What this is
We ran the same set of patient questions ("who is the best injector in [city]", "best med spa near [neighborhood]", "where should I get [procedure]") through Claude, ChatGPT, Gemini, and Perplexity. Fifteen questions per market, asked three times each, across 18 markets in three countries. About 2,800 answers in total.
For every answer we recorded two things: whether the practice was named in the recommendation, and whether the practice's own website was used as a source. Those two numbers turn out to tell very different stories.
Every number below is a real measurement of a real practice. Practice names are withheld. The method is repeatable and we re-run it per audit.
Do patients actually use AI to choose an aesthetic practice?
Short answer: the ones who spend the most do.
Bain & Company, working with Comite Colbert, surveyed luxury buyers in April 2026. Fifty-four percent of United States luxury buyers used AI during their most recent luxury purchase. The number rises with spending: 82% of the heaviest spenders used AI, compared with 28% of the lightest.
A 2025 Rater8 survey found 70% of United States adults are open to using or already use AI to research physicians.
What that means for a practice: the patient who books one modest treatment may never ask an AI anything. The patient who funds your year increasingly does.
Does market size help or hurt AI visibility?
It hurts. In our measurements, the smaller the market, the more often the practice gets named.
Practices in the largest, densest markets were named in a small share of answers: 8% in Manhattan, 7% in Dallas, 10% in Miami, 2% in Mexico City, 2% in Las Vegas.
Practices in mid-size and suburban markets were named far more often: 60% in the Twin Cities, 55% on the Mexican border, 50% in Boise, 46% in an LA suburb, 46% in a Vancouver suburb, 35% in London Ontario.
In a small market the effect is stronger still. In Erie, Pennsylvania, two practices were each named in more than half of all answers, and the top four practices absorbed nearly every recommendation.
The pattern held in the United States, Canada, and Mexico.
The practical reading: if you are outside a top-ten metro, the AI shelf in your market is winnable now, and a small number of practices are already taking it.
Why does AI read my website and then recommend someone else?
This is the finding practices are most surprised by, and it is common.
In one Chicago-area market the practice's website was used as a source in 47% of answers while the practice was named in 22%. In Pittsburgh, 38% versus 22%. In Las Vegas, 17% versus 2%.
Here is the mechanism. When an engine answers a question like this, it runs live searches, pulls somewhere between five and fifteen pages (review platforms, directories, and several practices' own websites), reads them, and then features a few practices by name.
So a practice with strong content makes the shortlist. Its pages are retrieved and read. It loses the feature slot at the last step, when the engine compares the candidates.
The engines tell you what they compare on. They write it into the answer. One verbatim example from our transcripts, justifying a pick: "strong recent review volume (about 4.8/5 from 200+ reviews)."
Across the corpus, the sources engines lean on for those comparisons are review platforms and directories: Yelp appeared 733 times, RealSelf 595, Healthgrades 188.
Bain observed the same shape at brand level: roughly 90% of the web addresses that large language models cite are not the brand's own site.
So your website earns the reading. Your review presence earns the recommendation. Most practices are investing in the first and assuming it buys the second.
Can I just check one AI engine and know where I stand?
No. The same practice, asked the same questions on the same day, scored very differently by engine.
One Los Angeles practice was named in 60% of Claude's answers and 20% of ChatGPT's. A Chicago practice: 26% and 13%. A Minneapolis practice: 68% and 46%.
Any report built on a single engine will tell you that you are winning or losing, and it has a good chance of being wrong. Ask across engines, and ask more than once, because answers vary between runs.
What are the AI engines actually looking at?
When we asked the engines what a practice should be doing, their advice assembled itself from the same ten areas every time. This is a fair map of what the machines consider your public presence to be.
Website fundamentals: structure, speed, mobile experience, real service pages.
Local presence: Google Business Profile, map pack visibility, consistent listing information.
Reviews and reputation: volume, rating, and whether the practice responds.
Paid advertising: whether the practice shows up in paid results at all.
Social media: cadence, engagement, and whether posts lead anywhere.
Email and text follow-up.
Content and patient education.
Positioning: whether the practice explains what makes it different.
Booking friction: how many steps from interest to appointment.
Measurement: whether anyone can tell which of the above is working.
Two of these carry unusual weight in AI answers specifically. Reviews decide the feature slot, as shown above. Google Business Profile decides whether the practice appears in the local layer the engines lean on.
What surprised us
Software companies outrank marketing agencies as sources. When a practice owner asks an AI how to get more patients, the engines cite practice-management software vendors and industry associations more often than they cite agencies. Those companies win by publishing plain, structured, evergreen explanations of how things work.
Video and forums carry real weight. YouTube and Reddit were among the most cited sources across those answers.
Nobody is measuring. Every piece of advice we pulled was generic. Not one answer cited a measurement of a specific practice in a specific market.
How to read your own position
Ask an engine the question your patient would ask, phrased for your city and your procedure. Ask it three times, on at least two engines.
Note two things separately: were you named, and was your website listed as a source. The gap between those two numbers is the finding.
If you are named: hold it, and watch it, because these answers change.
If you are cited but never named: your content is working and your comparison signals are not. Reviews are the first place to look.
If you are neither: start with the Google Business Profile and the review base, because that is the layer the engines read first.
About this measurement
Method: 15 patient-intent questions per market, 3 runs each, across Claude, ChatGPT, Gemini, and Perplexity in 7 markets and across Gemini and Perplexity in 11 markets. Roughly 2,800 answers, July 2026.
Limits worth stating. Markets are named at regional granularity rather than by municipality, because the practices behind them are published de-identified and a town name would undo that. Eleven of the eighteen markets were measured on two engines, so the engine-comparison numbers come from the seven full runs. Most markets tracked a single practice, with no full competitive set. AI answers change over time, so these are a July 2026 snapshot. And the relationship between reviews and being featured is what the engines say they are doing, which is not the same as a controlled experiment.
Fractional Aesthetics is a growth strategy practice for aesthetic medical businesses. We run this measurement as part of FAx, a 34-item audit of a practice's public presence.
See the same method applied to one practice
This measurement runs as part of FAx, our audit of what a patient actually finds when they look for a med spa, a plastic surgeon or a dermatology practice. Sample audits are published in full, with identifying details removed.