Entity & Knowledge Optimisation
Making sure every machine that looks you up finds the same company
Search engines and language models resolve you to an entity before they decide anything else about you. If your name, category, location and ownership are described three different ways across the web, you are three weak entities rather than one clear one.
What we find
Four ways a company stops resolving as one entity
Entity problems are rarely dramatic. They are accumulations of small inconsistencies, each individually defensible, which together make it hard for a machine to be confident about who you are.
- Somebody else has a name close enough to yours.
- A similar name in the same country, or an identical name in a different sector, is enough to split a record. Mentions get attributed to the wrong organisation, and neither company gets the benefit. Disambiguation here is a matter of making the differences explicit and machine-readable: legal name, registration number, sector, location, founding date, leadership.
- The rebrand is only half done.
- The website changed immediately. Company registries, directories, industry bodies, archived coverage, conference programmes and partner pages did not. Years of accumulated corroboration is therefore still attached to a name you have stopped using, while the new name has comparatively little behind it. This is the most expensive and most common version of the problem.
- You describe yourself four different ways.
- The homepage says one thing, the about page says something broader, the footer boilerplate is three years old and the LinkedIn description was written by someone who has left. None of them is wrong exactly, and together they make the category you belong to genuinely ambiguous. A machine deciding which businesses to include in an answer about a category needs to be able to place you in one.
- The markup claims more than the page shows.
- Structured data asserting awards, ratings, prices or services that appear nowhere on the visible page. The published guidelines are explicit that markup should represent visible content, so at best this is ignored and at worst it attracts a manual action. It is also the single most common thing we remove rather than add.
The work
Four parts, in the order they pay off
One canonical description
A single agreed paragraph describing what the business is, what it does, who it serves and where. Written once, approved once, then rolled out to every profile, listing, registry entry and press boilerplate. Unglamorous, cheap, and the largest single improvement available to most companies.
Deliberate disambiguation
Making the differences between you and anything similarly named explicit: legal name, registration number, founding date, leadership, sector and location, stated consistently and marked up where the page shows them. This is what stops mentions being attributed to the wrong organisation.
References you actually own
The sameAs property links your organisation to references that unambiguously identify it — official profiles, registry entries, verified accounts. It is a disambiguation aid rather than a ranking lever, and pointing it at pages you do not control or that are about something else makes the record noisier, not stronger.
Corroboration you do not control
Independent sources describing you the same way. This is the slowest input and the one that actually decides whether an entity is well established, which is why the honest version of this service spends more time on earned coverage than on markup.
The uncomfortable part
Knowledge panels, Wikipedia and what is actually available to you
This is where most conversations about entity work go wrong, because the two things clients most often ask for are the two things nobody can sell them.
A knowledge panel is generated from information a search engine has collected and is confident about. There is no submission form. If the corroboration does not exist, the panel does not appear, and no amount of markup manufactures the confidence that is missing. Where a panel already exists and shows something wrong, there are official channels for an authorised representative to suggest a correction, and we use them. That is the extent of what is available.
Wikipedia is stricter still, and correctly so. An article requires significant coverage in independent, reliable sources that are not connected to the subject. Editing about an organisation you have a financial relationship with is discouraged by policy, and paid editing carries a disclosure obligation under the terms of use. Undisclosed paid editing gets accounts blocked and articles deleted, and the reputational damage of being caught considerably exceeds the value of the article.
So our rule is simple and absolute: we do not edit Wikipedia on behalf of a client, we do not commission anyone else to, and we do not treat a Wikipedia article as a deliverable. What we do instead is the legitimate version of the same objective, which is making sure the independent coverage exists. If it does, an editor may eventually write about you. If it does not, the honest answer is that the article should not exist yet.
Wikidata is a different case and worth understanding separately. It is an openly editable structured database with its own notability rules, and correcting a factual error on an existing item, openly and with proper references, is a reasonable thing to do. Creating items that assert facts no published source supports is not, and we will not do it.
Rules
What we will and will not do with your entity data
What we do
- Write one canonical description and roll it out to every property you control
- Correct wrong or stale facts at the specific source that is publishing them
- Mark up organisation, location and person data that the page visibly shows
- Use sameAs only for references you genuinely own or that unambiguously identify you
- Use the official correction channels where a knowledge panel already exists and is wrong
- Build the independent coverage that makes an entity resolvable in the first place
- Tell you plainly and early when a knowledge panel is not realistic for your business
What we will not do
- Edit Wikipedia articles about our clients, or commission anyone else to do it
- Create Wikidata items that assert facts no published source supports
- Mark up awards, ratings, prices or services the visible page does not show
- Invent a schema type, or claim that any markup causes an AI citation
- Present llms.txt, or any other file, as a mechanism that produces rankings or mentions
- Promise a knowledge panel, a Wikipedia article, or a specific description in any system
Questions
The questions this work always raises
What is an entity, in plain terms?
A thing with an identity rather than a word on a page. Your company, your founder, your product and your office are entities. A search engine or a language model tries to work out which specific organisation a mention refers to, then attaches what it learns to that record.
The practical consequence is that everything said about you either accumulates onto one record or gets split across several. Split records are weak records, and a weak record is easy to leave out of an answer or to confuse with somebody else.
Can you get us a Wikipedia page?
No, and we will not arrange one through anybody else either. Wikipedia requires significant coverage in independent, reliable sources, and its policies discourage editing about subjects you have a financial relationship with. Paid editing carries a disclosure obligation under the terms of use, and undisclosed paid editing gets accounts blocked and articles deleted.
What we will do is the legitimate version: help build genuine independent coverage. If enough of that exists, an editor may write about you. If it does not, no agency can honestly change that, and the ones offering to are creating a liability rather than an asset.
What about Wikidata? Can we add ourselves there?
Wikidata is a structured, openly editable database with its own notability rules, and they are different from Wikipedia’s. An item generally needs to be either the subject of a page on a Wikimedia project, or a clearly identifiable entity that can be described using referenced, serious sources.
Where a client genuinely qualifies, correcting factual errors on an existing item with proper references is a reasonable thing to do openly. What we will not do is create items asserting facts that no published source supports, or treat Wikidata as a back door to a knowledge panel.
Can you guarantee a knowledge panel?
No. Knowledge panels are generated from information a search engine has gathered and is confident about, and there is no submission form that creates one. If the underlying corroboration does not exist, the panel does not appear.
What we can do is make the facts consistent, unambiguous and well supported across the sources a search engine draws on, and use the official channels to suggest corrections where a panel already exists and is wrong. If a panel is not realistic for your business, we will say so early rather than billing against it.
Is this not just adding schema markup?
Markup is maybe a fifth of it, and it is the fifth most people do first. Structured data helps a machine parse what is already on the page. It cannot make a claim true, and the published guidelines are clear that it should represent content visible to users.
The larger part of the work is off your site: registries, directories, profiles, coverage and third-party descriptions of you, all of which need to agree. That is unglamorous, and it is where the actual improvement comes from.
We rebranded two years ago. Is that still a problem?
Usually yes, and it is one of the most common findings. Your own site changed on day one. Company registries, directories, industry bodies, old coverage, conference listings, partner pages and other people’s footnotes did not.
That means the corroboration you spent years building is still attached to a name you no longer use. Finishing a rebrand is a source-by-source job with a tracked list, and it is the single highest-value entity project for most companies in that position.
Find out how many versions of your company are in circulation
We will pull every description of your business a machine could reach and put them side by side. Most companies are surprised by how many there are, and by how old the oldest one is.
Related
Where to go next
- AI search optimisation services
- brand messaging and boilerplateThe canonical description has to be written before it can be rolled out.
- local SEO and business profilesWhere physical-location entity data lives and goes wrong.
- reputation managementThird-party descriptions are entity data whether you like it or not.
- generative engine workThe service that depends most directly on this one.
- our content review policy
Last updated · Reviewed by Zubair Afzal