Cleareo research · 160 AI answers · 20 US cities · November 2026

We asked AI for the best dentist in town. ChatGPT named 1.7, on average.

In November 2026 we asked ChatGPT, Gemini, Claude and Perplexity “Who is the best dentist in [city]?” for 20 US cities, twice each. Here’s who they named, what they read to decide, and how to ask the same question about your own town.

  • Free
  • In your inbox in 2 business days
  • No card

Patients see a short list. On ChatGPT, it’s one or two names.

ChatGPT usually named one or two dentists per answer. In Austin and Portland it named exactly one. Claude named about ten. Which tool your patient opens decides how many seats there are, and there are rarely enough for everyone.

Read the full study

Practices named per answer

Average across 40 answers each
  • ChatGPT1.7 practices named per answer
  • Perplexity2.4 practices named per answer
  • Gemini5.2 practices named per answer
  • Claude10.1 practices named per answer
Each square = one practice named

Asked through each tool’s API with web search on and a US location set, 8–9 Nov 2026. API answers can differ from the consumer apps. 40 answers per tool: 20 cities, asked twice.

Four findings behind the short list. Each one drawn to scale.

The tools disagree with each other, change their minds when asked again, and read very different parts of the web. One thing they all bring up is insurance.

Agreement

The four tools almost never named the same dentist.

Across the 20 cities the tools named 489 different practices. 425 of them came from just one tool. Only 5 were named by all four.

Being on Gemini’s list tells you almost nothing about ChatGPT’s. You have to check each one.

Practices counted once per city, matched on name. Two branches of one group may count twice.

489 practices named

One square per practice
  • 425 named by one tool only
  • 51 named by two tools
  • 8 named by three tools
  • 5 named by all four

Repeat stability

Ask ChatGPT again and the list changes.

We asked every question twice, a few minutes apart. ChatGPT kept only 29% of the same practices, and its first pick in 7 of 20 cities. Perplexity was the steadiest, at 67%.

One check on one day isn’t enough. That’s why we ask every question more than once.

Repeat overlap: practices named in both runs ÷ practices named in either run, averaged across the 20 cities. Runs a few minutes apart.

Practices kept when asked again

Same question, minutes apart
  • ChatGPTfirst pick kept 7/2029%
  • Geminifirst pick kept 12/2037%
  • Claudefirst pick kept 8/2038%
  • Perplexityfirst pick kept 19/2067%

Source mix

Each tool reads a different web.

31% of ChatGPT’s links went to Reddit. Practice websites were 46% of Gemini’s links, but only 9% of ChatGPT’s. Claude cited Yelp in all 40 of its answers; the other three never linked to it.

So fixing your own website alone won’t get you named. The listings, lists and threads around you count as much.

Each link counted once per answer, including search results the tool returned and the links it footnoted. Domains sorted into six types with one shared set of rules, checked by hand.

Share of source links
  • Practice websites
  • Reddit and forums
  • Directories and reviews
  • Ranking sites
  • Local press
  • Other
  • ChatGPTPractice websites 9%, Reddit and forums 31%, Directories and reviews 15%, Ranking sites 11%, Local press 5%, Other 29%
  • ClaudePractice websites 14%, Reddit and forums 0%, Directories and reviews 55%, Ranking sites 16%, Local press 2%, Other 13%
  • GeminiPractice websites 46%, Reddit and forums 11%, Directories and reviews 24%, Ranking sites 11%, Local press 4%, Other 4%
  • PerplexityPractice websites 14%, Reddit and forums 0%, Directories and reviews 49%, Ranking sites 28%, Local press 7%, Other 3%

Claude answers that cited Yelp

One square per answer
  • 40 of 40 Claude answers cited Yelp. ChatGPT, Gemini, Perplexity: 0.

Insurance

Nearly every answer brought up insurance.

159 of the 160 answers mentioned insurance, from all four tools. It was the one topic that came up almost every time.

If your site, Google profile and directory listings don’t say which plans you accept, the AI has nothing to repeat.

Any mention of insurance, insurers or accepted plans in the answer text, all four tools.

Answers that mentioned insurance

One square per answer
  • 159 mentioned insurance
  • 1 didn’t

Getting named is about what the AI can read.

The practices that showed up were the ones that existed, with consistent details, on the pages each tool likes to read. Most of those pages are ones you can fix.

  • Your website

    Every tool reads it, some far more than others

    Practice websites were 46% of Gemini’s links and 9% of ChatGPT’s. Pages that say plainly who you are, which dentists work there, what you treat and which plans you take give the AI something to repeat.

  • Directories

    Your profiles are read, claimed or not

    Health directories and review sites were 55% of Claude’s links and 49% of Perplexity’s. If a directory lists a dentist who left, or no insurance, that’s what an AI may say.

  • Yelp and reviews

    Ratings get quoted with counts

    Claude cited Yelp in all 40 answers. “4.9 stars from 182 reviews” is the kind of line that makes it into an answer, usually copied from a directory or ranking page.

  • Reddit and lists

    Local threads and “best of” lists count

    31% of ChatGPT’s links went to Reddit. Ranking sites and city magazine lists were read too. You can’t plant posts, but you can see what they say about you.

Real answers from four of the cities. Then ask the same question.

The first run of each tool, with the practices in the order named. These are public businesses the tools happened to name, not Cleareo clients. Your answer today may differ from ours.

Austin, TX

Four tools, mostly different answers

ChatGPT and Perplexity both started with Blunn Creek Family Dentistry. Gemini and Claude both named Forest Family Dentistry. Past that, the lists barely overlap.

The question“Who is the best dentist in Austin, TX?”

Madison, WI

The rare city where all four agreed

All four tools named Artisan Dental first. Across the 20 cities, only 5 practices were named by all four, and this is one of them.

The question“Who is the best dentist in Madison, WI?”

Plano, TX

Two tools, no names in common

ChatGPT named three practices and Perplexity named four. Not one appears on both lists.

The question“Who is the best dentist in Plano, TX?”

Portland, OR

One pick each, and not the same one

Both tools gave a single name. A patient using ChatGPT and a patient using Perplexity would call different practices.

The question“Who is the best dentist in Portland, OR?”

A small snapshot, not a census.

We publish the method so you can judge the numbers, and repeat the test.

How we ran it

Question
“Who is the best dentist in [city]?”, one phrasing
Cities
20 mid-size US cities, from Austin to Tampa
Tools
ChatGPT, Claude, Gemini and Perplexity, through each tool’s API with web search on and a US location set
Runs
Each question twice per tool, a few minutes apart: 40 answers per tool, 160 in total
Dates
Gemini and Perplexity on 8 Nov 2026; Claude and ChatGPT on 9 Nov 2026
Counting
A second model pass pulled out every practice named, checked by hand on samples. Each source URL counted once per answer

What it can’t tell you

  • API, not the apps. Our earlier tests found API answers differ from what people see in the consumer apps. ChatGPT ran on a cheaper API model than the app may use.
  • Two days, two runs. US only, mid-size cities only, dentists only.
  • One phrasing. “Family dentist near me” or “dentist that takes Delta Dental” would pull different results.
  • Imperfect name matching. Branches of one group may be counted separately.
  • Google’s AI Overviews and AI Mode were not tested.

The full study: tables by city, top sources, and how to cite it

That was 20 cities. Now see yours.

Tell me your practice and town. I ask ChatGPT, Google’s AI, Claude and Perplexity five questions your patients ask, twice each, and email you who they named.

Get my free check
  • Free, no obligation
  • No card
  • In your inbox in 2 business days