Every generative answer starts with a resolution step nobody optimizes for. Before an engine decides whether to cite you, it has to decide what you are: which company, which product, which of the four organizations sharing your name, and which set of facts attaches to that node. Entity SEO is the work of making that decision fast and correct. Get it wrong and the best content on your site is filed against the wrong thing, or against nothing at all.
Read those four numbers as one finding. The things that correlate most strongly with being named in an AI answer are the things that describe a brand rather than link to it, and the pages that get cited most often are the ones that state plainly what an organization is. That is an entity problem wearing a content costume, and it is why entity SEO has moved from a niche technical interest to something enterprise teams now need a named owner for.
Chapter 1: What an entity is, and why resolution happens before retrieval
An entity is a uniquely identifiable thing: a company, a person, a product, a place, a concept. It is not a keyword. The phrase acme analytics is a string. The company Acme Analytics, founded in a particular year, headquartered somewhere, making a specific product, employing a specific chief executive, is an entity. An engine holds the second thing as a node with attributes and relationships, and it holds the first thing only as text that might point at that node.
The order of operations matters more than the definition. When a buyer asks which analytics platform handles server-side tagging best, the engine does not go looking for pages containing that phrase. It resolves the concepts in the question into entities, works out which entities are candidates for the answer, retrieves supporting material about those candidates, and only then generates prose. Retrieval is downstream of resolution. If your organization is not a well-formed node with clear attributes, you were filtered out one step before any of your content was considered.
This is also why two teams can look at the same ranking report and reach opposite conclusions about their AI visibility. Ranking measures whether a document competes for a string. Citation measures whether an entity survived a shortlist. They are different events with different inputs, which is the same separation we argued for in the case for treating grounding, citation and mention as distinct. A page can rank first and never be shortlisted, because the ranking was earned by the document and the shortlist is decided about the company.
Chapter 2: The entity home and the fact set that has to agree
Every entity needs one canonical page that an engine can treat as the authoritative statement of what the thing is. For a company that is usually the homepage or the about page. For a product it is the product page. For a person it is a real author or team profile, not a byline that points nowhere. That page is the entity home, and its job is different from every other page on the site: it is not selling, it is defining.
The Ahrefs finding that 23.8 percent of ChatGPT's top 1,000 citations go to homepages rather than deep content is the clearest evidence available that engines reach for definitional pages when they need to establish what something is. Most enterprise homepages are written as a value proposition and contain almost no extractable fact. That is a straightforward thing to fix and it is usually the single highest yield change in an entity program.
| FACT | WHERE IT MUST APPEAR | COMMON FAILURE |
|---|---|---|
| Legal and trading name | Entity home, Organization schema, third-party profiles | Trading name on the site, legal name everywhere else, no link between them |
| What the company does, in one sentence | First 100 words of the entity home | Replaced by a slogan that names no category |
| Category the company competes in | Entity home body copy and title | Invented category language no buyer or engine uses |
| Founding year and headquarters | Entity home, schema, directory profiles | Absent from the site, present and wrong on three directories |
| Named leadership | Team page with individual profiles | A grid of photographs with no structured detail |
| Products, as named things | Dedicated pages, one per product | One page listing six products, so none of them resolve |
| Verified external profiles | sameAs references from the entity home | Present in the footer as links, absent from the structured data |
The single mechanic worth implementing first is sameAs. In Organization schema it is the list of other places on the web that unambiguously refer to the same entity: the company's LinkedIn page, its Crunchbase entry, its Wikidata item where one exists, its verified social profiles, its listings on the review platforms buyers actually use. It is the closest thing available to telling an engine directly that these scattered records describe one thing. It is also frequently missing on sites that have otherwise complete markup, because it is the one property that requires someone to go and collect real URLs rather than fill in a field from the CMS.
The markup layer here overlaps with, but is not the same as, the work of making individual pages extractable. We set out the implementation standard for that separately in the enterprise standard for structured data in AI search. The distinction worth keeping: structured data describes a page, entity work describes a thing that exists whether or not the page does.
Chapter 3: Corroboration, the part of entity SEO you do not control
An entity that only your own site asserts is a claim. An entity that a dozen independent sources describe the same way is a fact. Engines behave accordingly, and this is where the correlation data gets uncomfortable for anyone whose authority program is measured in referring domains.
Correlation with AI brand visibility by signal type. Machine Relations Research, Spearman correlations across 75,000 brands, published July 2026, with the YouTube and mentions figures corroborated by Ahrefs' June 2026 analysis.
A 0.664 correlation for branded mentions against 0.218 for backlinks does not mean links stopped working. It means the link was always a proxy for something else, and engines that read text directly no longer need the proxy. A paragraph on a trade publication that describes what your company does, with no hyperlink at all, feeds entity resolution. A footer link from a directory with no descriptive text feeds almost nothing. We walked through the budget consequences of that in the case for shifting spend from backlinks to earned coverage.
Interest in the discipline has moved accordingly. Search Engine Land reports search interest in entity SEO growing by more than 1,000 percent, alongside a finding that three quarters of SEO leaders now believe backlinks influence appearance in AI answers even as the measured correlation for raw link counts sits near 0.2. That gap between belief and measurement is the honest state of the field, and it is why the practical advice here is to keep doing the earned media work while changing what you brief it to achieve. Search Engine Land's guide to link building under AI search makes the same case from the outreach side: diversity of description beats volume of links, and an unlinked mention that describes you accurately can be worth more than a followed link that says nothing.
There is a hard limit worth stating plainly. A meaningful share of the sources engines lean on hardest are ones no brand can influence directly. That is not a reason to skip the work. It is a reason to concentrate the influenceable effort on the places where the same facts can be stated by someone who is not you, which is a different brief from the one most link building programs are running. The distribution of those uninfluenceable sources is set out in our read of the citation source concentration data.
Chapter 4: Disambiguation when your name is not unique
Plenty of companies share a name with a town, a chemical, a film, a football club or three other companies. Ambiguity is the most damaging entity problem and the least diagnosed, because it does not look like a problem in any standard report. Your rankings are fine. Your traffic is fine. You are simply absent from answers where a differently named competitor appears, and there is no line in any dashboard called lost to name collision.
The test takes two minutes. Ask three engines to describe your company by name alone, with no other context, and read what comes back. Then ask again with the category attached. If the first answer describes something else entirely and the second is correct, you have a disambiguation problem rather than a coverage problem, and more content will not fix it.
| SYMPTOM | LIKELY CAUSE | FIX |
|---|---|---|
| Answer describes a different organization with your name | Stronger entity holds the bare name | Bind name to category in every definitional sentence, everywhere |
| Answer merges your facts with another company's | Two entities not separated in the graph | Distinct sameAs sets, distinct addresses, distinct leadership named |
| Answer is correct but generic | Entity resolved, attributes thin | Add checkable specifics to the entity home and the profiles |
| Cited for the brand, never for the category | Entity known, not associated with the category | Publish category-defining content under the entity, not just product copy |
| Product credited to the wrong company | Product entity not bound to organization entity | One page per product, explicit organization reference in markup and copy |
| Correct in one engine, wrong in another | Different corpora, different resolution | Fix the source each engine leans on rather than the site alone |
The general remedy for ambiguity is binding: never let the bare name travel alone in a definitional context. Acme in a headline is ambiguous. Acme, the server-side analytics platform, is not. Applied consistently across the entity home, the schema description, the profiles and the boilerplate that trails every press release, that pairing is what teaches an engine which node the string belongs to. It reads as slightly repetitive to a human and it is the whole mechanism to a machine.
“If an engine cannot tell you apart from a football club, no amount of content is going to be filed against the right company.”
Chapter 5: Measuring entity strength when no tool reports it
There is no entity strength metric. Anyone selling one has built a composite from proxies, and the proxies are more useful than the composite. Four measurements, run on a fixed cadence, tell you what you need without buying anything new.
First, description accuracy. Run a fixed prompt panel that asks each engine to describe the company, its category, its products and its customers, and score the responses for factual correctness rather than sentiment. Track the error types, not just an error count, because a wrong founding year and a wrong category are different failures with different fixes. Second, attribute coverage: of the facts you have decided matter, how many can an engine state unprompted. Third, source agreement: pick the ten highest-traffic third-party records describing your company and score whether they agree with each other. Fourth, resolution latency for new facts, meaning how long after a launch or a rebrand the engines start repeating the new version.
| MEASURE | HOW TO RUN IT | CADENCE | WHAT A BAD RESULT LOOKS LIKE |
|---|---|---|---|
| Description accuracy | Fixed prompt panel across four engines, scored by a human | Monthly | Category named incorrectly, or facts drawn from a predecessor company |
| Attribute coverage | Checklist of the facts that matter, marked present or absent per engine | Quarterly | Fewer than half the facts recoverable without a leading question |
| Source agreement | Ten highest-traffic third-party records, compared field by field | Quarterly | Three different employee counts and two different headquarters |
| Resolution latency | Date a fact changed against date engines repeat it | Per launch or rebrand | Beyond one quarter, which usually means no corroborating source carried the change |
Resolution latency is the one worth building a habit around, because it is the measurement that turns entity work into an operational discipline rather than a project. A rebrand that engines are still describing with the old name six months later is not a branding problem or a content problem. It is a corroboration problem: nobody outside the company restated the change in a place the engines read.
One more measurement worth adding once the first four are running: contradiction count. Take the facts an engine states about you and check each one against the record it most likely came from. A wrong employee count that traces to a stale directory profile is a fixable supply problem. A wrong category that traces to nothing identifiable is a resolution problem and needs corroboration rather than correction. Separating those two failure modes is what stops an entity program from turning into an endless round of correcting individual answers, which is not a strategy and does not scale past the first few.
Resist the urge to build a single score out of these. A composite hides exactly the information you need, which is which of the four is failing. A company can have excellent source agreement and terrible attribute coverage, which means the world agrees on a thin description, and the fix is publishing more checkable specifics. Another can have rich attributes and poor agreement, which means the site says a great deal and nobody else repeats any of it, and the fix is entirely external. Those two situations produce a similar composite score and require opposite quarters of work.
The 30 day entity audit
This sequence assumes one person at roughly half time, no new tooling, and an existing site. It is deliberately ordered so that the cheapest structural fixes land before any outreach effort is spent.
Two expectations worth setting before anyone commits to this. Entity changes propagate slowly, on the order of weeks to a quarter, because they depend on corpora being refreshed rather than a page being recrawled. And the first visible win is almost always accuracy rather than volume: engines start describing you correctly before they start naming you more often. A program judged on citation count at week four will be judged as a failure at exactly the point it is working.
The internal structure of the site does still matter to all of this, because discovery and categorization run over links even when resolution runs over facts. The companion piece on internal linking for AI search covers that layer. And where entity work has moved the needle fastest for us it has been in technical categories with crowded naming, the pattern behind the Arnica engagement, where clarifying what the company was preceded any change in what got cited.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Tyler leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.