What Does ChatGPT Actually Tell Patients About Chiropractors?

About 45% of patients now ask ChatGPT or a similar AI assistant for recommendations. That number gets repeated in every marketing pitch aimed at practice owners, usually followed by a service that promises to get you named. Nobody was showing the other half of the picture: when a patient does ask, what comes back?

So we measured it. Not opinions about it, not a handful of spot checks — the same questions asked in the same way across every metro we cover, with every named practice written down and matched against the real list of practices in that market. This page is the whole result, including the parts that undercut the sales pitch.

There is nothing clever here, and that is deliberate. Anyone can repeat it.

1. We built a list of real chiropractic practices in each metro, drawn from public business and license records.
2. In each metro we asked ChatGPT the same four questions a patient would actually type — a straight request for a recommendation, a best-in-the-city phrasing, a symptom-led phrasing, and a near-me phrasing.
3. We recorded every practice ChatGPT named by name in its answers.
4. We matched those names back against the real practice list for that metro, so a name it invented or a clinic in another state does not count as a hit.
5. We counted, per metro, how many practices were checked and how many were named.

That is what the word named means everywhere on this page. Not mentioned in passing, not linked in a citation — actually stated as a practice name in the answer a patient would read.

Here is the complete table. Nothing has been left out for looking bad, and nothing has been rounded to flatter the result.

The most useful finding is also the most boring one. ChatGPT answers this kind of question in a few paragraphs, and a few paragraphs do not have room for a list of clinics. A large share of its answers name no specific practice at all and instead explain how to choose: check credentials, read recent reviews, ask about technique, confirm insurance, be wary of anyone selling a long prepaid plan on day one.

That is genuinely reasonable advice, and it is terrible news for a practice hoping to be handed patients. When it does name practices, it typically names two or three. Everyone else in the market is not ranked lower. They are not in the answer.

We counted names per metro, not per question, so we cannot tell you which of the four phrasings works hardest for a practice. Anyone who claims to know the magic wording is guessing. What we can say is that all four were ordinary patient phrasings and the combined result across them was still a small number of named practices in every market.

There is also a category of answer that names nothing local and instead points the patient at a way to look — check a map listing, check reviews, check whether the clinic takes your plan. From a practice owner's point of view that outcome is identical to not existing, but it is worth knowing that it happens, because it is the most common thing an assistant does with this question.

Put the five thinnest markets next to the five richest and the pattern is impossible to miss. The last column is straight arithmetic on the two columns before it.

In Jacksonville we checked 254 practices and ChatGPT named 2. In El Paso we checked 33 and it named 15. Same assistant, same four questions, same week. The difference is the size of the pool competing for two or three slots.

It is worth being careful about what that does and does not prove. It does not prove that practices in small markets are better described or better run. The simplest explanation is arithmetic: a short answer holds a fixed number of names, so the fewer practices there are, the larger the share of them that fits.

This is the section most benchmark pages leave out. We recorded names. We did not record the reasons behind them, so any claim about why a particular practice was named would be us guessing in public.

• We did not record how many reviews the named practices had, so we cannot tell you a review count that gets you named.
• We did not record their website quality, their page speed, or how long they had been in business.
• We did not record whether they were cash-based, insurance-based, or part of a chain.
• We did not measure clinical outcomes, and nothing in this data speaks to quality of care in either direction.
• We did not test how the answers change for a patient with chat history, a logged-in account, or a different phone.

What the data supports is a count and a spread. Everything past that belongs in a different study.

Take the whole thing with the caveats it deserves. The run happened on 2026-07-03. It is one snapshot.

• One point in time. AI assistants change what they say from month to month, and a single run cannot show a trend. We re-run it, which is the only way to see movement.
• Four question phrasings. Real patients ask in a thousand ways, and a different phrasing can surface a different set of names.
• Two states. Florida and Texas are big and varied, but they are not the country, and nothing here should be read as a national figure.
• One assistant. We measured ChatGPT because that is what patients name when they describe what they used. Other tools will differ.
• Practice lists are imperfect. Clinics close, rebrand, and move, and a name that could not be matched to a real practice was not counted as a hit.

None of that makes the finding fragile. The headline gap between a couple of percent in the biggest markets and tens of percent in the smallest is far too wide to be an artifact of phrasing. But anyone who tells you a single benchmark run settles the question is overselling, including us.

Three practical conclusions come out of the numbers, and none of them require you to believe anything about the future of AI.

1. Stop treating this as a ranking problem. There is no position 12 to climb out of. There is a very short list, and you are on it or you are not.
2. Weight your effort by your market. In a metro at the top of the table, this is a long project, and the patients you already attract deserve the first attention. In a thin market, the door is genuinely open right now.
3. Measure before and after, with the same questions each time. A single lucky answer on a single day tells you nothing, which is the whole reason this benchmark exists.

If you want it handled rather than tracked yourself, Get Recommended is $97 a month while we do the work, and the full $495 a month only starts once ChatGPT is actually naming your practice. We priced it that way because of this data. In most of these markets the honest answer is that it takes a while.

1. Open a fresh chat, signed out if you can, so your own history does not steer the answer.
2. Ask for a chiropractor recommendation in your city, exactly the way a patient would say it out loud.
3. Ask again three more ways: best in town, a symptom you treat, and near a landmark or neighborhood.
4. Write down every practice name that appears, and how many of the four answers named anyone at all.
5. Repeat it in a month and compare the two lists. That is the only comparison that means anything.

If your name never appears, you now know where you stand, and you know from the table above roughly how crowded your market is. Both of those are more useful than any promise about what AI search will look like next year.