Entity Authority

The definition has two halves, and both are load-bearing. Resolvable means the engine can tell what the entity is. Corroborated means the engine has independent agreement about what is true of it. An entity can fail at either half separately, and the failures look different in generated answers.

An entity is a thing, not a string

In this context an entity is a thing with identity: a person, a company, a product, a place. The string that names it is just one of its properties, and usually an ambiguous one. The same string can name a bank and a riverbank; the same company can appear under a legal name, a brand name, and an old name it stopped using. A system that matched strings would conflate all of these. A system that resolves entities instead asks: which thing is this text about?

That question is answered by context, not by the name alone. An engine resolves a mention by checking which candidate entity the surrounding facts fit: the founder named in the next sentence, the city in the byline, the product category in the heading. Every consistent co-occurring fact narrows the candidates; every inconsistent one widens them. This is why entity work is mostly consistency work rather than naming work.

How an entity becomes resolvable

Three mechanisms, in increasing order of explicitness.

Consistent naming. One canonical name, used the same way everywhere the entity controls: site, profiles, directories, bylines. Variants are not fatal, but each variant spends some of the context budget on reconciliation that could have gone toward recognition. A rebrand, a translation, or a founder who publishes under two names all create exactly this cost.

Corroborating sources. Mentions on surfaces the entity does not control, agreeing on the basic facts. These do double duty: they help resolution by repeating the identifying context, and they feed the trust half of the definition, covered below.

Structured data with a stable @id. JSON-LD lets a site state its entities outright rather than leaving them to inference: this page is about a Person with this name, this role, this employer. The @id property gives the entity a durable identifier that every page, and in principle every cooperating site, can point at. The value of the @id is in its stability: one identifier, referenced from everywhere, accumulates all the statements made about the entity in one addressable node. Minting a new identifier per page forfeits exactly that accumulation. Google's introduction to structured data describes how such markup helps its systems understand page content; the general mechanism is the same for any engine that parses it.

Why corroboration must be independent

A language model deciding whether to repeat a claim faces a verification problem: it cannot check reality, only agreement among sources. A claim that appears on one site is an assertion. The same claim appearing across several surfaces that do not copy from each other starts to look settled, because independent agreement is unlikely to arise by accident or by one party's say-so.

The word independent carries the weight. Ten pages on one domain are one voice. A press release syndicated verbatim to fifty outlets is closer to one voice than to fifty, and detectably so, since the texts are identical. What moves a claim toward settled is corroboration across surfaces with different owners, different wording, and different incentives: coverage, directories, registries, profiles, citations in others' work. This is slow to build and hard to fake at scale, which is precisely why engines can afford to lean on it.

Authority is not popularity

The two are easy to conflate because both correlate with being talked about. They diverge in what they measure. Popularity is volume of attention: mentions, followers, traffic. Authority in this glossary's sense is confidence of identification and agreement: can the engine resolve the entity, and do independent sources concur on the claims about it?

The divergence shows at both extremes. A viral brand with inconsistent naming, no structured data, and contradictory coverage can be popular and unresolvable: engines mention it hesitantly, mix it up with namesakes, or omit it from answers where it belongs. A niche technical publisher with one stable identity and a decade of consistent third-party citation can be obscure and authoritative: rarely searched for, but cited confidently whenever its topic comes up. For answer engines, the second profile is the better position, and it is buildable without virality. The levers are the ones above: one name, one @id, and patient accumulation of independent corroboration. How that position shows up in measurement is the subject of citation rate.