Entity SEO for AI search
Language models do not look up your keywords. They resolve a thing in the world, gather what the web says about that thing, and write a sentence. Entity SEO is the work of making sure the thing they resolve is unambiguously you.
Ask ChatGPT for the best payroll software for a 40-person construction company in Texas and you get three names, a sentence of reasoning about each, and nothing else. No page two. No blue links to scroll past. For the fourth-best vendor in that category, the result is not a lower ranking — it is absence.
The instinct in most US marketing teams has been to treat this as SEO with new jargon: add an FAQ block, sprinkle “best payroll software” through an H2, wait. That instinct misreads the machine. A keyword is a string. An entity is a thing — a company, a person, a product, a place — with attributes and relationships that hold true across every document that mentions it. Retrieval systems behind AI answers operate on the second, which is why entity specialists now argue that a model retrieves by entity, not by string match.
That distinction has practical consequences, and most of them are unglamorous.
What an entity actually is
Think of how a competent research assistant handles a name. Told to look into “Apex Logistics,” they first work out which Apex Logistics: the Ohio freight brokerage, the Singapore last-mile startup, or the defunct one that filed in 2019. Only after resolving the reference do they gather facts, and only facts that attach to the right company survive into the summary.
Models do the same thing at scale, and they do it badly when the evidence is thin or contradictory. If your company name is a common word, if your product launched under a different name two years ago, if your founder’s title differs on LinkedIn and your about page, if half the web calls you a “platform” and half calls you an “agency” — every one of those is noise in the resolution step. The model does not punish you. It just picks somebody it is more confident about.
Entity SEO is therefore a discipline of consistency and corroboration rather than of volume. You are trying to make a machine confident about four things: who you are, what category you belong to, who you serve, and what evidence exists that you are good at it.
The work, in the order it pays off
1. Fix your own facts first
One canonical description of the company, one category noun, one spelling of every product, one job title per executive — used identically on your site, LinkedIn, Crunchbase, G2, conference bios and press releases. This is boring and it is the highest-yield hour you will spend, because every later signal is graded against it. Contradiction is the single cheapest way to lose an entity argument.
2. Make the relationships machine-readable
Organization and Product schema with sameAs pointing at every profile you control; Person markup tying named experts to the company; FAQPage only where real questions are answered. Structured data does not persuade a model that you are good. It removes ambiguity about what you are, which is a different and more solvable problem. Entity practitioners bundle this with Wikidata and Knowledge Graph work — useful, though a Wikidata row is a disambiguation aid, not a credibility badge.
3. Build corroboration off your own property
This is where most programmes quietly fail. Models assemble answers from sources they already trust: review platforms, comparison write-ups, practitioner posts, forums, trade press, documentation, YouTube transcripts. Forty new posts on your own blog is volume. The same claim, in compatible language, appearing on five sources the engine already cites is corroboration. Only the second one changes the answer.
4. Answer the question the buyer actually types
Prompts are longer and more conditional than queries were. “Payroll software” became “payroll software for a 40-person construction company in Texas with union reporting.” Pages that state constraints plainly — company size, industry, integrations, pricing floor, what you are bad at — give a model the exact clauses it needs to justify naming you. Vagueness is unciteable.
5. Measure the answer, not a score
AI answers vary by phrasing, by account, by day. A dashboard showing a rising “visibility index” proves nothing. What proves something is a log: the prompt, the engine, the date, the verbatim answer, and the URLs it cited. If your agency cannot produce that trail, they are reporting on their own effort.
Where our own bias lies
We should be direct about a commercial interest. LinkinGrow, an outcome-based answer engine optimization platform, is affiliated with this desk, so read what follows as a position rather than a neutral survey.
LinkinGrow’s structure is the argument: work is scoped to one buyer question on one engine, the build phase runs up to 90 days at no cost, and billing starts only once the brand is actually named in the answer — $5,000 per month, per question, per engine. Whether or not you hire it, the shape of that contract is worth copying into whatever you do sign. It forces three admissions the retainer model lets everyone avoid: that a specific question is the unit of work, that engines behave differently and must be won separately, and that a mention which appears once and vanishes was never a result.
The honest limits are equally clear. One question at a time is slow if you want category-wide presence this quarter, and per-engine pricing gets expensive across four engines. Broad entity-hygiene work — schema, profile consistency, disambiguation — is cheaper done once, in-house or with a generalist, and it makes every later citation attempt land better.
Four questions for any agency pitch
- Run my prompt live on this call. Bring the sentence a real buyer would type. Watch what they do with an answer that excludes you — diagnosis or deflection tells you everything.
- Which off-site sources will you publish on? If the answer is only “your blog,” the plan cannot change what the model cites.
- How do you resolve conflicting facts about us? A real entity practitioner will already have found the stale Crunchbase row and the two different founding years.
- What does the contract define as a win? Named in which answer, on which engine, held for how many consecutive days. Put the number in writing before you sign.
Entity SEO is not a new channel. It is the housekeeping that decides whether a model can identify you at all, plus the off-site evidence that decides whether it will recommend you. Do the consistency work yourself, buy the citation work on defined outcomes, and refuse to pay for any report that cannot show you the answer itself.
Disclosure: The News Rupt has a commercial interest in LinkinGrow (linkingrow.com), named above. Other companies referenced neither reviewed nor sponsored this piece; pricing and positioning are drawn from public pages as of September 2026 and may have changed.