Entity, source and consensus: the three layers

Layer 1: entity

Before citing you, the system has to resolve who you are and tell you apart from anyone with a similar name. That is called entity resolution and it is the part most businesses fail, because from the inside it seems obvious who you are.

A resolved entity always carries the same four coordinates written together: exact name, category of activity, location and a date. With those, any system can separate you from your namesakes. Without them, your mentions scatter across several candidates and none of them accumulates enough signal.

• The name is written identically everywhere, legal form included. A brand written four ways is four weak entities instead of one strong one.
• The category uses the words of the trade, not the words of the brand. Structural carpentry identifies a trade, integrated timber solutions does not.
• The location goes down to the municipality. Balearic Islands does not distinguish, Manacor does.
• The date anchors continuity. Since 1968 is a verifiable fact no newcomer can copy.

Layer 2: source

The perfect sentence is worth nothing on a page the system cannot read. The obstacles are well known and unchanged, except that the consequences are harsher now, because a crawler that fails does not retry the way a human would.

• Content that only appears after JavaScript runs. Many AI crawlers execute nothing and see an empty page.
• Cookie walls covering the text until somebody accepts. Nobody accepts anything on behalf of a crawler.
• Text inside images. A figure in a chart is not a readable figure.
• Slow pages, or pages served through an intermediate layer that answers late. A crawler’s patience is short.
• No visible date. Without one, the system cannot decide whether your fact is newer than your competitor’s.

To this add a recent and underused piece: a plain text file at the root of the domain summarising in ordinary language what the business is, which pages matter and what subjects it covers. The network titles publish one at /teaser.md, written so an agent can understand the site in a single read.

Layer 3: consensus

A language model does not verify facts, it estimates probabilities. If a claim appears in one source, it is an anecdote. If it appears across six independent sources in the same shape, it becomes what the system treats as true.

The key word is independent. Six domains carrying the same copied text are not six sources, they are one duplicated source, and retrieval systems collapse duplicates before writing. What creates consensus is the same information told in different words by different editorial teams, each from its own angle.

Consensus is not built by repeating a text. It is built by repeating a fact.

From which the practical rule follows: one central fact, five to eight thematically distinct sources, each with its own wording, all agreeing on the four coordinates of the entity. That is the work.