BlogFindings
Organization schema type: 277 of 448 identity nodes declared no category, and 55 of 63 subtypes matched the industry
Lantad requested the home page of each of the 1,419 hostnames in this repository's two committed corpus seed files on 17 October 2026, as LantadBot with no JavaScript executed, and ran the shipped entity assessor over the 1,084 that delivered HTML. 448 declared an identity node. 171 stated what kind of organisation they are and 277 did not. Of the 63 industry-corpus sites that stated it through the type itself, 55 chose a subtype belonging to their own sector.
On 17 October 2026 this scanner requested the home page of all 1,419 hostnames in the two committed corpus seed files in this repository, as LantadBot/1.0 with redirects followed, a twenty five second timeout and no JavaScript executed, then ran the shipped extractor and the shipped entity assessor over every page that answered with HTML. 1,084 did. What follows is how many declared an identity at all, which organization schema type each one chose, how many never chose one, and whether the choices track the sector a site is actually in. The last question had an answer in this repository already, written from 24 pages, and this run tests it on 448.
In short
- The organization schema type is where most sites stop short: of the 448 home pages that declared an identity node when Lantad scanned 1,419 hostnames on 17 October 2026, 337 typed it exactly Organization, and 258 of those 337 stated no category by any route the scanner reads.
- Google's Organization structured data documentation, last updated 2026-09-08, states that it recommends using the most specific schema.org subtype of Organization that matches your organization. 89 of the 448 identity nodes did so.
- Where a site did pick a subtype, it picked the right one: 55 of the 63 subtype choices across the eight industry strata named a type belonging to that site's own sector, including 19 of 19 in news and 11 of 11 in government.
- SaaS is the exception and the vocabulary is why. 85 of 118 readable SaaS home pages declared an identity node, the highest rate of any stratum, and only 3 used a subtype, none of them a software type. 22 of them stated their category through a sibling SoftwareApplication or Product node instead.
- schema.org release 30.1, dated 2026-09-16, lists 20 direct subtypes of Organization. The IDENTITY_TYPES list in core/src/entity.ts was built against release 30.0 and carries 18 of them, omitting Cooperative and PoliticalParty. No page in this corpus declared either, so the gap cost nothing here.
What we requested, and what came back
The frame is the same one every corpus measurement on this blog uses, and it is committed rather than assembled per run: 392 hostnames across ten platform strata in worker/seeds/corpus-seeds-platform.json, generated on 30 July 2026, and 1,027 hostnames across eight industry strata in worker/seeds/corpus-seeds-industry.json, generated on 3 August 2026. One home page was requested per hostname, at the apex, with redirects followed. Nothing was rendered, which is the point rather than a shortcut: an AI crawler that does not run JavaScript sees the bytes this run saw, and the whole premise of the entity diagnostic is the crawler view.
1,084 hostnames delivered an HTML document on a 2xx status, 1,074 of them on HTTP 200 and 10 on 202. The rest did not answer with a page. 232 returned HTTP 403, 52 returned 503, 7 returned 429, 5 returned 404, 2 each returned 401 and 406, one each returned 498 and 500, and 33 requests failed inside the client, 22 at the connection and 10 on the twenty five second timeout with one connection terminated mid body. The 403s are not spread evenly, and the pattern repeats what this blog has measured every time it has counted them: 62 of the 128 news hostnames refused LantadBot outright, along with 50 of 130 in ecommerce and 33 of 130 in travel, against 9 of 130 in SaaS. A scanner is not a citation crawler, so that is a statement about how those sectors treat unfamiliar user agents, which is the subject of the crawlability study rather than of this post. What it costs here is denominator: the news and ecommerce figures below rest on 61 and 63 readable sites.
Of the 1,084 pages, 459 carried no application/ld+json block at all and 625 carried at least one, for 1,167 blocks between them. 1,156 of those blocks parsed as JSON and 11 did not, spread across 11 separate pages, which is the ordinary rate of broken markup this corpus shows and is itself a schema defect worth its own count. Flattening the parsed blocks with the shipped traversal gave 9,238 typed nodes on 620 pages, using 141 distinct type strings. 21 of the 1,084 pages ran past the two megabyte capture limit and were read only to that point, so for those 21 an absent signal means absent from the part we read. That caveat is applied by the assessor itself rather than added here.
| Outcome | Hostnames | What it means for this study |
|---|---|---|
| HTML on a 2xx | 1,084 | The denominator for everything below. 1,074 on 200 and 10 on 202 |
| HTTP 403 | 232 | Refused the scanner. Concentrated in news, ecommerce and travel |
| HTTP 503 | 52 | Unavailable. Not requested again |
| HTTP 429 | 7 | Rate limited on a single request |
| HTTP 404, 401, 406, 498, 500 | 11 | No page at the apex, or an error answer |
| Failed in the client | 33 | 22 at the connection, 10 on timeout, 1 terminated mid body |
Which organization schema type did each site choose?
448 of the 1,084 pages declared what the scanner counts as an identity: a node carrying an Organization-family type, a name, and an absolute url, preferring a node whose own url sits on the page's host so that a retailer is not scored as the brand it stocks. 337 of those 448 nodes reported their type as exactly Organization. A further 22 reported Corporation or Brand, which the scanner reads as generic too, on the grounds that every identity node already implies a company exists. That leaves 89 nodes carrying a type narrower than the family name, and the distribution inside those 89 is narrow in a second sense: 15 distinct types carried all of them. Eight of the 15 are among the 20 direct subtypes schema.org defines for Organization in release 30.1 dated 16 September 2026, and the other seven are deeper descendants rather than direct subtypes.
The four most used are the four you would guess, and the guess is the finding. NewsMediaOrganization appeared on 22 nodes, LocalBusiness on 18, GovernmentOrganization on 12 and MedicalOrganization on 10. Those four account for 62 of the 89. Then the tail: OnlineStore on 4, and ProfessionalService, EducationalOrganization, FinancialService, ResearchOrganization and InsuranceAgency on 3 each, Dentist, NGO and Airline on 2 each, and GovernmentOffice and Store on one apiece. Named cases read exactly as the counts suggest: cnn.com, aljazeera.com, cbsnews.com and cnbc.com all type their identity NewsMediaOrganization, ca.gov and rijksoverheid.nl and nsw.gov.au use GovernmentOrganization, boston.gov uses GovernmentOffice, and clevelandclinic.org, mdanderson.org and nationwidechildrens.org use MedicalOrganization.
Google's position on which to pick is not ambiguous. Its Organization structured data documentation, last updated 2026-09-08, states that it recommends using the most specific schema.org subtype of Organization that matches your organization, and gives an ecommerce site choosing OnlineStore over OnlineBusiness as its worked example. The page also states that there are no required properties, which is the part that explains the gap: nothing fails, nothing warns, and a bare Organization node validates perfectly. 89 of 448 followed the recommendation through the type. That is the whole of the first finding, and it is a fact about defaults rather than about effort, because the sites getting it right are mostly not the ones trying hardest. 12 of the 54 readable Wix and Squarespace sites stated a category and all 12 did it through a subtype, 11 as LocalBusiness and one as MedicalOrganization, because that is what the template emits. blacksheepbikes.com and scottiespizzaparlor.com did not make a schema decision. Their builder made it, correctly, and it is the same decision LocalBusiness markup gets right and then leaves half finished.
| Type declared | Nodes | Example from this corpus |
|---|---|---|
| NewsMediaOrganization | 22 | cnn.com, aljazeera.com, cbsnews.com |
| LocalBusiness | 18 | blacksheepbikes.com, 400rabbits.co.uk |
| GovernmentOrganization | 12 | ca.gov, rijksoverheid.nl, nsw.gov.au |
| MedicalOrganization | 10 | clevelandclinic.org, mdanderson.org |
| OnlineStore | 4 | shein.com, mercadolivre.com.br, westwing.de |
| InsuranceAgency | 3 | statefarm.com, newyorklife.com, lemonade.com |
| FinancialService | 3 | fidelity.com, emiratesnbd.com |
| Nine further subtypes | 17 | ResearchOrganization, EducationalOrganization, ProfessionalService, Dentist, NGO, Airline, GovernmentOffice, Store and one more |
The 277 that named the company and stopped
277 of the 448 identity nodes stated no category by any of the six routes the assessor reads. By reported type they are 258 typed exactly Organization, 16 typed Corporation and 3 typed Brand. Those are three of the four types the scanner treats as generic because every identity node already implies as much, the fourth being OnlineBusiness, which no identity node in this corpus used. Corporation says a company is incorporated. Brand says a brand exists. Neither narrows what the entity does, and the scanner is explicit in the code about not awarding the point for them.
The names in that group are the ones worth sitting with, because they are not small sites with no budget. cloudflare.com, asana.com, docker.com, box.com, circleci.com, browserstack.com and canonical.com all ship a clean Organization node and never say what they sell. forbes.com, irishtimes.com, handelsblatt.com and hollywoodreporter.com publish journalism and do not type themselves as news organisations, while 19 of their readable peers do. texas.gov, japan.go.jp, metoffice.gov.uk and worldbank.org are governments and institutions whose identity nodes say Organization. Not one of those pages is broken. Each one answers the first question and declines the second, and because Google states there are no required properties, nothing anywhere tells them.
It is worth being precise about what this costs, because the honest answer is that nobody outside the engines can measure it. This blog has published that structured data came third in a 1,100 URL citation audit rather than first, and nothing in this run observed a citation, a ranking or an answer. What can be said is narrower and still useful: the category is a claim a machine can read without inference, and the alternative is inference from prose. The scanner's own entity confidence diagnostic prices it accordingly, at 10 points of 100, the smallest of its five signals, behind identity at 35, sameAs at 25, and @id and logo at 15 each. Those are weights somebody chose, recorded in core/src/config.ts, and not a measurement of what any engine does with the property. The same file is candid about why it is smallest: it was the rarest signal in the sample the weights came from. Treating it as a 10 point improvement to a diagnostic score is accurate. Treating it as 10 percent more visibility would be an invention, which is the distinction the methodology page exists to keep.
- Identity declared 35 pts 448 of 448, by definition: a typed Organization-family node with a name and an absolute url. The other 636 readable pages carried no such node in the HTML a crawler receives.
- Logo declared 15 pts 371 of 448. The most common of the four dependent signals. Google lists logo among its recommended Organization properties.
- External identifiers (sameAs) 25 pts 325 of 448 carried at least one absolute sameAs URL on the identity node itself, rather than on some other node that mentions them.
- Stable identifier (@id) 15 pts 230 of 448. An absolute @id lets other nodes and other documents point at one identifier instead of repeating the details.
- Category declared 10 pts 171 of 448, the rarest of the five on this corpus, and the smallest weight for the same reason in the sample the weights came from.
The six routes to a stated category, and which ones sites actually use
The subtype is the first route and not the only one. The assessor tries six rules in a fixed order, from the identity node saying what it is to some node in the document saying what the site offers, and the first match wins. Written out, the ladder is: a specific subtype on the identity node, then an additionalType URL on it, then a standard industry code in naics or isicV4, then knowsAbout naming subject areas as values, then any node in the document carrying one of the eight types that say what is on offer, and last an applicationCategory value on the identity node. Prose is deliberately not a rule. A description field saying "we are a dental practice" earns nothing, and the comment in the code records that it used to, which made a property found on five of 24 pages look common until the escape hatch was removed.
Across the 171 pages that stated a category, 89 did it through the subtype, 59 through a sibling node, 17 through knowsAbout, 3 through an industry code and 3 through additionalType. The sixth rule decided nothing: no identity node in this corpus reached applicationCategory without an earlier rule having already matched. The two small groups are the interesting ones because they are deliberate in a way a template never is. capitalone.com declares naics 522291, icicibank.com declares isicV4 6419, and bindwell.ai declares naics as well. unacademy.com points additionalType at a Wikidata entity, which is the only form in the whole corpus that resolves to a third party record rather than restating something the page already said, and is the same move as naming a reference entry in sameAs. edreams.com also uses additionalType, and points it at its own home page, which satisfies the rule and conveys nothing.
The sibling node route carries 59 pages and 43 of those are a SoftwareApplication node, with Product on 7, Service on 6, FinancialProduct on 2 and WebApplication on one. This is how software companies answer the question, and it is a real answer rather than a loophole: a document that declares an Organization and a SoftwareApplication in the same graph has stated what kind of thing it is about. datadoghq.com, dynatrace.com, auth0.com, fly.io, dropbox.com and monday.com all do exactly that, with a bare Organization for the company and a typed application node beside it. The limit is that the node has to carry real properties to be worth anything, and this blog has already measured that 4 of 57 SoftwareApplication declarations carried every required property. A category stated by a node that is itself a stub is a weaker claim than it looks, which is the same failure as a schema value that is not on the page.
Flow: Identity node found (first) to Specific subtype: 89; Specific subtype: 89 (else) to additionalType URL: 3; additionalType URL: 3 (else) to naics or isicV4: 3; naics or isicV4: 3 (else) to knowsAbout values: 17; knowsAbout values: 17 (else) to Sibling offer node: 59; Sibling offer node: 59 (else) to applicationCategory: 0; applicationCategory: 0 (else) to No category: 277.
Does the subtype match the industry the site is in?
This is the question the run was built to settle, and the reason it needed settling is recorded in this repository. The comment above ENTITY_SIGNAL_WEIGHTS in core/src/config.ts states that a specific subtype appeared on only 5 of the 24 pages in the sample the weights were derived from, and that it clusters entirely by vertical, listing three news publishers using NewsMediaOrganization, both commerce entities using commerce subtypes, and every SaaS company using a bare Organization. 24 pages is small enough that the note says so itself. The industry corpus exists to test claims of that shape, because it is stratified by sector rather than by platform.
It holds. 63 of the subtype choices fall in the eight industry strata, and 55 of those 63 named a type belonging to that site's own sector. All 19 news sites that chose a subtype chose NewsMediaOrganization. All 11 in government chose a government type, 10 GovernmentOrganization and one GovernmentOffice. All 4 in ecommerce chose a commerce type, 3 OnlineStore and one Store. All 5 in finance chose FinancialService or InsuranceAgency. 9 of 12 in healthcare chose MedicalOrganization, and the three that did not are defensible rather than wrong: cancer.org and ama.com.au type themselves NGO, which they are, and moh.gov.sg is a health ministry that types itself GovernmentOrganization, which it also is. Four of the eight apparent mismatches are sites that genuinely sit in two sectors at once: the three in healthcare, and visitbritain.org in travel, a public body running a tourism site that types itself GovernmentOrganization. One is a real oddity, frontiersin.org, an academic publisher in the education stratum typed NewsMediaOrganization. The other three are all in SaaS, and they have a different cause.
The exception is SaaS, exactly as the 24 page note said, and the cause is the vocabulary rather than the sites. 85 of 118 readable SaaS home pages declared an identity node, the highest rate of any stratum by a wide margin and well above the 448 of 1,084 overall. Only 3 used a subtype, and none of the three used a software type, because schema.org does not define one as a subtype of Organization: misa.vn types itself ProfessionalService, outsystems.com and webnames.ca type themselves LocalBusiness. The route those companies actually use is the sibling node, 22 of them, with knowsAbout taking 8 more. So the clustering claim survives contact with 20 times the sample, and the exception survives with it, now with a mechanism attached rather than an observation. A SaaS company cannot pick a fitting Organization subtype because there is not one to pick, which is worth knowing before you go looking for the missing word on your own site. The same split shows up in how much prose each sector sends a crawler at all, which the supplier unit measurement counted on this corpus, and in the per stratum AI visibility grades on the score breakdown.
| Stratum | Seeded | Readable | Identity | Category | Subtype | Own sector |
|---|---|---|---|---|---|---|
| news | 128 | 61 | 44 | 19 | 19 | 19 |
| government | 129 | 85 | 17 | 11 | 11 | 11 |
| healthcare | 123 | 94 | 30 | 13 | 12 | 9 |
| education | 130 | 106 | 19 | 7 | 6 | 5 |
| finance | 127 | 93 | 40 | 10 | 5 | 5 |
| ecommerce | 130 | 63 | 27 | 6 | 4 | 4 |
| travel | 130 | 77 | 21 | 4 | 3 | 2 |
| saas | 130 | 118 | 85 | 33 | 3 | 0 |
What this run does not show, and one thing it found in our own code
Six limits, stated plainly, because the figures above are only worth what their bounds allow. Home pages only were read, so a site declaring a richer organization schema type on its about or contact page is counted by what its home page says. One URL per hostname was requested at the apex, so subdomains are outside every count. No JavaScript was executed and no rendered view was supplied to the assessor, so every figure describes the bytes a crawler receives and nothing here distinguishes a site with no markup from a site whose markup arrives only after a browser runs, a distinction this blog has measured separately and found 27 of 404 pages gained schema when a browser ran. Presence was checked and never truth, so nothing above distinguishes a correct category from a confident wrong one. No engine citation of any of these pages was observed, so no sentence here connects a declared category to being named in an answer, which is the line the how to get cited guidance keeps on the right side of. And the 1,419 hostnames are two committed sampling frames rather than a random sample of the web, so no proportion generalises beyond them.
One more, specific to the day: 232 hostnames refused LantadBot with HTTP 403 and 52 answered 503, so 1,084 of 1,419 is the real denominator and the sector rows vary in strength because of it. Our own user agent and what it asks for is published on the bot page, and a reader who wants to see the crawler view of their own page rather than trust ours can run what GPTBot sees.
The run also found something in our own code, which is the part worth acting on. schema.org release 30.1, dated 16 September 2026, lists 20 direct subtypes of Organization. The IDENTITY_TYPES set in core/src/entity.ts carries a comment saying it was built from schema.org v30.0, dated 19 March 2026, and it holds 18 of those 20. Cooperative and PoliticalParty are absent, so a site typing its identity node as either would be told by this scanner that it declared no identity at all, not merely that it declared no category. The honest size of that defect in this corpus is zero: neither string appeared among the 141 distinct types on any of the 1,084 pages, so nothing in the figures above is affected. It is still a list that lags the vocabulary it is built from by one release, found by running the measurement rather than by reading the file, and the fix is smaller than this paragraph. That is the kind of thing a schema markup type census turns up and a code review does not, and it is why the counts in these posts are computed with the shipped functions instead of a script written for the occasion.
-
Category declared171 of 448 277 identity nodes named the company and stated no kind of organisation by any of the six routes the shipped assessor reads. -
Subtype fits sector55 of 63 Where an industry-corpus site chose a subtype it chose one belonging to its own sector, including 19 of 19 in news and 11 of 11 in government. -
Our own type list18 of 20 IDENTITY_TYPES in core/src/entity.ts omits Cooperative and PoliticalParty. Zero pages in this corpus declared either, so no figure here moves. -
Effect on citationsNot measured No answer engine was queried and no citation was observed. Nothing here links a declared category to appearing in an answer. -
Rendered markupNot measured No JavaScript was executed and no rendered blocks were passed to the assessor, so every count is the crawler view of delivered bytes.
Lantad
Published .
A machine reading your home page has to answer two separate questions, and the second one is the one sites forget. The first is who is behind this page, which an Organization node with a name and a URL answers. The second is what kind of organisation that is, which the bare word Organization does not answer at all, because every company on the web is one. That second answer is a structured data decision that takes one word, and it is the difference between a machine knowing a hospital is a hospital and a machine knowing only that a hospital exists.
Common questions
What is the organization schema type, and which one should I use?
It is the @type value on the schema.org Organization node that identifies the site owner. Google's Organization structured data documentation, last updated 2026-09-08, states that it recommends using the most specific schema.org subtype of Organization that matches your organization, and gives an ecommerce site using OnlineStore rather than OnlineBusiness as its example. Release 30.1 of schema.org, dated 16 September 2026, defines 20 direct subtypes, among them NewsMediaOrganization, GovernmentOrganization, MedicalOrganization, EducationalOrganization and LocalBusiness with its own large subtree.
Is a bare Organization type wrong?
No. It is valid, it passes every validator, and Google's own page states that there are no required properties, which is why so little pressure exists to improve it. When Lantad scanned 1,419 hostnames on 17 October 2026, 337 of the 448 identity nodes found were typed exactly Organization and 258 of those stated no category by any route. The cost is that the markup answers who you are without answering what you are, leaving the second question to be inferred from prose.
What if no Organization subtype fits my company?
Then say it another way, which is what the software companies in this corpus do. schema.org defines no SaaS subtype of Organization, so 22 of the 33 SaaS sites that stated a category did it by declaring a node beside the Organization that says what they offer, 16 of those a SoftwareApplication and 4 a Product, and 8 more used knowsAbout. Two other routes exist and were used by 6 sites between them: an additionalType URL pointing at a resolvable record such as a Wikidata entity, and a standard industry code in naics or isicV4.
Does declaring a subtype improve AI visibility?
This run cannot say, and no figure in it should be read that way. It measured what 1,084 home pages declared on 17 October 2026 and queried no answer engine, so it observed no citations, rankings or answers. Lantad's Entity Confidence diagnostic weights a declared category at 10 points of 100, the smallest of its five signals, and that weight is a design decision recorded in core/src/config.ts rather than a measurement of any engine's behaviour.
See what AI can read on your site
Run a free scan and get a graded report of exactly what AI crawlers can and cannot read, with ranked fixes.
Structured data for AI (schema markup)
Structured data is schema.org markup, usually written as JSON-LD, that labels a page's facts in a machine-readable format so AI systems can extract and attribute them without having to interpret the surrounding prose.
Entity confidence
Entity confidence is a free diagnostic on every graded Lantad report that scores, out of 100, how completely the JSON-LD in your crawler-visible HTML declares one identity and how far that identity is anchored by identifiers which resolve outside the page.
AI visibility
AI visibility is whether AI systems can access, read, and faithfully represent your web pages, the prerequisite for being cited in an AI answer.
Schema markup types: 88 of the 939 schema.org defines reached 626 home pages, and the top three name the page
Lantad asked all 1,419 hostnames in this repository's two committed corpus seed files for robots.txt on 16 October 2026, then read every permitted home page as LantadBot with no JavaScript executed. 1,085 answered a 2xx with HTML, 631 carried at least one JSON-LD block and 626 held one that parsed. Across the 2,181 typed nodes those pages declared, 88 distinct type strings appeared, against the 939 classes schema.org release 30.1 defines. WebSite, Organization and WebPage accounted for 1,029 of the 2,227 type declarations, and five of the 88 strings are not types schema.org defines at all.
sameAs schema: 85 of 317 organizations named a reference entry, and 310 named a social profile
Lantad read the home page of 1,419 hostnames on 25 September 2026 with no JavaScript executed. 1,068 answered HTTP 200 with HTML, 433 carried an identity node the scanner recognises, and 317 of those declared sameAs on it. Those 317 pointed at 185 distinct hosts, of which the six largest are social networks: 310 named a social profile and 85 named a Wikipedia or Wikidata entry.
Entity SEO: 107 of 163 schema topic statements named the site itself
Lantad asked all 1,419 hostnames in this repository's committed corpus for robots.txt on 8 October 2026 and read each home page it was allowed to read, with no JavaScript executed. 1,074 answered HTTP 200 with HTML and 612 of those carried parseable JSON-LD. 129 of the 612 used about, mainEntity or knowsAbout to name a subject, and of the 163 about and mainEntity values across 104 sites, 107 pointed at the site's own organisation, service or founder rather than at a topic.
