citations. If the same directory, booking site or review page keeps appearing, look at your listing there before you commission a new piece of content.
  2. Give each main treatment or condition its own page, in language patients use, with the areas you cover. A clinic that only looks like a sports injury clinic may be skipped for everything else, even when the assistant has found the site.
  3. Keep one practice name, used the same way on the main website, the Google Business Profile and every listing. Put each treatment on that main site, not only on a second domain.
  4. Ask every patient for an honest review, offer nothing, and never tell them what to write. Ratings and review counts were quoted in 29 of the 52 dental answers. The legal rules are in reviews and AI search.
  5. Do not treat a strong Google rating, or the fact that your own site was read, as a mention.

How to run the same kind of test is in how to check if ChatGPT recommends your business. The AI visibility checklist is a wider list. Advertising rules for clinic pages are in AI search for clinics.

Frequently asked questions

Did you test ChatGPT on these clinics?

Not in the published figures above. Both checks used Claude, made by Anthropic, with live web search. I also ask ChatGPT the same kind of questions on client checks, logged out, and the report names the assistants used. A result on one assistant is not a result on another. AI results vary.

Why publish source types if the clinics are anonymous?

Because the useful part is what was read, not who the clinics were. Directories, condition pages and review figures are things another clinic can check without seeing a name.

If no Google pages were pulled in the dental check, can I ignore Google Business Profile?

No. That finding is only about the URLs returned in that run. It is not evidence that a complete profile is wasted. Google says a Business Profile can help local details appear in Google's own AI responses, and other listings often copy it.

Can you get my clinic named next time?

Nobody can promise that. AI results vary between runs, apps and users. I can show you who is named now, work on the pages and directories an assistant is already reading, and re-test.

Want someone to run the checks for you?

A Who-Gets-Named Check asks the patient questions for your town and shows who is named and which pages are cited. Published examples, with the names removed, are in the sample report. If you would rather hire someone, how to choose an AI visibility service lists the questions to ask, including of me. AI results vary.