Ask ChatGPT or Perplexity to recommend a med-spa near you, and you'll get a list. Three names, maybe five. Clean, confident, no hedging. What you won't get is a straight answer to the obvious follow-up question: why those names?
The assumption most owners make is that the AI is doing some version of what a careful patient would do — checking who's board-certified, who's got a medical director, who's licensed to inject neuromodulators versus who's just supervising an aesthetician. That's the story the technology tells about itself. Ask a question, get a considered answer.
But that's not what's happening. And the reason matters if you run one of these practices or market for one.
The credential data barely exists in a usable form
Here's the problem. Med-spa qualifications are real — state licensing, medical director requirements, RN and NP credentials, physician oversight rules that vary by state. All of that is legally significant and clinically meaningful.
It's also almost entirely absent from the text an AI model was trained on, and mostly absent from the pages it retrieves at query time.
Think about where credential data actually lives. State medical and nursing boards keep license lookups behind clunky search forms, often not indexed, often not even scrapeable in a clean way. A practice's "medical director" might be named once, in small type, in a footer. Certifications get listed as logos — images, not text — on an About page. There's no standard schema. One practice writes "physician-supervised," another writes "board-certified team," another says nothing at all because the owner assumes it's obvious.
So when a model tries to assemble an answer about who's qualified, it's reaching for a signal that's thin, inconsistent, and frequently missing. The credential corpus is weak.
So the model falls back on what's abundant
When the strong signal is missing, models don't return "insufficient data." They answer anyway, using whatever signal is plentiful. And in the med-spa category, the plentiful signal is mentions.
Not clinical mentions. Marketing mentions. How often a practice's name shows up across the aggregators — Yelp, RealSelf, Groupon, "best med-spas in [city]" listicles, directory pages, local blog roundups, review platforms. That footprint is dense, text-rich, and easy to retrieve. It's exactly the kind of repeated, cross-referenced name-dropping that a language model treats as a signal of relevance.
So the named roster you get back is, functionally, a ranking of marketing reach. The practices that show up in the most places get named. Whether the top-named practice has a stronger medical director than the one that didn't get named is a question the model didn't — and mostly couldn't — evaluate.
This isn't a knock on the models. It's what any system does when the signal it wants isn't there and a correlated proxy is. The proxy for "qualified" becomes "frequently mentioned." And frequently mentioned is a marketing outcome, not a clinical one.
Why this is easy to miss
The uncomfortable part is that the output looks the same either way. A confident list of names reads identically whether it was built on credentials or on aggregator density. There's no asterisk that says "ranked on footprint."
So a practice with a genuinely strong clinical setup — a well-credentialed medical director, the right licenses, a careful supervision structure — can be left off the list entirely, and never know why. The owner assumes the AI just doesn't know about them yet, or that it's random. It isn't random. They're being scored on a footprint they never thought to build, against a signal the AI can't actually read.
And the practice that is getting named often assumes it earned the spot on merit. Maybe it did. But if the win is coming from a broad aggregator presence, that's a very different competitive position than a credential moat — and it can erode the moment a competitor buys more directory placements or runs a bigger review push.
Both practices are operating on a guess about why the list looks the way it does.
You can't reason about this — you have to look at the list
Here's the practical move, and it's simpler than the problem sounds. You can't infer how you're being ranked from first principles. The only way to know what AI is actually saying about your practice is to read the named list — repeatedly, across the assistants people actually use.
Run the queries a real patient would run. "Best med-spa in [your city]." "Where should I get Botox near [neighborhood]." "Med-spa with the best reviews in [metro]." Run them across ChatGPT, Claude, Gemini, Perplexity, and Google AI, because they don't agree with each other and they don't retrieve from the same places. Then look at the actual roster of names that comes back.
You're looking for a few specific things:
- Are you named at all, and in which assistants?
- Who's named alongside you — and are they clinically stronger than you, or just more visible?
- Does your credential story show up anywhere in the reasoning, or is it all "highly rated" and "popular"?
- Does the list shift week to week, which tells you it's tracking something volatile like review volume rather than something stable like qualifications?
That named list is the measurement. It tells you the ground truth of what these systems are saying about your market right now — before you decide whether to do anything about it. Most owners have never actually looked. They've formed a theory about their AI visibility without ever reading the output.
This is the gap LLMClarity exists to close: watching the actual named roster across those five assistants over time, so you can see your share of voice and where you sit relative to the practices getting named. It doesn't change what the AI says. It shows you what's being said — which, given how thin the credential signal is, is usually more surprising than owners expect.
The takeaway isn't that AI recommendations are broken. It's that in a category where the qualification data barely exists online, the named list is measuring reach, not merit — and the only way to find out which one is carrying you is to read the list.