- 0:00 The short answer: a label, not a lever
- 0:18 The field data: 284 of 527 had no schema at all
- 0:43 Only 18 of 527 were complete
- 1:28 What schema actually is, and when it helps
- 2:21 Why AI placement cannot be bought
- 2:38 The false address experiment
- 3:26 Audit 1: read the live output, not the plugin screen
- 3:48 Audit 2: do the fields match the visible page?
- 4:08 Audit 3: coverage, and naming the right subject
- 4:52 Audit 4: who owns the block when the page changes
- 5:32 Implementation is not outcome: two separate scoreboards
- 6:18 What a clean before-and-after test needs
- 6:55 What a practice should do
- 7:16 What a practice should refuse
- 7:34 The measured conclusion
- 7:54 Where the open data lives
What is the finding, in one paragraph?
Schema markup is a structured label attached to a page, not a way to buy visibility. In Citevio's dental practice web anatomy study of 527 US dental practice websites across seven US metros plus nine smaller nearby towns, 6-26 June 2026, 284 sites, 53.9%, had no LocalBusiness or Dentist schema at all. Another 225, 42.7%, had LocalBusiness present but no FAQPage. Only 18, 3.4%, had the complete measured configuration: both LocalBusiness and FAQPage together. This counts how common each implementation state was; it does not show that missing markup caused a practice to be absent from an AI answer, or that complete markup caused one to be named.
What did the measured result show?
Among 527 US dental practice websites scanned 6-26 June 2026, more than half, 53.9%, had no LocalBusiness or Dentist schema at all. Another 42.7% had LocalBusiness present but were missing FAQPage. Only 3.4%, 18 sites, had both together, the complete measured configuration used in this study. Raw records were 611; 84 sites with no dental signal were dropped, leaving the published denominator of 527.
| Schema state | Sites | Share of the 527 |
|---|---|---|
| No LocalBusiness or Dentist schema at all | 284 | 53.9% |
| LocalBusiness present, no FAQPage | 225 | 42.7% |
| Both LocalBusiness and FAQPage present | 18 | 3.4% |
Source: Citevio dental practice web anatomy study, 6-26 June 2026, n=527. Full methodology, per-metro breakdown and raw CSV at citevio.com/data.
What does schema markup actually do, and when does it help?
Schema is a structured label that can state a business's type, name, address and relationships in explicit fields, so a machine receives a clean declaration instead of rebuilding the identity from marketing copy. It cannot establish that an outside system will trust, retrieve or cite the claim.
It becomes most useful when a homepage lists several services, locations and brand phrases but never gives one plain statement of identity. Structured data can connect those facts and help another system distinguish which fact belongs to which entity. Think of the markup as a compact claim sheet whose entries must be supported by the page a visitor can read. It can reduce ambiguity inside the document; it cannot buy a citation, force a recommendation, or guarantee that an assistant says your name. There is no paid slot in an AI recommendation.
Can an AI assistant repeat wrong information from invalid schema markup?
Yes, in at least one outside test. Researcher Mark Williams-Cook placed a false address inside deliberately invalid structured data on a test page whose visible text carried no address at all, and both ChatGPT and Perplexity repeated the false address. His reading was that the assistants lifted the text straight out of the HTML without parsing it as schema, so the markup being invalid was not what decided whether the fact was used.
That experiment does not show schema has no value; it shows that "schema makes AI cite you" is too strong a claim. A single test page is also different from Citevio's field audit, and the two pieces of evidence should be kept separate. One marks a limit on what schema can promise; it is not a reason to abandon the infrastructure.
Source: Mark Williams-Cook, 2026.
How do I audit my own dental practice website's schema markup?
Run four checks against the live page, not the plugin settings screen. Each one asks a different question, and a page can pass one while failing another.
- Audit 1 — Read the live output, not the plugin screen. Ask for the rendered output or use a validation tool on the public URL. Do not accept a plugin toggle as proof; the thing an external machine receives is the live output.
- Audit 2 — Do the fields match the visible page? A clean schema block with an old address, wrong business type, or invented service is not an improvement. It can simply become a clean version of the wrong fact.
- Audit 3 — Coverage, and naming the right subject. Check whether type, name, address, telephone and page relationships are explicit and consistent, and whether the block names a clinic, a clinician or a service correctly — the three are related but not interchangeable.
- Audit 4 — Who owns the block when the page changes? Someone must know which system generates the markup and what happens when the visible page changes. An unowned block ages quietly and can start disagreeing with the page it describes.
This is implementation only: we did not measure whether adding, fixing or completing schema causes an assistant to name, cite or recommend a practice, nor how long any change would take to appear. It is not an intervention: this audit did not hold a question set, engines, wording and run rules fixed across a before-and-after comparison, so nothing here supports a causal claim. We did not measure which schema type, if any, assistants prefer for dental practices. And this is a dated snapshot of one regional sample, not a live feed and not a national figure.
Implementation and outcome are two separate scoreboards. The implementation check is binary and inspectable: the markup exists, validates, and matches the page, or it does not. The outcome check is different: the practice is named, cited, or absent on a recorded assistant run. A clean implementation can coexist with a missing name — that coexistence is not a contradiction, it is two different questions. Never use success in the first check as a substitute for observing the second.
What is the practical conclusion?
Make the live markup accurate, explicit, and consistent with the visible page, keep a copy of the before state, then measure assistant outcomes separately under a fixed protocol. Refuse any promise that adding one block of code will produce a named mention, ranking, patient, or deadline — that is speaking beyond what this study measured.
The measured conclusion is modest. Most sites in the sample lacked the measured machine-readable identity, and very few had the complete measured state during 6-26 June 2026. That is a real implementation gap worth inspecting. It is not, on its own, an outcome effect.
How did we measure this?
In Citevio's dental practice web anatomy study, 527 US dental practice websites across seven US metros plus nine smaller nearby towns were scanned 6-26 June 2026. Each site was checked for LocalBusiness or Dentist schema and for FAQPage. "Complete" means both were present; "incomplete" means LocalBusiness present and FAQPage missing. Raw records were 611; 84 sites with no dental signal were dropped, leaving the published denominator of 527. The aggregated counts, per-metro breakdown, method and known limitations are published at citevio.com/data. No practice name or domain is published in the aggregate package, and no client data is included.
Common questions
Does adding schema markup make ChatGPT or another AI assistant recommend my dental practice?
No. Schema markup can make your practice easier to describe, but it cannot buy a citation, force a recommendation, or guarantee that an assistant says your name. Citevio's web anatomy study measured how common each implementation state was, not whether adding or completing schema changes an AI assistant's answer.
How many dental practice websites actually have schema markup?
In Citevio's dental practice web anatomy study of 527 US dental practice websites, 6-26 June 2026, 284, 53.9%, had no LocalBusiness or Dentist schema at all. Another 225, 42.7%, had LocalBusiness present but no FAQPage. Only 18, 3.4%, had both.
How do I audit my own dental practice website's schema markup?
Ask four questions. What structured data is actually in the live page source, not the plugin settings screen? Do the fields match the visible page, or is it a clean version of a wrong fact? Does coverage include the practice identity, type, name, address, telephone, page relationships, and does it name the right subject? And who owns the block when the page changes?
What exactly counts as "complete" schema in this study?
Complete means a site had both LocalBusiness or Dentist schema and FAQPage present together; only 18 of 527 sites, 3.4%, met that bar. Incomplete means LocalBusiness or Dentist schema was present but FAQPage was missing, which covered another 225 sites, 42.7%. Both figures come from Citevio's dental practice web anatomy study, 6-26 June 2026.
How did the study get from 611 raw records down to the published 527?
Citevio started with 611 raw records from the same crawl and dropped 84 sites that showed no dental signal at all, leaving the published denominator of 527. That trimming happened before any schema counting, so the 53.9%, 42.7% and 3.4% figures are all measured against the same 527, not against the original 611.
Does a schema validator tool prove my markup is being read correctly?
Not by itself. This page's own audit treats reading the live output or running a validator as only the first check, confirming what code is present, while a second, separate check asks whether the fields match the visible page. The transcript notes that a validator may approve the syntax while the underlying statement is still wrong or hard to interpret, so passing validation is not the same as being accurate.
Is a website with no schema markup at all the same problem as one with invalid schema?
No, they're two different problems. Having no LocalBusiness or Dentist schema at all was the most common state in Citevio's study, 284 of 527 sites, 53.9%. Invalid or wrong schema is a separate issue: in an outside test, researcher Mark Williams-Cook showed ChatGPT and Perplexity repeating a false address planted in deliberately invalid structured data. Missing schema is an absence; invalid schema is a fact that can travel into an AI answer.
What's the difference between "implementation" and "outcome" when it comes to schema markup?
Implementation is binary and checkable on the page itself: the markup exists, validates, and matches the visible content, or it does not. Outcome is different: whether the practice is actually named, cited, or absent on a recorded AI assistant run. This page's study only measured implementation across 527 dental practice websites; a clean implementation can still coexist with a missing name, because the two are separate questions.
Who should own keeping schema markup updated when practice details change?
Someone at the practice or its vendor needs to know which system generates the markup and what happens when the visible page changes, such as a new address, clinician or service. This page's audit treats that as ownership, the fourth check, separate from whether the markup is currently valid or complete. An unowned block ages quietly and can start disagreeing with the page it describes.
Was this a controlled test of schema's effect on AI visibility, or a snapshot of what practices currently have?
A snapshot, not a controlled test. Citevio's web anatomy study counted how common each schema implementation state was among 527 sites during 6-26 June 2026; it did not hold a question set, engines, wording and run rules fixed across a before-and-after comparison. That means nothing in this study supports a causal claim that adding or completing schema changes whether an AI assistant names a practice.
Does this study say which schema type AI assistants prefer for dental practices?
No. Citevio measured whether LocalBusiness, Dentist and FAQPage schema were present, and the completeness state that combination produced, but did not test whether assistants prefer one schema type over another for dental practices. Turning the counts into a recommended "best" type would claim more than the study measured.
Is the 53.9% / 42.7% / 3.4% breakdown a national figure for all US dental practices?
No. These percentages come from Citevio's dental practice web anatomy study of 527 US dental practice websites across seven metros plus nine smaller nearby towns, scanned 6-26 June 2026, a dated snapshot of one regional sample, not a live feed and not a national figure.
What happens to a schema block that nobody maintains over time?
It ages quietly. If no one owns updating the markup when the visible page changes, a new address, service or clinician can make the page and the structured data disagree, so a once-correct block stops matching what's actually true. This page's audit treats ownership as its own separate check for that reason.
Should I trust a vendor who promises adding schema will produce a set number of AI mentions or patients?
No. This study did not measure any causal effect of schema on named mentions, rankings, patients or timelines, so a vendor promising a specific number or deadline from adding one block of code is speaking beyond what this study measured. The honest position is to make the markup accurate and then measure assistant outcomes separately, under a fixed protocol.
Does the study explain why so many dental websites are missing schema markup?
No, it doesn't. Citevio's web anatomy study counted how common each schema implementation state was, 53.9% missing, 42.7% incomplete, 3.4% complete, but it did not investigate why practices ended up in one state or another. We haven't measured that.
Where can I see the full schema methodology and per-metro breakdown behind these numbers?
At citevio.com/data, where Citevio publishes the full methodology, per-metro breakdown and raw CSV for the dental practice web anatomy study, 6-26 June 2026. No practice name or domain is published in that aggregate package, and no client data is included. See the open data page directly.
Full transcript
Schema markup can make your practice easier to describe. It cannot buy a citation, force a recommendation, or guarantee that an assistant says your name. That is the short answer to “does schema markup help AI find my dental practice?” The honest role is narrower and more useful.
Start with the field data. In Citevio’s dental practice web anatomy study of 527 US dental practice websites across seven metros plus nine smaller nearby towns from 6–26 June 2026, 284 of 527 sites, or 53.9%, had no LocalBusiness or Dentist schema.
The complete group was small. In the same Citevio study from 6–26 June 2026, 18 of 527 US dental practice websites, or 3.4%, had the complete measured configuration. From 6–26 June 2026, another 225 of 527 sites in that audit, or 42.7%, were incomplete under the measured definition.
What does that tell us? It tells us how common each implementation state was in this sample. It does not tell us that missing markup caused a practice to be absent, or that complete markup caused a practice to be named by an AI assistant.
Schema is a structured label attached to a page. It can state the business type, name, address, and relationships in explicit fields. That lets a machine receive a clean declaration instead of rebuilding the identity from whatever it can infer from marketing copy.
That becomes useful when a homepage contains several services, locations, and brand phrases but never gives one plain statement of identity. Structured data can connect those facts. It can also help another system distinguish which fact belongs to which entity. Think of the markup as a compact claim sheet whose entries must be supported by the page a visitor can read. It can reduce ambiguity inside the document. It cannot establish that an outside system will trust, retrieve, or cite the claim.
But a label is not a lever. Valid markup does not make a claim true, and a complete block does not compel an engine to cite the page. AI placement cannot be bought. There is no paid slot in an AI recommendation.
An external experiment helps define the limit. Researcher Mark Williams-Cook placed a false address inside deliberately invalid structured data on a test page whose visible text carried no address at all. ChatGPT and Perplexity repeated the false address anyway. His reading: the assistants lifted the text straight out of the HTML without parsing it as schema.
That experiment does not show schema has no value. It shows that “schema makes AI cite you” is too strong. A single test page is also different from Citevio’s field audit. The two pieces of evidence should be kept separate. One marks a limit; it is not a reason to abandon the infrastructure.
The first audit question is simple: what structured data is present in the live page source? Ask for the rendered output or use a validation tool on the public URL. Do not accept a plugin toggle as proof. The thing an external machine receives is the live output, not the plugin settings screen.
The second question is whether the fields match the visible page. A clean schema block with an old address, wrong business type, or invented service is not an improvement. Machine-readable does not mean safe from scrutiny. It can simply become a clean version of the wrong fact.
The third question is coverage. A site can carry a generic organization block and still omit the practice identity a local retrieval system needs. Check whether the type, name, address, telephone, and page relationships are explicit and consistent. We have not measured which fields create visibility, so frame this as a useful machine-readable identity for the practice. Coverage also means choosing the right subject. A clinic, a clinician, and a service are related, but they are not interchangeable. If the block blurs them together, a validator may still approve the syntax while the underlying statement remains hard to interpret.
The fourth question is ownership. Someone must know which system generates the markup and what happens when the visible page changes. An unowned block ages quietly. A new address or service can make the page and the structured data disagree, preventing a once-correct measured configuration from staying accurate. Put the check into the same workflow that changes hours, clinicians, services, and locations. That is maintenance, not optimization theater. The value comes from keeping the statement accurate enough to audit, not from adding more fields simply because a plugin offers them.
Now separate implementation from outcome. The implementation check is binary and inspectable: the markup exists, validates, and matches the page, or it does not. The outcome check is different: the practice is named, cited, or absent on a recorded assistant run. A clean implementation can coexist with a missing name. That coexistence is not a contradiction. The two checks answer different questions. One asks whether the site states facts coherently. The other asks what a changing external system returned under recorded conditions. Never use success in the first check as a substitute for observing the second.
A clean before-and-after test would keep the question set, engines, wording, and run rules fixed. It would record errors separately and compare enough scheduled runs to avoid treating one answer as a trend. Citevio’s web anatomy audit did not run that intervention, so the conclusion should stay descriptive.
We also did not measure which schema type, if any, assistants prefer for dental practices. We measured LocalBusiness, Dentist, and the defined completeness state in a regional sample. Turning that into a universal recipe would pretend the instrument answered more.
So what should a practice do? Make the live markup accurate, explicit, and consistent with the visible page. Keep a copy of the before state. Then measure assistant outcomes separately under a fixed protocol. That turns a technical cleanup into something testable without turning it into a guarantee.
What should a practice refuse? A promise that adding one block of code will produce a named mention, ranking, patient, or deadline. We have not measured any of those causal effects. A vendor making that promise is speaking beyond what this study measured.
The measured conclusion is modest. Most sites in the sample lacked the measured machine-readable identity, and very few had the complete measured state during 6–26 June 2026. That is a real implementation gap worth inspecting. We have not measured the outcome effect.
The aggregated schema counts, definitions, and limitations are published at citevio.com/data. The files are dated snapshots, not a live feed. The market research is public; any client record is not. That boundary keeps the audit honest and useful.
Narration in this video is synthesized; the data, methodology and limitations are our own and are linked below.
Prefer to run the audit by hand? See how to test if AI can read your dental website.