What entity authority is, and why it decides whether the AI cites you
Before recommending a brand, an engine has to recognise it as an entity. Entity authority is the work of building that recognition, and it weighs most in the model's training memory.
Open in ChatGPT
Open in ClaudeAn AI engine does not treat your brand as a set of letters. It treats it as an entity: a node with an identity of its own, linked to a sector, to people, to a place, to other entities.
Before recommending a brand, an AI engine needs to recognise it as an entity: to know who it is, what it does, and to have that confirmed by independent sources. Entity authority is the work of building that recognition. It is what separates a brand the model cites with confidence from one it ignores or confuses, and it weighs above all in the models' training memory.
- Models cite entities they recognise, not strings of text
- Consistency of description across every platform is the base signal
- It weighs more in the training memory than in live web search
- For a small company, it starts at Wikidata and schema with sameAs, not at Wikipedia
From text to entity
When somebody asks for a company that does what yours does, the model does not look for the string of your name; it looks for entities it recognises as an adequate answer, and recommends the ones it can pin down with confidence.
Entity authority is the degree of that confidence. A brand with high entity authority is described the same way everywhere, has its official profiles linked to one another, appears in sources the model considers reliable, and is not confused with namesakes. A brand with low authority is one the model barely knows, describes hesitantly, or swaps for another.
Why it decides the citation
It is worth separating the two modes in which an engine answers, because the entity weighs differently in each. In live web search the engine looks for and reads pages in the moment; here what is published and reachable counts for more than the model's memory.
In the answer from memory, with no search, the model falls back on what it absorbed in training. And what it absorbs clearly is what it saw described consistently across many sources. A brand with a contradictory identity between the site, LinkedIn, the directories and the press gives the model conflicting signals, and the result is hesitation or omission. A brand described the same way everywhere settles as a stable entity. That is why entity authority weighs above all in the training memory, the layer hardest to move.
The signals that build it
Entity authority is not bought and is not declared; it is built with signals the models verify. The main ones, in order of how reachable they are for a Portuguese company:
Consistency of description
The same name, the same activity, the same address and contacts, on every public platform. It is the cheapest signal and the most ignored.
Structured data with sameAs
Organization schema on the site, with the
sameAsproperty linking to the official profiles (LinkedIn, sector directories, Wikidata). It tells the model, unambiguously, that these presences are the same entity.A Wikidata record
Reachable by any company and read heavily by the models. It is the first structured step towards pinning the brand down as an entity.
Third-party mentions
Independent coverage, citations in the sector's reference sources, presence in real discussions. The entity gains weight when it is not only the company talking about itself.
Where a small company starts
The temptation is to aim at Wikipedia, and for most that is the wrong advice. Wikipedia demands notability demonstrated by independent coverage, it is out of reach in the short term for most small companies, and promotional articles get deleted, which burns credibility on the platform.
The realistic path is the reverse. First, sort out the consistency: one single description of the company, replicated everywhere. Then the schema with sameAs on the site and the Wikidata record, both reachable right away. Only when press coverage justifies it does Wikipedia enter the conversation, and never forced. The entity is built from the bottom up, with patience, and it is precisely the kind of advantage that consolidates rather than evaporating overnight.
Entity authority is not bought and is not declared; it is built with signals the models verify.
Frequently asked questions
It is the degree to which the engines recognise a brand as a verifiable entity: who it is, what it does, confirmed by independent sources. Before recommending a brand, the model needs to pin it down as an entity. Entity authority is the work of building that recognition, with consistency of description, structured data and presence in reference sources.
It is not compulsory, but what Wikipedia stands for does matter: an entity described by third parties, with sources. For most small companies Wikipedia is out of reach in the short term. The reachable path is Wikidata, schema with sameAs, and brand consistency across every platform. Wikipedia comes when press coverage justifies it, and should never be forced.
Because the model's training memory is made of what it absorbed consistently across many sources. A brand with a contradictory identity between platforms confuses the model; a brand described the same way everywhere settles clearly. In live search the entity weighs less, because the engine reads the web in the moment, but in memory it is decisive.
See also: where the AI learns about your brand and the two answer modes.