AI SEO · 26 September 2026 · 9 min
Entity SEO for ChatGPT and Perplexity: make the brand the source they quote
Entity SEO for ChatGPT and Perplexity starts with facts, not prompts. How brand mentions and consistent pages become the source that gets quoted.
Entity SEO is how a brand becomes a thing a model can resolve: one name, one offer, one set of facts, repeated until the system does not have to guess which company is speaking. ChatGPT, Perplexity and Google's summaries assemble an answer from pages they can retrieve and from descriptions that agree. If the name is generic, the about page is a slogan, and a directory still lists a product you retired, you will be confused with a rival or left out.
Entity SEO starts with a fact sheet, not a prompt
Start with a one-page record a stranger could check. Legal name, trading name, what you sell, who it is for, where you operate, what you do not do, and the proof points you will repeat for a year. Named people, if they are why a buyer should trust the firm. The canonical URL for each major offer. Kill the adjectives. "Innovative global partner" is not a fact. "Commercial SEO retainers for multi-market sites, from a desk in Birmingham" is.
That sheet becomes the source for the about page, the footer, the organisation details in your schema, the author bios, and the opening of each service URL. When those surfaces disagree, every system that reads you learns to hedge, and hedging is how a model names the category and cites a publisher instead. ChatGPT SEO, in practice, is less about whispering into a prompt and more about removing contradictions a retrieval system would otherwise average away.
Keep the sheet boring. Package names change; the identity should not. A real rebrand has to land everywhere in the same week, including profiles you control. A half-finished rename is why answers keep using the old name.
What an answer engine has to resolve
A model quoting the web has a quieter question first: which entity is this? Shared names make it hard. A manufacturer, a software tool and a studio can occupy the same two words. Attach a stable modifier and use it where the bare name would be ambiguous: the title, the opening sentence, the organisation name you publish. Do not stuff it into every heading. Do put it where a citation has to tell you apart from the other result.
Perplexity citations and Google's overviews are retrieved. They prefer a URL that states the claim, and a web that repeats it. An ungrounded chat answer is different. It may describe you from older pages, or name nobody. You cannot edit an ungrounded memory on a timetable, and you should not pay anyone who claims they can. Make the crawlable record consistent enough that a system which looks you up finds one description.
Ask the buyer's question, not your slogan. "Who handles international SEO for a UK firm selling into the Gulf?" is useful. "Best SEO agency" is a popularity contest you will misread. Log the wording used for you, the URL cited, and the rival preferred when you are absent. That log is the brief. It is not a ranking report, and it should not be presented as one.
Brand mentions that corroborate
A mention corroborates when it repeats a fact you already published, on a site a careful person would recognise. A trade piece that describes the offer, a conference listing with the right name and URL, a partner page that says what the partnership is: these help a retriever connect the brand to the work. They also help classic rankings when the mention is a real link to a real URL. Editorial link building is the right tool, aimed at pages that deserve the reference. Mentions on sites that exist only to sell mentions are a future clean-up.
Unlinked brand mentions still help recognition. They are not a substitute for links on competitive commercial terms. Keep them in a separate column of the same programme: publication, sentence, whether the facts match the sheet, whether a link is present, and which URL should have received it. Correct the errors. A muddled description left live becomes the version that gets quoted.
Do not invent a knowledge panel, a Wikipedia page you are not notable for, or a stack of awards. Paid edits and fake profiles collapse when someone checks. Point schema `sameAs` only at profiles you actually control.
Pages a model can quote without inventing you
The fact sheet gets you recognised. The citable page gets you used. Each priority URL should match the sheet, then answer the questions a summary fans out into: who it is for, what the work includes, what it refuses, how it is sequenced, and how it sits next to neighbouring services. That passage design is the same job as Google AI Overviews optimization. Entity work without a page worth lifting produces a correct name and an empty citation.
Stop improvising the boilerplate. The first hundred words of a service page are where drift starts: a new metaphor, a softened exclusion, a market you do not serve. Lock those words to the sheet. Let the rest carry proof, method and internal links. Content briefs that include the required facts, the forbidden claims and the target URL will beat thought-leadership that never describes the firm the same way twice.
Authors need the same treatment when the person is part of the trust. A bio that says what they have done, on pages they wrote, is an entity. A stock portrait and the word "expert" is not. If several people publish, keep their topics distinct so the firm is not an anonymous committee.
What we measure, and what we refuse
Run a fixed prompt panel each month, in the engines buyers actually open. For a single-market firm that might be a dozen questions in ChatGPT, the same dozen in Perplexity, and the Google queries already tracked for overviews. Record four fields: named or not, description accurate or not, URL cited or not, rival cited if you were skipped. Paste the answer. A memory of what a chatbot usually says is useless a fortnight later.
Read that panel beside ordinary rankings for the same offers. If you are unnamed and also invisible in search, the problem is retrieval, not a cleverer sentence. If you rank and the chat description is wrong, fix the crawled URLs, then earn mentions that repeat the correction. Accurate and uncited usually means the source page is vague, blocked, or aimed at another query — a classic SEO job, not a brand campaign.
We refuse doorway pages for every city you would like to be associated with, prompt tricks aimed at the public web, and any report that leads with a "share of model" figure you cannot audit. We refuse pages written only for the machine. International firms should not publish one global mush of facts. The entity can be shared and the market pages must differ, which is the point of international SEO.
Tie the entity back to the ranking URL
The commercial test is dull. Does the resolved brand point at a URL that can rank and convert? An about page that never links to the service, beside a service page that describes a different company, fails. So does a citation to the homepage when the query was about a specific offer. Internal links tell people and machines which URL owns which fact. Use anchors that match the offer. Keep them inside the market you are targeting. Stop sending every mention to the root.
Write the fact sheet, make the site agree with it, align the profiles you control, rebuild the priority passages, then earn mentions against those URLs. Put the prompt panel and the ranking report in the same monthly note. When the description stabilises and the right URL starts to appear, continue. When the description keeps mutating, stop publishing and repair the contradictions first.
SEO Jackers Global keeps this on one desk with the ranking work, not as a side experiment in prompts. If the brand is being summarised badly, or not at all, start with AI SEO and send the fact sheet with the order form.