We Asked AI for a Dentist 15 Times. Here's What It Was Actually Reading.
A walkthrough of what ChatGPT actually pulled from when we asked it to recommend a dentist fifteen different ways, and what it means for any local service business.
The Point
AI doesn't have an opinion about which dentist is good. It has a handful of sources it trusts, and it repeats what those sources say. We ran the same basic question fifteen different ways for one city's dentists to see what was actually driving the answer, and the pattern was obvious once you looked at the sources instead of the results.
The Test
We picked a mid-size city and asked ChatGPT variations of "who's a good dentist in [city]" fifteen times over a few days — some phrased as "best dentist near me," some as "dentist for a nervous patient," some as "family dentist with good reviews." Different wording, different angle, same basic ask.
We weren't trying to trick it. We were trying to see what it does when a real person, in a real chat window, tries to find a dentist the normal way.
The Same Few Names, Every Time
Out of fifteen prompts, the same three practices showed up in almost every answer. Call them Practice A, B, and C. A fourth practice showed up about half the time. Everything else was noise — one-off mentions that didn't repeat.
That's the first thing worth sitting with: AI isn't sampling the whole market and giving you a fresh take each time. It's converging on a short list and sticking to it, the same way a front desk person would if you asked them ten different ways to recommend a dentist — they'd give you their two or three go-to answers no matter how you phrased the question.
What It Was Actually Pulling From
When we asked the model to explain its reasoning or pushed it on why it picked those names, the sources it leaned on broke down into a few buckets:
Google Business Profile and review volume. Practice A had the most reviews and the highest average rating of anyone in the set. It came up first almost every time. Review count and rating aren't the only thing that matters, but they're clearly load-bearing.
Directory sites — Healthgrades, Zocdoc, WebMD's provider listings. These showed up as background sources even when the model didn't name them directly. Practices with complete, consistent profiles on these sites got mentioned more. Practices with thin or outdated directory listings didn't, even when their actual reviews were strong.
Local "best of" content. One practice — Practice C — got a boost that didn't track with its review count. Digging in, it had been named in a local news "best dentists in [city]" roundup from a couple years back. That single article was still doing work.
Reddit and forum threads. A couple of answers referenced a local subreddit thread where someone asked for dentist recommendations and got real replies. AI treats that kind of unscripted, multi-person conversation as a strong signal — it reads like independent confirmation, not marketing.
The practice's own website. This mattered less than you'd think for the recommendation itself, but it mattered for the details — hours, insurance accepted, whether they take new patients. If the site didn't have that info clearly stated, the model either left it out or got it wrong.
What Didn't Show Up
The more useful finding was a practice that should have made the list and didn't. Good reviews, been in the neighborhood over a decade, solid reputation by every normal measure. It never came up once in fifteen tries.
We looked at why. Its Google Business Profile was unclaimed and out of date — wrong hours, no recent reviews showing, a phone number that didn't match the website. It had no presence on the directory sites at all. No local press. Nothing on Reddit. As far as the AI could tell, it barely existed. Good service in real life doesn't matter if nothing on the open web backs it up.
The Pattern Underneath
Strip out the dentist framing and the mechanism is the same one that decides which HVAC company, plumber, or electrician gets named when someone asks an AI tool for a recommendation. The model isn't judging quality directly — it's judging what other sources say about quality, and it trusts a narrow set of those sources: your Google profile, a couple of directory sites, any local press you've picked up, and whatever people are saying about you in places like Reddit or Facebook groups.
If those sources are thin, outdated, or contradictory, you're invisible no matter how good the actual work is. If they're solid and consistent, you get named — even against competitors with a bigger ad budget.
This also means there's no trick to it. It's not about gaming the model. It's about making sure the handful of places it actually reads are accurate, current, and say the same thing about your business everywhere.
What To Do With This
Start with your Google Business Profile — claim it, fix the hours, make sure the phone number and service area match your website exactly. Then check the two or three directory sites relevant to your trade. Then see if there's any old local press or "best of" mention you can point back to. That's most of what these tools are reading.
If you want to see exactly what AI is pulling up for your business right now — which sources it's citing, what it's getting wrong, and what's missing entirely — run the free scan at hollar.ai/scan. It shows you the same kind of breakdown we just walked through, but for your own name.