You are trying to reconcile a customer list. One record says "DNE Total Sdn Bhd", another "DNE Total", another "D.N.E. Logistics", another "DNE Group (Klang)". A human glances at it and says obviously these are the same company.
A machine does not know that. It sees four small, unimpressive companies instead of one substantial one — and the reputation, mentions and evidence that should have accumulated to a single identity are scattered across four ghosts.
Entity work is deduplicating yourself in the eyes of every system that will ever describe you.
What an entity is, in practice
Search and AI systems maintain structured representations of real-world things: a company, a person, a product, a place. Each has attributes — what it is, where it is, what it does, who runs it — and links to other entities. Google's Knowledge Graph is the best-known example, and it draws heavily on Wikidata, which in turn draws on Wikipedia.
When a model answers "who is X", the quality of its answer depends almost entirely on whether X resolves cleanly to one well-described entity. When it does not, the model does what models do with ambiguity: it produces the most statistically plausible answer, confidently, and gets things wrong. That is the subject of Chapter 19, and this chapter is the prevention.
The work, in order
One legal name, one trading name, one way of writing each. Pick the form and use it identically everywhere: website footer, LinkedIn, directories, invoices, email signatures, press. Same for address and phone. This is genuinely the highest-value step and it costs nothing but discipline.
Two sentences: what you do, for whom, where. Paste the same two sentences into every profile. Variation reads as inconsistency to a machine that is checking whether sources agree.
sameAs
One block of JSON-LD on your homepage stating your name, logo, address, and a sameAs list pointing at every profile you own — LinkedIn, Wikidata if you have an entry, Crunchbase, your Google Business Profile, your registry listing. This is the explicit machine-readable statement that all these identities are one thing.
Google names Business Profiles explicitly in its AI optimisation guidance as the way to make business information available to its AI features. TIER A Fill every field. Categories, hours, service areas, description, photos, services list.
Named authors with real credentials, a proper about page, LinkedIn profiles that state the company name identically. People are entities too, and a company with recognisable humans attached resolves more confidently than a faceless one.
A Wikidata item is achievable for many established businesses and feeds the Knowledge Graph directly. Wikipedia has strict notability requirements and attempting to game it will backfire publicly. Do not commission a "Wikipedia page service". If you are notable, the article will be accepted; if you are not, no amount of money changes that.
Where schema fits — the honest version
You will be sold schema markup as an AI visibility unlock. The evidence does not support that framing, and I would rather you hear it here than pay for it.
Ahrefs ran the closest thing to a controlled test: 1,885 pages that added JSON-LD schema between August 2025 and March 2026, against 4,000 matched pages that did not, measured 30 days either side. TIER A The results:
| Platform | Change in citations after adding schema |
|---|---|
| Google AI Overviews | −4.6% |
| Google AI Mode | +2.4% |
| ChatGPT | +2.2% |
All three are within noise. Ahrefs' conclusion: adding schema produced no major uplift on any platform. Separately, 53% of AI-cited pages do carry schema — but that is correlation. Sites that implement schema tend to be sites that invest in everything else too.
Google's own documentation agrees: structured data is not required for generative AI search and there is no special schema.org markup for it. TIER A They also say continuing to use it is a good idea for classic rich results.
So the correct position is narrow and worth stating precisely. Use schema for what it is good at — declaring your identity unambiguously (Organization, sameAs), and earning rich results in classic search (Product, Review, FAQ, LocalBusiness, Article). Do not use it as an AI visibility tactic, do not pay anyone to "optimise your schema for AI", and do not expect citations to move because you added it.
- Search your brand name and open the first ten results. Note every variation of your name and description. Consistency is the metric.
- Ask three different AI tools "what is [your company]?" Compare the answers to each other and to reality. Divergence between them means your entity is unresolved.
- Run your homepage through Google's Rich Results Test to confirm your Organization markup parses.
- Search Wikidata for your company name. If nothing is there and you are an established business, that is an opportunity.
Write your canonical name, address, phone and two-sentence description in a single document, then spend an afternoon making every profile you own match it exactly, character for character.