Original research · September 2026
We asked four AI engines to recommend a med spa. Here is what they actually cited.
We asked the four AI tools people actually use (Perplexity, ChatGPT, Gemini, and Google AI Mode) to recommend a med spa in Salt Lake City, four ways each, and recorded every business named and every source cited. 16 answers, a 25-business field, and a screenshot of every one. All figures below are measured.
different sources were cited across just 16 AI answers. No single one appeared in even 40%. There is no one place to "win" AI recommendations. The engines assemble each answer from a scattered mix of directories, business websites, and Google's own map data.
Key findings
Five things the data shows.
- 01
Across the 16 answers we counted 30 different sources. No single one appeared in even 40% of them. The most-cited source, a business's own website, showed up in just 6 of 16. AI builds a local recommendation from a fragmented web, not one authority.
- 02
Third-party directories are the connective tissue. MedSpaScout, Medical Spa Locator, ClinicCompass, GlowUpFinder, AllMedSpas, and ThreeBestRated were the pages AI leaned on to decide who counts as "top." Being listed on the right directories is not optional.
- 03
The engines behave very differently. Gemini cited only Google's own Maps data, zero third-party web pages across all four prompts. ChatGPT leaned hardest on directories. Perplexity cited the widest mix. Google AI Mode led with the top pick's own website.
- 04
Phrasing changes the sources. A "best" query pulled the businesses' own sites, while a "top / who are the top" query pulled third-party directories. The same business needs both to be covered.
- 05
Remove the branded probe and the most-recommended name was Nuance Aesthetics + Wellness (10 of 12 discovery answers), even though it did not have the most reviews in the cohort. Raw review count is not what decides the AI's pick.
The sources AI actually cites
When AI recommends a local business, these are the sources behind the answer.
Share of the 16 AI answers that cited each source. The bars are short on purpose: the citation layer is fragmented, and directories you may never have heard of carry as much weight as the businesses' own sites.
Who got recommended
A few names show up again and again.
Share of the 12 discovery answers (the three "recommend / top / which providers" prompts, across all four engines) in which each business was named. The leader was not the business with the most reviews.
Methodology
How we ran this, in plain terms.
We built a field of 25 real Salt Lake City med spas, then asked each of 4 AI engines (Perplexity, ChatGPT, Gemini, and Google AI Mode) the same four fixed prompts in September 2026: three ways of asking for a recommendation, plus one branded probe about a single named business. That is 16 recorded answers.
For every answer we recorded exactly which businesses were named, in what order, and which web sources the engine cited, then saved a screenshot. The branded probe is kept separate from the discovery numbers because it only ever names the one business we asked about.
Limits worth stating plainly: AI answers vary from run to run and by account history, and two of the four engines ran in a signed-in session. This is one dated snapshot of one category in one city, not a universal law. We report it exactly as measured, and we hold our own work to the same standard we ask of yours.
What this means for your business
Being cite-able is the new being findable.
There is no single lever. To get recommended by AI you have to be present in the directories these engines pull from, keep your Google Business Profile strong (Gemini used nothing else), and make your own site readable by AI so it can be quoted directly. That is exactly what we do. See how AI visibility works.