Named for Cleanings, Invisible for Implants: AI Names a Dentist in 89% of Answers, and 94.7% on Only One Platform

AI named a dental practice in 89% of answers. In Dallas, 68% of practices were never the name.

Two measurements look like a contradiction. In April 2026 Local Vitals put 750 patient-style queries to five platforms (ChatGPT, Perplexity, Google AI Overviews, Gemini, Claude) across 10 Florida metros. AI named a dental practice in 89% of them (Local Vitals). Local Vitals sells local-visibility services, so weigh a study finding that AI answers are full of practice names accordingly.

In July 2026 we measured the Dallas metro: 262 practices, 30 patient-style questions, across ChatGPT, Gemini and Perplexity. 177 of the 262, or 68%, were never named in a single answer. The average score was 1.6 out of 100, the best in the metro 27.0, and a typical answer named about six practices (the Dallas dental benchmark). Both results are true: assistants do name dentists, and they name a small, unstable set of them.

Nobody has measured how many patients ask AI for a dentist. The closest figure is healthcare-wide: rater8's June 2026 survey of about 1,000 US adults put the share of patients using AI to find a new provider at 47%, up from 31% in late 2025 (rater8), from a vendor selling review-collection software to medical practices. It was not measured on dental patients.

94.7% of 2,548 practices appeared on exactly one platform. One dentist, in ten metros, appeared on all five.

The same study collected 2,548 recommended dental practices across its 750 queries, and 94.7% of them appeared on just one of the five AI tools. Exactly one dentist, in the whole sample of ten Florida metros, appeared on all five (Local Vitals). Local Vitals sells local-visibility services and the study is not replicated, so hold the decimal point loosely. The shape of the finding is what matters, and it is the one number in this article that changes what an owner should do on Monday.

Because it means there is no such state as "visible in AI." A practice is named by ChatGPT, or by Perplexity, or by Google AI Overviews, and on these numbers those are close to independent events. The dentist who opens one assistant, sees the practice named and closes the laptop has sampled one of five results — and the study says that result repeats nowhere else 94.7% of the time. The reassurance is worth about one fifth of what it feels like.

The disagreement has a mechanical cause. The platforms are reading different source pools. Local Vitals reports ChatGPT pulls 60% to 70% of its local data from Foursquare, that Gemini leans on Google Maps, and that Perplexity cites Yelp and Healthgrades most often on dental queries (Local Vitals) — all vendor claims, none replicated. Take even the direction of that as true and the consequence follows: a corrected Google Business Profile is not a ChatGPT fix, and a strong Yelp presence is not a Gemini asset. Work that pays off in one answer can be invisible in the next four. What does not split by platform is the fact itself: a treatment, a price, a plan, an emergency policy has to be stated somewhere before any engine can repeat it, and the only place you can state it is your own site — which is also where, in our Dallas run, 79% of everything the three engines cited came from (the Dallas dental benchmark).

The single dentist who appeared on all five is not a case study to copy. The study does not report what that practice did, and one observation is not a method — anyone selling you "how the all-five dentist did it" is writing fiction. What the number does establish is that cross-platform presence is rare enough to be a real gap rather than table stakes, in a field where 89% of answers hand out a name to somebody.

So the unit of measurement is not the practice. It is the platform crossed with the question. Four patient question clusters across five platforms is twenty cells, and a single "dentist near me" check on a single assistant fills one of them, then leaves you to assume the other nineteen. That assumption is precisely what the fragmentation number kills, and it is the arithmetic behind the seven-question check further down.

A practice can be named for cleanings and invisible for implants: the four questions you are scored on

"Dentist near me" is the query owners test, and close to the least useful to win. Patients ask at least four different dental questions, and each is a separate contest fed by different sources.

No study breaks AI dental recommendations down by query type, so treat these clusters as a diagnostic frame, not a measured finding. But emergency and implant questions are answered from different pages, directories and review text, and the Local Vitals fragmentation result shows dental visibility does not transfer.

• Emergency. "I chipped a tooth tonight in [city], who can see me tomorrow?" Decided by hours, same-day availability and emergency wording that sits in readable text, not inside a booking widget.
• Insurance. "Dentist in [city] that accepts Delta Dental / Cigna / MetLife." The assistant answers from whatever plan data it can find, and the freshness of that data is not under your control.
• High-ticket procedure. "Best implant dentist in [city]", veneers, Invisalign, sedation. Separate pages, separate competitors, separate review vocabulary. A practice known for family dentistry can be invisible for all of them and never notice: the hygiene schedule stays full while the consults never arrive.
• The anxious or pediatric patient. "Dentist in [city] who is good with kids", "sedation for dental anxiety". Decided by what reviews say in words, not by a star average. "Gentle" and "patient with my son" are things a model can read; a star average is not.

The insurance answer is only as fresh as the record behind it, and nobody has told you which record that is

Ask an assistant which dentist in your city takes Delta Dental and you will get names. What you will not get is a citation to your practice management system: the assistant does not have one. It infers from your own site, insurer pages and directory profiles, which go stale separately — and only the first of those is a record you write.

Local Vitals reports ChatGPT pulls 60% to 70% of its local data from Foursquare and Gemini leans on Google Maps (Local Vitals) — a vendor claim, unreplicated, and neither source is where anyone maintains insurance participation. An assistant saying you are in network on a plan you dropped two years ago is not a bug. It is a stale record read correctly.

The instruction is not "publish an insurance page and this is fixed". Publishing the plan list in readable text is the only way the claim exists in your own words at all, and it is still not a fix, because the stale directory keeps talking. So: the claim is checkable, so check it. A wrong answer turns vague invisibility into a specific error with a named source behind it — a different class of problem, because it has an address.

Ranked by what each source can actually say about you: your own pages first, directories as corroboration

The only dentistry-specific vendor study is one (Local Vitals, April 2026); the other hard numbers below come from healthcare or local business generally, with no dental breakout, and nothing here is a proven dental ranking factor. The order is not by citation volume. It is by what each source can carry: your own site is the only place a fact about the practice — a treatment, a plan you take, who the dentist is, what happens if a tooth breaks tonight — can originate at all, and the directories below either corroborate it or, when they go stale, contradict it. Work it in order anyway; the cheap items stay cheap.

• 1. Your own website. The only source in this list where a fact about the practice can be stated in the first place: the treatments you do today, the plans you take, who the dentist is, what happens if a tooth breaks tonight. It is also, in the one primary measurement anybody has published for dentistry, where the engines went most often. Across the dental answers in our July 2026 Dallas run, of 1,367 cited sources 1,085 (79%) were practices' own websites. The largest third-party platforms trailed far behind — Healthgrades at 61, Zocdoc at 59, Yelp at 45 — and directories and professional sites, taken together, contributed 81 (the Dallas dental benchmark). Dallas, July 2026 — one market, one day, three engines. It is a primary measurement, not a survey: every raw answer is stored, the method is published, and the run can be repeated against any market. No other vendor has published a citation-source measurement for dentistry at all. The nearest third-party figure agrees in direction: Yext's October 2025 analysis of 6.8 million citations across 1.6 million AI answers found Gemini takes 52.15% of its citations from brands' own sites, ChatGPT 48.73% from third-party listings (Yext); Yext sells listings management, so a finding that listings decide everything sells listings management. Do this: one readable page per treatment, plus one for insurance participation, in crawlable text — not inside a booking widget.
• 2. Google Maps and your Google Business Profile, for Gemini. Local Vitals reports Gemini leans on Google Maps; vendor claim. The listing an engine most readily treats as current, which makes it the fastest corroboration of what your own pages already say. Do this: category, hours, and every treatment you sell, named the way a patient says it ("dental implants", not "restorative solutions") and worded the way your own page words it, so the two records agree rather than argue.
• 3. Healthgrades, for Perplexity. Local Vitals names it alongside Yelp as a leading Perplexity source on dental queries; vendor claim, and the largest single third-party source in our own Dallas run at 61 citations (benchmark). Do this: list procedures and accepted insurers, not just a name and phone number, and word them exactly as your own site words them. A fact two independent sources state identically is the closest thing to verification an engine has.
• 4. Zocdoc and Yelp, for Perplexity. Local Vitals reports Perplexity cites Yelp most often on dental queries; in our Dallas run Zocdoc drew 59 citations and Yelp 45 (benchmark). Yelp's general weight: 512,680 citations across 28 million AI answers in Q4 2025, more than BBB, Angi, Thumbtack, HomeAdvisor and Nextdoor combined (Foundation Inc. with AirOps, a content agency and an AI-search vendor, no dental cut). Yelp is also the largest body of text in which a claim of yours can be contradicted. Do this: claim both; keep the service list matching your own pages; reply within the HIPAA limits below.
• 5. Foursquare, for ChatGPT. Local Vitals reports ChatGPT pulls 60% to 70% of its local data from Foursquare (Local Vitals, 750 queries): a vendor claim, unreplicated, and the record almost no dental practice has looked at. Note that a source can feed a model without being cited by name, so the claim is not testable from a citation count, ours included. Fifth because all it can do is repeat or contradict what you have already published: a stale address or a category you outgrew argues with your own site. Do this: fix the address, hours and category.
• 6. Dental-association and niche provider directories: plausible, unproven. The claim that provider directories decide these answers rests on Zocdoc leading provider queries in Perplexity, ahead of Healthgrades and Vitals (rater8 on the Yext data) — rater8 sells review-collection software to medical practices and the underlying analysis is Yext's, a listings-management vendor, so both parties profit from a finding that profiles decide the answer. Measured across healthcare, not dentistry. No study confirms a dental-association find-a-dentist directory as an AI source. Do this: complete the free ones so they repeat what your site already says; log them as a hypothesis, not a channel. A profile can confirm a fact or contradict it; it cannot state one you never published — and in our own Dallas run, directories and professional sites together drew 81 citations out of 1,367 (benchmark).

Seven questions a patient types, and a 15-minute check that covers all four clusters

Any dentist can run this free, this week. Open ChatGPT, Gemini and Perplexity, use a fresh chat per question, paste the seven below with your city in the brackets, and write down every practice named. That is 21 answers. Count how many name you, note which cluster the misses fall into, and check accuracy too: ask each assistant which insurers it thinks you accept.

One pass is one moment: the same question asked twice returns different practices, and a notepad cannot track a competitor's share over a quarter. It is free, and a bad snapshot is still a finding.

• "I chipped a tooth tonight in [city]. Which dentist can see me tomorrow?"
• "Which dentist in [city] accepts Delta Dental and is taking new patients?"
• "Who is the best implant dentist in [city]?"
• "Veneers or Invisalign in [city]: which dentists do both, and what do they charge?"
• "My son is terrified of the dentist. Who in [city] is good with anxious kids?"
• "Dentist in [city] that offers sedation for dental anxiety."
• "Is [your practice name] in [city] any good, and what do they do?" That shows what the engines already believe about you.

Elite Dental, New Vision Dental and U. Phillip Igbinadolor: three review-reply penalties, all of them dental

Reviews with real text are what an assistant reads, and owner replies are what patients read. Three of the enforcement actions the Office for Civil Rights has published over review replies landed on dental practices — we make no claim about how many others exist. The Office for Civil Rights, which enforces HIPAA federally in the United States, settled with Elite Dental Associates for $10,000 in 2019 (HHS) and New Vision Dental for $23,000 in 2022 (HHS), and imposed a $50,000 civil money penalty on U. Phillip Igbinadolor, D.M.D. & Associates in 2022 (HHS). Each involved patient information disclosed in a reply to an online review.

Whether HIPAA reaches your practice turns on whether it is a covered entity, and penalties turn on the facts and the harm, not a price list. This is not advice about your situation: confirm your status with counsel. The mechanics of a compliant reply are in the HIPAA review trap for chiropractors.

The second constraint is professional, not federal. The ADA's 2025 Code update addresses social media and influencer claims: they must not be "false or misleading in a material respect", and paid partnerships must be disclosed (ADA News, February 24, 2025). The ADA Code is an association code, not a statute; the rules that bind you come from your state dental board and vary by state, so check yours or ask counsel. Conveniently, the specific, sourced content an engine rewards is what a board is least likely to object to.

Six things dentists say when we show them this, including the ones that go against us

These come up in nearly every conversation with an owner.

• "68% invisible just means the field is wide open." Half true. In Dallas the best practice scored 27.0 out of 100 (benchmark), so nobody has locked this down. But the 89% figure says the answers are not empty: six names get handed out per answer, and they go to somebody.
• "I'm booked out three weeks. Why do I care?" For cleanings, you may not. The exposure sits where patients research before calling: implants, veneers, Invisalign, sedation. A full hygiene schedule says nothing about those.
• "My marketing agency handles this." Ask which of the five platforms they checked, and what the other four said. On the Local Vitals numbers, a one-platform report describes a result that 94.7% of the time repeats nowhere else.
• "The models change every month, so what is the point." They do, which argues for measuring repeatedly, not for skipping it. Nobody can promise a position in an AI answer; there is no #1 to buy. What you can know is whether you are named, where, and for which question.
• "Is running these checks a HIPAA problem?" No. Asking a public assistant which dentists it recommends involves no patient and no protected health information. The exposure runs the other way, into what you write back to a reviewer.
• "Will fixing my Foursquare listing get me recommended?" Unknown, and we will not pretend otherwise. The 60% to 70% claim is one vendor's and unreplicated, and sitting in the source pool is not the same as being picked from it. An hour's work and a bet, not a fix. A listing can confirm a fact or contradict it; it cannot state one your own site has never made.

Six names per answer, sixty-seven practices in the city: which measurement do you have?

A dental practice has Google Search Console for search traffic and nothing for the answers assistants give about it. You cannot see whether ChatGPT tells a patient you take their plan, or which practice Gemini names when a parent asks who is gentle with a frightened child.

Start from a measurement rather than a purchase. RecoSignal's free AI Visibility Snapshot puts patient-style questions to ChatGPT, Gemini and Perplexity, then reports how often your practice is named, which practices are named instead, and the sources those answers lean on. It reads public answers and touches no patient data. For what moves the odds, see how to improve AI visibility; for why a strong Google position does not carry over, see AI visibility versus local SEO. It shifts probability, not certainty.