An entity map is a documented model of the things your business consists of and how they relate to each other, expressed so a machine can read it. Your company is a thing. So are your people, services, locations and the subjects you claim to know about.
Search the term and you will find several competing frameworks with proprietary names, most of which describe the same underlying activity. Very few define what an entity map contains, and almost none explain how you would check whether yours is working.
This guide does both. It also includes a worked example using a real company, with figures you can verify against public records. It sits under our complete guide to AI search optimisation, which covers where entities fit in the wider picture.
Why entities replaced keywords
Classical search matched strings. If somebody typed a phrase and your page contained it, you were a candidate.
Retrieval systems reason about things instead. When a model decides whether to recommend a company it is checking whether it holds a coherent enough idea of that company to assert something about it. Text matching plays almost no part. No coherent idea means no recommendation, however good the copy is.
That explains a pattern which otherwise looks unfair. Two firms with comparable content and comparable links, one named constantly by AI assistants and the other never. The difference is usually that one is legible as an entity and the other is a collection of pages.
What an entity map actually contains
Four layers, in the order they matter.
Identity
Who you are, stated once and stated the same way everywhere. Legal name, trading name, registered address, incorporation date, registration number. These are boring and they carry more weight than anything else, because most of them can be checked against an external register.
Corroboration
Where else the same facts appear. This is what sameAs is for in schema markup. A list of verified profiles telling a machine that these separate accounts describe one organisation rather than several similar ones.
One page asserting something about you is a claim. Fifteen independent sources agreeing is a fact, as far as a retrieval system is concerned.
Relationships
How your entities connect. Who founded the company. Who writes the content. Which services belong to which parent category. Which locations you serve.
People are worth particular attention because individuals are easier to establish as entities than companies, and their credibility flows back. A named specialist with an external profile is worth more than an anonymous corporate byline.
Expertise claims
What you assert competence in, expressed as a specific list rather than a category. This is knowsAbout in schema terms, and it is where most entity maps quietly fail.
The specificity problem
Almost every business describes itself too broadly to be retrievable.
“Full service digital marketing agency” names a category rather than an entity. It gives a model no reason to select you over several thousand identical claims, because there is nothing in it that could only be true of you.
Narrow claims retrieve better precisely because they exclude people. “Paid search and marketing automation for UK aesthetics clinics” is a smaller market and a far stronger entity signal, because when somebody asks about exactly that there is a defensible reason to name you.
Breadth feels commercially safer. In retrieval it approaches invisibility.
A worked example
Rather than describe this abstractly, here is an entity map for an invented company. Brightwater Clinics is fictional and the values below are illustrative, but the shape is exactly what a real one looks like.
| Property | Illustrative value | What it would be checked against |
|---|---|---|
| name | Brightwater Clinics | Consistent across every profile |
| legalName | Brightwater Clinics Limited | Companies House |
| identifier | Company registration number | Companies House register |
| foundingDate | Incorporation date | Companies House |
| address | Registered office address | Companies House and Google Business Profile |
| sameAs | LinkedIn page, Companies House record, industry body listing | All publicly resolvable |
| founder | Named individual, linked to an author page | Their LinkedIn profile |
| knowsAbout | Four named clinical specialisms | Published content on each |
The important property is the address. In a working entity map it appears on the contact page, in the privacy policy, in the structured data and on the public register, identically. A machine can then check it without trusting the website.
Note what is absent. No awards, no certifications and no client names, because Brightwater cannot evidence them externally. A smaller entity map containing only checkable facts outperforms a larger one padded with unverifiable claims.
One caveat from practice. Where a registered office is a home address, think about where you repeat it. It belongs in the privacy policy and on the contact page because the law requires it. It does not need to appear in every article you publish.
The failure mode nobody writes about
While building the map above we found the Person node advertising a Cloudways staging domain as an associated URL. WordPress had picked it up from a user profile field nobody had edited since the site was first set up.
For however long that had been live, the structured data was telling every machine that read it that the founder was associated with a sandbox subdomain. Nothing on the visible page was wrong. The error existed only in the markup, where nobody looks.
Stale entity data is common and close to invisible. Old office addresses, former trading names, staging URLs, a founder who left in 2019. These quietly contradict your current claims and reduce a machine’s confidence in asserting anything about you at all.
How to build one
Step one, write down what is true
A spreadsheet is fine. Every fact about your organisation that could be stated publicly, with a column for where it can be verified. Anything without a verification source is a marketing claim rather than an entity property, and should be marked as such.
Step two, find the contradictions
Search your company name and read the first three pages properly. Directory listings, old profiles, review sites, industry bodies. Note every place a fact disagrees with your current position.
This is tedious and it is where most of the value sits. Contradictions do more damage than omissions. A machine handling conflicting information about you tends to assert nothing at all.
Step three, express it as schema
Use Organization for the company and Person for named individuals, with sameAs arrays pointing at profiles you control or that are publicly verifiable. Add identifier where a registration number exists, because it is the single most checkable property available.
If you already run an SEO plugin that emits an entity graph, extend its nodes rather than publishing your own alongside. Two Organization blocks describing one company gives crawlers contradictory claims, which does more harm than a single incomplete one.
Step four, correct facts where they originate
Where an external source carries a wrong fact, correcting your own site does not remove it. Ask the third party to update. Slower, and the only thing that works.
How to check whether it worked
Almost nothing published on this subject covers verification, which is odd given it is the only part that tells you anything.
- Ask each major AI assistant what your company does, who runs it and where it is based. Record the answers.
- Read the descriptions for factual errors rather than for flattery. Being described wrongly costs more than being absent.
- Check your rendered schema in a structured data validator and confirm every
sameAsURL still resolves. - Search your registration number and see whether your own site appears alongside the official record.
- Repeat quarterly, because entity data goes stale silently.
Description accuracy belongs on your reporting next to visibility. Being named correctly is what makes being named worth anything.
On the proprietary frameworks
Several agencies market named entity mapping methodologies. Read past the branding and they describe the same four layers set out above, in a different order and with fresh vocabulary.
That is not necessarily a criticism, because a named process is easier to sell and easier to follow. It is worth knowing that you are buying a packaging of public technique rather than a discovery, and that nothing in this discipline currently requires a proprietary model.
What nobody can prove yet
- How much weight
sameAscarries against plain unlinked mentions across the web. - Whether
knowsAboutinfluences retrieval directly or only tidies the graph for human inspection. - How quickly a corrected fact propagates through training data, as opposed to live retrieval.
- Whether a registration number materially helps, or simply happens to correlate with organisations that also do everything else properly.
- How much of the observed benefit is the entity map itself rather than the discipline required to build one.
That last question is the honest one. Documenting every fact about your business and correcting the contradictions would probably help even if no machine ever read the schema.
