Google and Schema.org have published a monthly dataset showing which Schema.org types and properties are used across millions of domains. For people responsible for technical GEO work, that changes the markup conversation: there is now an observable adoption signal to put beside feature documentation, implementation cost and content quality instead of treating every valid Schema.org term as equally strategic. ppc.land
The dataset does not establish that any term earns a ranking benefit, a rich result, or inclusion in an AI-generated answer. Google’s own documentation is explicit that its Search documentation, not Schema.org documentation, is definitive for Google Search behaviour. That distinction is the useful corrective to speculative AEO claims: vocabulary validity, web adoption and Google Search support are three different questions. Google Search Central
What Google and Schema.org actually released
The first release was announced on June 4, 2026. It is a joint effort drawing on Google’s web crawling infrastructure, with CSV, JSON and JSON summary files published through Schema.org’s GitHub repository and expected to be updated monthly. Each record identifies whether a term is a Type or Property, gives its URI, and assigns it to a domain-count range. ppc.land
The ranges are deliberately broad: fewer than 1,000 domains; 1,000 to 10,000; 10,000 to 100,000; 100,000 to 1 million; 1 million to 10 million; and more than 10 million. Schema.org does not publish precise counts because crawl observations fluctuate and because exact numbers could expose information about crawling patterns. This is adoption telemetry, not a market-share table. ppc.land
Crucially, the unit is the domain. A site using Product markup on hundreds of thousands of URLs counts once for Product; a site using a term only once also counts once. That makes the data useful for judging breadth of deployment, but not for estimating page-level prevalence, markup depth, implementation quality, or the number of entities marked up. ppc.land
The adoption pattern is more useful than a list of terms
The May 2026 release contains 958 item types and 4,587 predicates. Twelve item types sit in the more-than-10-million-domain bucket: BreadcrumbList, EntryPoint, ImageObject, ListItem, Organization, Person, PropertyValueSpecification, ReadAction, SearchAction, Thing, WebPage and WebSite. That is a strong indication that the most widely deployed layer of structured data is foundational site, page, identity, image, navigation and action description—not an exotic vocabulary built for a narrow AI use case. ppc.land
In the 1 million to 10 million bucket, the dataset includes Product, Review, AggregateRating, Article, BlogPosting, FAQPage, LocalBusiness, VideoObject and Question. The equivalent property picture is similarly practical: name, description, image, url, headline, datePublished, dateModified, author, publisher, breadcrumb and logo are among the properties used by more than 10 million domains. aggregateRating, availability, articleBody, acceptedAnswer, address, brand, price and ratingValue are in the next range down. ppc.land
I would not read this as a universal implementation checklist. It is evidence that broad deployment clusters around the information a page must already express clearly: what it is, who published it, when it changed, what entity it concerns, how it relates to a site, and—where relevant—what is offered or reviewed. That is compatible with Google’s explanation that structured data gives explicit clues about a page’s meaning and helps Google understand page content and information about the web and world. Google Search Central
Why this matters for GEO prioritisation
GEO work often attracts a familiar failure mode: adding large volumes of technically valid markup because a term sounds semantically expressive or may, in theory, help a machine understand an answer. This release gives teams a way to ask a more disciplined question before they build support into a CMS, component library or publishing workflow: is this term meaningfully deployed across domains, or is it in a long tail of specialised or new vocabulary? The dataset reports 485 item types in the fewer-than-1,000-domain range. ppc.land
Low adoption is not proof that a term is worthless. Schema.org notes that low-use terms can be new or serve naturally limited communities, including medical or government publishers. A specialist publisher should not discard a term merely because general-web adoption is low. But a low bucket is valid evidence against presenting that term as established, default markup for every site. ppc.land
For GEO teams, the dataset is therefore best used as a prioritisation layer. It can help distinguish work that strengthens the common semantic baseline from work that is specialised, experimental or difficult to justify. It cannot demonstrate downstream retrieval, citation, answer selection or visibility in Google AI features. The supplied evidence says the data measures use of terms on domains; it does not connect a term to performance in AI Overviews, AI Mode or any other result type. ppc.land
Adoption is not support, and support is not eligibility
This is where implementation decisions still need Google’s documentation. Google says that most Search structured data uses Schema.org vocabulary, but there are Schema.org attributes and objects that are not required by Google Search and may instead be useful to other search engines, services, tools or platforms. A type’s presence in the new files is not confirmation that Google offers a corresponding Search feature. Google Search Central
Even documented Search markup is conditional. Google says all required properties must be included for an object to be eligible for enhanced display. It also advises that fewer complete and accurate recommended properties are more important than trying to supply every possible recommended property with incomplete, malformed or inaccurate data. That is a better operational standard than maximising the count of properties emitted by a template. Google Search Central
The same constraint applies to content parity. Structured data must be in-page markup describing the page to which it applies; Google says not to create empty pages merely to hold markup or add data about information that is not visible to users. Schema.org’s own guidance similarly says to mark up content visible to page visitors rather than hidden content. For GEO, machine-readable claims should follow the published page, not substitute for it. Google Search Central Schema.org
A practical decision model for teams
I would use the new dataset in a sequence rather than as a scorecard:
- Start with the page’s real entities and facts. Schema.org provides shared vocabularies that can describe items such as organizations, people, products, events, reviews and creative works, using Microdata, RDFa or JSON-LD. Schema.org
- Check Google’s applicable feature guidance. For Google Search behaviour, treat Search Central as the authority, including required and recommended fields. Google Search Central
- Use adoption buckets to order engineering work. Widely used foundational terms deserve scrutiny and maintenance; low-adoption terms need a specific business or content rationale. ppc.land
- Validate and monitor after release. Google recommends the Rich Results Test during development and Rich result status reports after deployment because templating or serving issues can break valid pages. Google Search Central
Format should not become a false strategic debate. Google supports JSON-LD, Microdata and RDFa, and recommends choosing the format that is easiest to implement and maintain; it says all three are equally acceptable to Google unless documentation states otherwise. The adoption dataset also does not distinguish those formats, so it cannot tell a team whether a widely used term is predominantly implemented in one syntax. Google Search Central ppc.land
The limits should shape the conclusion
This is Google-indexed-web data, not a neutral census of every site. Sites blocked by robots.txt are excluded, and crawler scope and indexing methodology affect the picture. The initial contribution comes from Google, although Schema.org has invited other crawlers and indexers to contribute in the same format. Until other contributors do, the dataset is authoritative evidence of the terms observed through this particular infrastructure—not the final account of how the whole web uses Schema.org. ppc.land
That limitation does not reduce its value. It defines it. The dataset gives structured-data teams a missing middle ground between an abstract vocabulary and anecdotal implementation advice. Use it to make markup roadmaps more evidence-led, while resisting the leap from “widely used” to “guaranteed to influence AI search.” The evidence supports the first claim. It does not yet support the second. ppc.land