Most Top Sites Never Say Who They Are in Structured Data

Why measure this at all

Entity coverage gets written about as an advanced topic. Disambiguation, knowledge panels, sameAs graphs, the shape of an organisation across the web. All of that assumes the first step is done, which is a site saying in machine readable terms who it is.

So rather than start at the interesting end, I started at the boring one. One request per homepage, parse every JSON-LD block, and record what is actually declared.

What came back

The first row is the one worth sitting with. 109 of these sites publish nothing at all in the format search engines and assistants read most easily, on the single page most likely to be crawled and cited. These are not small sites.

The half that does it does it well

The second finding softens the first. Among the 60 sites that declare an Organization, the work is not token: 57 give a name, 52 a logo, 44 at least one sameAs, and the median sameAs list is 6 entries with one site listing 32.

So this is not a story about everyone doing it badly. It is a split. A site either treats structured data as part of how it publishes, and then does it properly, or it does not do it at all. Very few sit in between with a broken half-attempt.

That matters for where to spend your effort. If you have nothing, the first hour is worth more than anything else on this page: a name, a URL, a logo, and the handful of profiles that are unambiguously you.

What to publish, in order

One, an Organization node with name, url and logo. That is the machine readable version of who owns this site, and it is the anchor everything else references.

Two, sameAs pointing at profiles you control. The point of sameAs is disambiguation, so list the accounts that are unambiguously the same entity and skip the ones that merely mention you. Six is a normal number. Thirty two is not better.

Three, a WebSite node. The site and the company are different things and conflating them is the most common modelling mistake. Add SearchAction only if your site search actually works from a query string.

Four, breadcrumbs on deep pages. Nine percent of top homepages carry a BreadcrumbList, which is low, and on a homepage it is arguably unnecessary. On a category or article page it is what replaces the raw URL in a result.

Then check the thing that decides whether any of it is read: whether a crawler can fetch the page at all. In the access study a fifth of these same sites did not serve GPTBot real content, and markup on a page nothing can reach is a note to nobody. The current crawler names to check against are in the user agent reference.

What this does not say

It does not say these sites are hard to understand. A company with a large brand footprint is resolved from many signals, and the biggest names here are understood with or without markup. Structured data is a shortcut, not the only route.

It also measures the homepage only, in one pass, from one location. A site can carry rich markup on product or article templates and nothing on the homepage, which for a retailer is a defensible choice. Read this as the state of the front door, not of the whole house.

Questions people ask about this

Method: one homepage request per host on 19 September 2026 from a single residential connection in the United States, reading up to 700,000 bytes. Every application/ld+json block was parsed and every @type recorded, including nested nodes and @graph members. 180 of the 184 hosts returned a 200. Blocks that failed to parse were counted as absent, so these figures are a floor.