Google recommends JSON-LD as the format for Organization structured data, and the property that does the heavy lifting is sameAs: a list of URLs that point to your Wikipedia page, your Wikidata entry, and your verified profiles (Google). Those links are how an engine matches your domain to a known entity in its knowledge graph instead of treating it as one more anonymous website.
Organization schema in JSON-LD, with a sameAs array pointing to Wikipedia, Wikidata, and verified social profiles, tells AI engines your brand is a specific known entity rather than an unidentified domain. sameAs is the disambiguation link: it connects your site to authoritative records engines already trust, so they consolidate the mentions of you into one thing they can cite.
What is entity anchoring?
An AI engine does not think in websites. It thinks in entities: people, organizations, places, products, each a node connected to other nodes. When your content says “Acme,” the engine has to decide which Acme, and whether this Acme is the same one it read about on three other sites. If it cannot resolve that, your mentions stay scattered and none of them accrue to a single recognized thing.
sameAs is how you tell it. The schema.org definition is precise: sameAs is a URL that unambiguously indicates the item’s identity, typically its Wikipedia page or Wikidata entry (schema.org). Point your Organization markup at your Wikidata record and your verified profiles, and you hand the engine the resolution it was going to guess at. Wikipedia and Wikidata belong in sameAs specifically because engines use them as the reference layer for the knowledge graph. You are not describing yourself. You are citing the authorities that already describe you.
Why does this matter more for AI than for classic search?
Because the model has no memory of you. This is the part founders underestimate. In our own testing, we asked Gemini 2.5 Flash, with live search switched off, about a site that had been online for four months. It answered “UNKNOWN.” Not a hallucination, not a wrong guess: a flat admission that the brand was not in the model’s trained knowledge at all.
That is the default state of every young company. The model was trained before you existed, so its memory of you is empty, and its only way to say anything true is to retrieve it in the moment. Retrieval is the entire surface. And retrieval works far better when the engine can resolve your pages to a known entity than when it has to reason about a domain it has never encountered. Entity markup is how a four-month-old site borrows the credibility of records the engine already trusts, instead of waiting a year to maybe get absorbed into the next training run. We go deeper on that trained-versus-retrieved split in why the model’s memory of your brand refreshes on a yearly clock.
What does the markup actually look like?
A minimal, correct Organization block, rendered into the initial HTML:
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "Acme Analytics",
"url": "https://acme.example",
"logo": "https://acme.example/logo.png",
"sameAs": [
"https://www.wikidata.org/wiki/Q000000",
"https://www.linkedin.com/company/acme-analytics",
"https://github.com/acme-analytics"
]
}
Two rules keep it honest. First, the sameAs URLs have to actually be you: the LinkedIn company page you control, the Wikidata entry that describes this organization, the GitHub org you run. Pointing sameAs at pages you do not own is not a growth hack, it is a correctness bug the engine can catch. Second, the block has to land in the HTML the server sends, not get injected by JavaScript after the page loads. AI crawlers read raw HTML and do not run scripts, so schema added client-side is schema that, to them, does not exist. If you take one thing from this, make it server-rendered.
The Wikidata entry is worth the effort even though it is unglamorous. A Wikidata record is a public, structured, machine-readable statement that your organization exists, with a stable identifier engines already reference. Creating one is free and mostly a matter of filling in fields. It is the closest thing to registering as a citizen of the knowledge graph.
The practical next step: view-source on your homepage, confirm an Organization block is present in the raw HTML with a populated sameAs, and if you have no Wikidata entry, create one this week. To see whether AI engines currently recognize your brand as an entity or a stranger, run a free scan.