There isn't a reliable pattern to copy from the handful AI does name: 3 practices in ChatGPT, 18 in Perplexity, too few to generalize from responsibly. No positive result was recorded for 507 of 527, 96.2% of the sample; this includes failed checks and is not a confirmed non-recommendation rate. Across the full sample, high ratings and open robots.txt files were common. The infrastructure findings also describe the full sample: 53.9% have no LocalBusiness or Dentist schema on their site at all, and only 3.4% have both tested schema types.
Why not just describe what gets a dental practice recommended?
Because there aren't enough winners to describe one. ChatGPT named 3 of the 415 practices it returned a reading for, and Perplexity named 18 of 297. A pattern built from 3 or 18 data points isn't a pattern. It's a guess wearing a data costume.
Those two denominators are smaller than our 527-practice sample for an unglamorous reason: the engine check failed on the rest. ChatGPT returned a rate-limit error for 112 practices and Perplexity an authentication error for 230, so we dropped those from the engine rows instead of filing them as "not recommended". An error is a missing reading, not a finding.
What we can do honestly is flip the question. Instead of inferring a winning formula from a few named practices, we can describe infrastructure across the full sample. These totals do not establish what distinguishes named from unnamed practices. If 53.9% of 527 practices are missing the same basic piece of infrastructure, that's worth knowing even without a clean before-and-after comparison, because it describes the ground more than half of the practices we scanned are standing on.
The full method behind these numbers, including how we handle scans that come back as an error rather than a real reading, is in how we measure AI visibility. City-level versions of the same scan are in our AI visibility report for Charlotte dentists.
| Signal | Measured, June 2026 | What it suggests |
|---|---|---|
| Star rating 4.3 or higher | 506 of 523 (96.7%) | High ratings are common; causation was not tested |
| robots.txt open to AI crawlers | 492 of 527 (93.4%) | robots.txt permission alone does not establish live access |
| No LocalBusiness/Dentist schema at all | 284 of 527 (53.9%) | A common infrastructure gap, not a proven cause |
| LocalBusiness present, FAQPage missing | 225 of 527 (42.7%) | Partial machine-readable description |
| Both present | 18 of 527 (3.4%) | Both tested schema types present |
Is it their reviews?
This scan does not isolate their effect. 96.7% of the practices we scanned, 506 of the 523 with a rating on file, already sit at 4.3 stars or higher, the level ChatGPT-recommended locations averaged in SOCi's 2026 index. That average is not an entry threshold and does not establish whether reviews influence recommendations.
The rounded 97% describes how common that rating level is in this sample. It does not compare the 3 or 18 named practices with the rest while controlling for other factors, and it cannot establish a review threshold.
Is their website blocking AI crawlers?
For most, no. 93.4% of the practices we scanned, 492 of 527, have a robots.txt file open to AI crawlers. Only 15 practices (2.8%) were actively blocking one or more AI bots, and 20 (3.8%) had no readable robots.txt at all.
Open yoursite.com/robots.txt and check rules for the relevant search crawler, including OAI-SearchBot or PerplexityBot, and the paths they cover. OpenAI distinguishes OAI-SearchBot for search from GPTBot for training; blocking GPTBot alone does not block ChatGPT search. A robots.txt check also does not establish whether the server actually serves the page to that crawler. A block on your own site does not remove information about your practice from other websites. A full breakdown of how blocks happen, including sites that look open but quietly refuse AI bots anyway, is in are dental websites blocking AI crawlers?
Source: OpenAI bot roles.
So what is common across the scanned practices?
A common gap is missing business schema, not an absence of all machine-readable information. 53.9% of the 527 practices we scanned have no LocalBusiness or Dentist schema markup at all, and only 3.4% have both LocalBusiness and FAQPage schema in place. Those are whole-sample infrastructure counts, not a comparison that isolates why particular practices were recommended.
Schema markup isn't a switch that turns citations on. In 2026, Mark Williams-Cook hid a made-up address inside deliberately invalid schema on a test page, with nothing matching in the visible text, and both ChatGPT and Perplexity still repeated the fake address back. The address was in the embedded JSON-LD, not the visible copy. The test shows that the assistants could extract it despite invalid Schema.org vocabulary; it does not prove they ignore schema entirely or rule out a schema-aware system falling back to text extraction.
What schema does is state basic facts, what a business is, where it is, what it treats, in a form a machine can parse without guessing from prose. A practice with none of that is asking every engine that visits to reconstruct who it is from whatever text happens to be on the page. Clear business facts in visible text remain useful to readers whether or not schema is present. This scan did not test whether schema changes retrieval, accuracy or recommendations, so it cannot attribute missing recommendations to missing markup. Watch a walkthrough of the schema findings on video, with the four-question audit explained step by step.
Source: Mark Williams-Cook, 2026; Citevio scan, June 2026.
| Surface | Locations recommended nationally | Our June 2026 scan |
|---|---|---|
| ChatGPT | 1.2% | 3 of the 415 we could read (0.7%) |
| Perplexity | 7.4% | 18 of the 297 we could read (6.1%) |
| Gemini | 11% | Not measured |
| Google local 3-pack | 35.9% | Not measured |
Our recorded rate sits below SOCi's national share of locations recommended on both engines. These studies use different populations and query methods, so the comparison cannot explain the gap or attribute it to schema, third-party mentions or crawl history. The ChatGPT and Perplexity rows above already show one engine recommending a practice far more often than another; how often AI engines agree with each other looks at that gap directly.
One limit worth stating on the Perplexity row: 3 of those 18 come from a Columbus batch where the city label inside the query was malformed, so the engine was answering a slightly different question than we intended. They are reported rather than dropped because a reading is kept or excluded on a written rule set before the scan, not on how it reads afterwards; the number should be taken as 15 clean plus 3 qualified rather than a flat 18. What that batch does and does not support is written out in the AI visibility report for Columbus dentists.
The dataset supporting the comparison above is preserved under DOI 10.5281/zenodo.23143214.
Does fixing my schema guarantee I'll get recommended?
No, and no page should promise that. What the data supports is narrower: schema is foundational infrastructure that most of the practices in this scan are missing, not a lever that reliably produces a citation. This scan does not isolate the influence of mentions elsewhere, crawl history or page text.
We cover the mentions side of this in more depth in why isn't my dental practice showing up in ChatGPT, and the step-by-step fixes in getting your practice recommended by ChatGPT.
When your details disagree across the web, which version does an engine repeat?
The Google help page cited here does not specify a conflict-resolution rule for these sources. Your practice's information does not sit in one place — Google says profile information is compiled from crawled web content, licensed third-party data, user contributions and Google's own interactions with the business. That document describes Business Profiles, not how every AI assistant chooses an answer. Our recommendation is to correct conflicting details to reduce opportunities for errors; this is not evidence of a ranking effect.
Source: Google Business Profile Help, how Google sources and uses information in Business Profiles, read 18 August 2026. That page documents the sources; it does not state how conflicts between them are resolved.
Our own scan gives a sense of how much of this surface is even readable, and the honest headline is: less than you would expect. Directory presence was checked on two services, and on both the number of practices we could get a reliable reading for was far below the full sample.
| Directory | Practices with a usable reading | Listing in order | Incomplete or missing |
|---|---|---|---|
| Bing Places | 291 of 527 | 256 (88.0%) | 35 incomplete (12.0%) |
| Foursquare | 146 of 527 | 66 (45.2%) | 47 incomplete (32.2%), 33 missing (22.6%) |
What this does not show is that inconsistent details are why an engine skipped a practice. No such measurement exists here: nothing was corrected and re-measured, and no practice was tracked before and after a fix. Anyone presenting a consistency audit as the reason you are invisible is selling a correlation they have not established, and the honest version of the argument is duller — information gets compiled from sources that can disagree, the cited Google help page does not specify a conflict-resolution rule, so fewer disagreements means fewer ways for the wrong version to travel.
Where this stops being about consistency and starts being about an engine stating something false about your practice — wrong hours, a treatment you do not offer, a provider who left — that is a different problem with a different measurement behind it, and it is covered in what AI gets wrong about you. The Google Business Profile side, including the limits of the cited Google documentation, is in where your Business Profile fits. The directory counts above are in the downloadable Citevio dataset.
Citevio's dental market research is published openly under CC-BY-4.0 with a DOI, so the numbers can be checked and reused; client data is never published.
Crawler access and schema both leave a trace on your own site, and the checker below reads it in under two minutes.
Prefer a deeper look? The full AI visibility checker for dentists tests what ChatGPT, Perplexity, Gemini and Google AI Overviews actually say about you.
One reminder before you go. Ahrefs' correlation work across 75,000 brands found mentions of a brand on other sites tracking AI Overview appearances at r=0.664, against 0.218 for links pointing at the site, a sign that what's on your own site is one part of the picture, not the whole one. And any snapshot like this one ages: Profound's historical June-to-July comparison found that roughly 40-60% of domains in the later answers were absent from the earlier answers for identical prompts, which is why this page carries a scan date instead of a number we treat as permanent.
Sources: Ahrefs, May 2025; Profound, July 2025.
Common questions
Does my star rating affect whether AI recommends my practice?
Not on its own. 96.7% of the practices in our June 2026 scan that had a Google rating on file, 506 of 523, already sit at 4.3 stars or higher, the level ChatGPT-recommended locations averaged in SOCi's 2026 index. That is an average, not an entry threshold; this scan does not isolate whether reviews influence recommendations.
Is my website blocking AI crawlers without me knowing?
Check the relevant search crawler and page path rather than assuming that all bot rules have the same effect. In our scan, 93.4% of practices had robots.txt open to AI crawlers and only 2.8% were actively blocking one. Open yoursite.com/robots.txt and check OAI-SearchBot or PerplexityBot rules. GPTBot concerns training, not ChatGPT search; an open file alone does not establish live access.
If I add schema markup, will ChatGPT start recommending me?
There's no guarantee, and any claim otherwise is worth doubting. What we can say is that 53.9% of the 527 dental practices we scanned have no LocalBusiness or Dentist schema at all, so most practices are missing this basic layer of machine-readable description. It's foundational, not a lever with a promised payoff.
How many dental practices does this data cover?
527. The raw diagnostic set held 611 unique website scans, and 84 of those were not dental practices at all, such as retailers, salons, dental laboratories and plastic surgery practices, so they were removed before anything was counted. The remaining 527 are spread across 7 US metros, Charlotte, Austin, Raleigh, Nashville, Columbus, Tampa and Salt Lake City, plus 9 smaller towns, scanned June 6-26, 2026.
Out of the 527 practices, how many were actively blocking AI crawlers?
Very few. 15 of the 527 practices (2.8%) were actively blocking one or more AI bots in robots.txt, and 20 (3.8%) had no readable robots.txt at all. Combined with the fact that 93.4% had an open robots.txt, these robots.txt results alone do not establish live access or explain why practices go unnamed.
What share of the scanned practices have complete schema markup versus none at all?
The largest group has no business schema: 53.9% (284 of 527) have no LocalBusiness or Dentist schema at all, 42.7% (225 of 527) have LocalBusiness present but are missing FAQPage schema, and only 3.4% (18 of 527) have both in place. Having both tested schema types is rare across the sample; that is not a test of every form of machine-readable information.
How does your scan's recommendation rate compare to SOCi's location-level rate?
Lower on both engines we measured. SOCi's 2026 index found ChatGPT recommended 1.2% of locations nationally and Perplexity 7.4%; in our 527-practice scan, ChatGPT named 3 of the 415 practices it could read (0.7%) and Perplexity named 18 of the 297 it could read (6.1%). Gemini and Google's local 3-pack weren't measured in our scan, so no comparison exists for those.
Why is your recommendation rate lower than SOCi's location-level rate?
We cannot explain the gap from these studies. SOCi measured locations of multi-location brands nationally; our scan used dental practices and a different query method. The 53.9% missing-business-schema figure describes our sample, not SOCi's locations, and does not establish a cause of the difference.
Is the Columbus figure in your Perplexity count fully reliable?
Treat it as directional rather than exact. 3 of the 18 practices Perplexity named come from a Columbus batch where the city label inside the query was malformed, so the engine was answering a slightly different question than intended. We report the number as 15 clean plus 3 qualified rather than a flat 18, because a reading is kept or excluded on a rule set fixed before the scan, not on how convenient it looks afterward.
When my business details disagree across different websites, which version does an AI engine repeat?
The Google help page cited here lists Business Profile sources: crawled web content, licensed third-party data, user contributions and Google's interactions with the business. That page does not specify which source wins a conflict or how every AI assistant chooses an answer. Correcting conflicting details is our error-reduction recommendation, not a measured ranking intervention.
How many of the scanned practices have a complete Bing Places listing?
We could get a usable reading for 291 of the 527 practices. Of those, 256 (88.0%) had their listing in order and 35 (12.0%) were incomplete. These listing counts do not establish which sources ChatGPT used for an answer. The broader diagnosis is covered on why isn't my dental practice showing up in ChatGPT.
What about Foursquare listings?
Coverage was much thinner. We could get a usable reading for only 146 of the 527 practices; of those, 66 (45.2%) had their listing in order, 47 (32.2%) were incomplete and 33 (22.6%) were missing entirely. Bing Places and Foursquare returned usable readings for different numbers of practices, so the two rates can't be compared directly to each other.
Does inconsistent business information explain why AI skipped a practice?
We don't have a measurement that shows that. Nothing was corrected and then re-measured, and no practice was tracked before and after a fix, so there's no before-and-after evidence linking inconsistent details to being skipped. What we can say is duller but honest: information gets compiled from sources that can disagree, the cited Google help page does not specify a conflict-resolution rule, so fewer disagreements means fewer chances for the wrong version to travel.