How Much of the Web Describes Itself in Machine-Readable Form

Structured data does not have to be estimated. It is extracted and counted. The Web Data Commons project at the Data and Web Science Group, University of Mannheim, runs a parser over the Common Crawl corpus and publishes the totals. Its most recent release, published in January 2025 and covering the October 2024 crawl, records that 16,525,070 of 37,447,141 pay-level domains carry structured data, or 44.12%. Under half the measured web describes itself in a form a machine can read without inference, and the markup that exists is thinner than the totals imply.

The size of the count

The release is large enough to treat as a census rather than a sample. Web Data Commons reports 2,391,039,772 parsed HTML URLs and 73,993,669,093 triples for the October 2024 crawl, containing 15,647,463,083 typed entities and occupying 1.4 terabytes compressed. At page level the coverage is higher than at domain level: the project's January 2025 announcement on the W3C semantic-web list states that markup was found on 1.3 billion of the 2.4 billion pages in the crawl, or 51.25%.

The two rates differ because markup is not distributed evenly within a domain. Templated sites emit it on every page; the rest emit it on a home page and nowhere else. Domain coverage is the stricter number and the more useful one, because it counts organisations rather than URLs.

Growth, and a decline that is not one

Adoption is rising. Web Data Commons recorded 14,646,081 domains with structured data in the October 2023 crawl and 16,525,070 in October 2024, a rise of 12.8%. Over the same interval the triple count fell, from 97,689,391,384 to 73,993,669,093. That fall is an artefact of corpus size and the publisher says so directly: the October 2024 crawl is the smaller of the two. The domain count is the growth metric; the triple count is not comparable between releases.

One caution attaches to the whole series. There is no release later than the October 2024 crawl. Any figure presented as a 2025 or 2026 Web Data Commons measurement does not exist.

Which formats carry it

Format share is concentrated. Web Data Commons counts embedded JSON-LD on 11,562,359 domains against 7,599,792 for Microdata, 3,522,517 for Microformats hCard and 474,635 for RDFa, and its announcement puts JSON-LD in use by 70% of websites that annotate structured data with Microdata at 46%, hCard at 23% and RDFa at 3%. The shares exceed 100% because sites commonly emit more than one format at once. Volume follows the same ordering: JSON-LD accounts for 47.98 billion of the triples extracted, Microdata for 21.83 billion.

An independent page-level measurement agrees on direction. The HTTP Archive's 2024 Web Almanac found JSON-LD on 41% of pages, up from 34% in 2022, with Microdata at 26% and Open Graph at 64%. The same chapter reports RDFa at 66%, which is a syntax-level detection that sweeps in Open Graph og: properties rather than a count of semantic RDFa, and should not be read as adoption of the format.

What the markup says

The class distribution is the finding. Counting pay-level domains that emit at least one entity of each schema.org class in JSON-LD, Web Data Commons records WebSite on 8,461,613 domains, ImageObject on 6,873,617, Organization on 6,574,284 and SearchAction on 6,573,017. The commonest assertion on the web is that a site exists. The second declares an image. Markup naming the organisation comes third, on almost exactly the same number of domains as a declaration that the site has a search box. Three of the four most widely emitted classes describe the website rather than the body that runs it.

Page-level types show the same shape. The Web Almanac's mobile figures put WebSite at 12.73% of pages and Organization at 7.16%, followed by BreadcrumbList at 5.66%, LocalBusiness at 3.97%, ItemList at 2.44%, BlogPosting at 1.40% and Product at 0.77%. Commercial description is the thinnest layer in the stack. Fewer than one page in a hundred carries product markup.

Two totals that do not agree

Schema.org states on its own home page that over 45 million web domains mark up their pages with over 450 billion Schema.org objects as of 2024. It publishes no method for that figure. Web Data Commons, which publishes its method, measured 16.5 million domains over the same period. The two numbers differ by almost a factor of three and cannot both describe the same population. The measured figure is the one that can be checked; the discrepancy should be named wherever both appear.

Implication

Coverage is a differentiator because the base rate is low. On the domain measurement, a majority of the web emits nothing. Specificity is a second differentiator because, among sites that do emit something, the markup is mostly navigational furniture. An organisation whose pages assert the narrower classes — what it sells, where it operates, what an item costs — has placed itself in a small subset, and that position is observable from outside without the organisation's cooperation.

What the assertion buys is documented, not measured. Google Search Central states that structured data is used to understand the content on a page and show it in a richer appearance, and names twenty-five eligible features, among them Article, Breadcrumb, Event, Job posting, Local business, Organization, Product, Review snippet and Video. That is a platform's description of its own behaviour and should be read as such.

No published figure states what share of an organisation's machine-readable description is supplied by the organisation rather than derived by the platform, and none of the sources above measures it. The measurable part is coverage and class specificity, across a named peer set, at a stated date. Where a competitor emits Organization and Product and a firm emits only WebSite, the difference is a fact about the two firms, countable in an afternoon.