A sitemap shows which pages exist on your website. What it does not show: which terms your brand owns, how you define them and where the evidence is. That is exactly what AI systems ask when they decide whether to name you in an answer.
In this chapter you get the entity map: what it is, what mine for bernard.digital looks like, what it uncovered and how to build yours in five steps.
Key takeaways
- Definition: An entity map records which terms a brand owns, how it defines them, where the evidence is and how the terms connect.
- Building blocks: entity, definition, evidence, relation.
- Benefit: It makes website, structured data, profiles and Wikidata consistent. LLMs benefit from that, even if they never read the map itself.
- Scope: 15 to 20 terms are enough for most brands. Depth beats breadth.
- File: There is a format, entitymap.json. There is no evidence that AI systems read it.
- Live example: My map for bernard.digital has 17 entities and uncovered three gaps.
On this page:
- What is an entity map?
- What does an entity map look like?
- What did the map uncover?
- How do you build your own entity map?
- What does the entitymap.json file do?
- FAQ
- Next step
- Sources
What is an entity map?
An entity map is an overview of the entities your brand should stand for. Each term comes with a definition, real evidence from your website and its relations to other terms. AI systems think in terms like these and their connections, not in single pages.
SITEMAP /en/ /en/geo-handbook/ /en/ai-visibility/ a list, no connections ENTITY MAP GEO ─depends on─▶ SEO ├─ achieves ─▶ AI visibility │ ▲ │ └─ measures: share of voice └─ evidence: passage + URL what your brand knows
An entity map has four building blocks:
- Entity: a term your brand owns, such as a service, a topic, a product or a person.
- Definition: one to three sentences on how you understand the term.
- Evidence: real passages from your website, with URL.
- Relation: how terms connect, for example “measures”, “depends on” or “offers”.
The map is the work, the file is a by-product.
In short: an llms.txt tells an AI which pages exist. An entity map tells it what your brand knows. The comparison is in the chapter llms.txt: do you need it?
What does an entity map look like?
This is the structure of my own map for bernard.digital: 17 entities, 35 relations, 23 pieces of evidence, all quoted verbatim from my pages. The interactive version, where you can click every entity, is in the German edition of this chapter.
BRAND bernard.digital ─offers─▶ ├─ GEO consulting ├─ SEO consulting └─ reputation management CONCEPTS GEO ─depends on─▶ SEO GEO ─achieves─▶ AI visibility share of voice ─measures─▶ AI visibility PRODUCTS AND CLIENTS kywrd.ai, GEO handbook Placetel (Q104655598)
Each entity gets a simple strength score. It shows at a glance which terms carry weight and which are only claimed:
BUILDING BLOCK POINTS Evidence 30 Definition 25 Relations 25 Wikidata anchor 20 Total 100 0-39 weak 40-69 developing 70+ strong example GEO: 90, kywrd.ai: 20
What did the map uncover?
While building it, three gaps came up that I had never noticed on the website itself. That is the real value: the map forces decisions.
GAP bernard.digital no Wikidata Bernard Zitzer no Wikidata THIN EVIDENCE GEO consulting █░░░░ 1 source kywrd.ai █░░░░ 1 source goal: 3 to 5 per core term MANY NAMES GEO, AI SEO, AEO, LLMO └─▶ one entity
No Wikidata entry
Neither I nor bernard.digital have a Wikidata ID. The two most important nodes of the map have no external anchor. Terms like GEO (Q134083964) or SEO (Q180711) have one, the brand does not.
Thinly evidenced services
GEO consulting and kywrd.ai each have only one piece of evidence. Core services and own products should have three to five: definition, example, number.
Four names for one field
GEO, AI SEO, AEO and LLMO are scattered across many pages. The map bundles them as alternate names of one entity, so they do not count as four weak signals.
My recommendation: start with the terms you earn money with. A service with only one piece of evidence is little more than a claim to an AI.
How do you build your own entity map?
You do not need a tool, a spreadsheet is enough to start. The order matters: define first, then evidence, then reconcile.
1 COLLECT 15 to 20 terms ▼2 DEFINE 1 to 3 sentences, fixed page ▼3 EVIDENCE 1 to 5 passages + URL ▼4 CONNECT measures, offers, part of ▼5 RECONCILE website = schema = profiles every conflict blurs the entity
- Collect terms: 15 to 20 terms your brand should stand for. Brand, people, services, core topics, own methods.
- Define: One definition of one to three sentences per term, on a fixed page of your website.
- Evidence: One to five passages with URL per term, ideally mixed: definition, example, number.
- Connect: Record relations. Which service pays into which topic, which metric measures what?
- Reconcile: Do website, structured data (Organization with sameAs), profiles and Wikidata match? Every contradiction blurs the entity.
Tip: you find contradictions fastest in numbers. On my own website one page said “five clients”, two others said “six”. Exactly this kind of mismatch blurs an entity.
Free tools
WikidataCheck whether your brand, your services and your topics already have an ID. These IDs belong in your structured data as sameAs.
Schema Markup ValidatorShows which structured data your pages deliver today.
EntityMap ValidatorChecks a finished map in entitymap.json format, no sign-up needed.For clients, I build the entity map as part of my GEO consulting. I am also building it into my own tool kywrd.ai, so the map stays current after every crawl.
What does the entitymap.json file do?
There is a file format for the map, entitymap.json. It was proposed by Fred Laurent and Dixon Jones (InLinks, Waikay) at entitymap.org. I use the format as a scaffold because it separates the four building blocks cleanly. The effect comes from the work before it.
WORKS TODAY ✓ clear definition per term ✓ uncovers contradictions ✓ reconciles schema + Wikidata ✓ template for briefings NOT PROVEN ? AI systems read the file ? adoption or measurement proposal by InLinks/Waikay
Watch out: there is no evidence so far that ChatGPT, Gemini or Google read entitymap.json, and the proposal comes from a vendor, not a neutral body. Publishing costs nothing, but promises nothing.
From my practice
I see the bigger lever in the entity map. The models probably do not read it directly either. But it creates clarity about the entities, on the client’s website and in their content, and that is exactly what LLMs benefit from.
FAQ
What is an entity map?
An overview of the terms a brand owns, with definitions, evidence from its own website and the relations between the terms. It makes content and signals consistent, whether or not you publish it as a file.
What is the difference between an entity map and a sitemap?
A sitemap lists pages so search engines can find them. An entity map describes terms and how they connect, so it is clear what a brand stands for.
How many entities does an entity map need?
For most brands 15 to 20, for large websites up to 30. A few well evidenced terms are worth more than many thin ones.
Is entitymap.json an official standard?
No. It is an open proposal at entitymap.org, initiated by Fred Laurent with support from Dixon Jones. Whether AI systems evaluate the file is not proven.
Do I need Wikidata for this?
Not necessarily, but it helps. A Wikidata ID as sameAs in your structured data gives an entity a clear external anchor.
Next step
I also build the entity map for your brand, as part of my GEO consulting: I crawl your website, define with you the terms your brand should own, and show you where definitions are missing, evidence is thin or facts contradict each other. You get a strength score per term, the list of gaps and an export as entitymap.json.
Result · Placetel
Organic visits per month
Want this for your brand? In 30 minutes we find out whether it can become the market leader.
Next in the handbook: With a clear map you know which terms your content has to carry. The next chapter is about content that AI cites. Until it is out, the handbook overview shows what comes next.
Sources
- EntityMap: Specification v1.0, Fred Laurent and Dixon Jones
- Schema.org: Organization (sameAs)
- Wikidata: Generative engine optimization (Q134083964)
