AI assistants recommend businesses they can verify. That means consistent company facts across the web, pages that answer one question each with concrete numbers, and independent sources (reviews, directories, discussions) that repeat your story back. Classic SEO still matters, because most AI tools read search results, but it is no longer the whole job.
The shift is measurable. When Google shows an AI Overview, clicks to the top result drop by about a third, 34.5% in Ahrefs’ study of 300,000 keywords. A growing share of buyers never see your site: they see what an AI says about you. This guide covers what the AI engines cite, what to change on your site, and what to skip.
If the acronyms themselves are the confusing part, start with AEO, GEO and SEO explained and come back. This guide is the practical half.
What AI engines cite
Semrush’s AI visibility study looked at what ChatGPT and its rivals reference when they answer. Two content types dominate. Community sources come first: Wikipedia, Reddit, review platforms, places where people other than the brand say what the brand does. Separate research from 5W found Wikipedia and Reddit alone drive over a quarter of ChatGPT citations in the US.
Structured factual pages come second: clear specifications, transparent pricing, verifiable claims. The Semrush team’s phrasing is worth keeping: AI models cite content they can “confidently reference”. A page that says “affordable websites for ambitious brands” gives a model nothing to hold. A page that says “a 5–10 page business site costs £3,500 to £8,000 and takes three to five weeks” gives it a fact, and models repeat facts.
Polished marketing pages, the study notes, rarely make the citation lists at all.
Make your company verifiable
An AI deciding whether to recommend you runs a background check with the patience of a machine. It reads your site, then checks whether the rest of the internet agrees.
Start with entity consistency. Your business name, address, registration and services should match everywhere they appear: your site, Companies House, your Google profile, every directory. Conflicting facts read as risk, and models hedge on risk. That same consistency is the foundation of local SEO, so the work earns you map visibility while it earns AI trust. Mark the facts up with Organization structured data and point sameAs at your real external profiles so the connections are explicit rather than guessed.
Then give the machine somewhere independent to look. Reviews on platforms like Clutch or Trustpilot, a filled-in directory profile, a client site crediting your work: each one is corroboration you don’t control, which is why it counts. This is slow work with no shortcut, since fake reviews and padded profiles are exactly what the models are trained to discount.
Publish your proof where a crawler can read it. We keep company registration, ICO registration and insurance on a public trust page, so a model checking whether ICE WIND is real can settle the question in one request.
Finally, check your robots.txt. OpenAI’s search crawler, Perplexity’s and Google’s need access before any of this pays off. Blocking OAI-SearchBot while hoping ChatGPT recommends you is a mistake we’ve seen on live client sites.
Make your content citable
Structure your site so each page answers one question a customer asks. One intent per page beats one page listing ten services, which is why we split our web development, web design and other services into separate pages: a model matching “who builds e-commerce sites in London” needs a page about that, not a homepage about everything.
On each page, put the direct answer in the first paragraph and let the detail follow. Assistants lift their quotes from openings.
Publish numbers you can stand behind. Prices, timelines, project sizes. Vague content protects you from nothing and gets cited for nothing; our own most-referenced page is the one that states plainly what a website costs in the UK. Keep dates on articles and refresh stale facts, because models discount pages that look abandoned.
And keep passing classic SEO: crawlable HTML, working sitemap, sensible titles, internal links. Perplexity and ChatGPT lean on search indexes to find candidates before they cite anyone. GEO extends SEO; it doesn’t replace it.
Can a machine actually read your site?
All of the above assumes the crawler gets your words. Often it does not.
If your pages build themselves in the browser, which is what most React and Next.js sites do by default, then whatever arrives before the JavaScript runs is what some crawlers keep. Google renders JavaScript. Several of the AI crawlers do not, or do it inconsistently, and none of them publish a guarantee. You cannot see this from your own browser, because your browser runs the script. Fetch your page with JavaScript disabled and read what is left. If the answer is a loading spinner, that is your citation problem.
We build brochure sites as static HTML for this reason more than any other. The words are in the file. There is nothing to execute and nothing to get wrong.
Then there is llms.txt, a plain text file at your site root that lists your key pages and states in a sentence what the business does. Be clear about its status: it is a community proposal, not something any model vendor has committed to reading, and nobody can currently show you it moved a number. We publish one anyway. It takes twenty minutes, it costs nothing, and if adoption arrives you are already there.
What to skip
Skip AI-bait pages written for robots: keyword-stuffed “best agency in London” lists, thin location pages, FAQ farms. Models learn from human-quality signals, and content no human would read gets filtered the same way search engines learned to filter it years ago.
Skip inflated claims outright. Unverifiable awards, invented client counts and borrowed logos now carry a new cost: an AI that finds a claim contradicted elsewhere downgrades its trust in everything else you say.
Skip obsessing over any single tool’s answer. Responses vary by phrasing, day and model. The work above raises your odds in every tool at once, and it keeps paying off as new assistants launch on the same public data.
How to tell whether any of it worked
This is where most AEO advice stops, which is convenient for whoever is selling it. Three things are worth watching, and they answer different questions.
Whether you get named at all is the first. Ask the questions your buyers ask and count how often your business appears in the answer. That is your baseline, and it is usually zero for a while.
Where you land inside the answer is the second, and it matters more than people expect. Being the first company named and being mentioned in a closing aside are not the same result, even though both count as an appearance.
How the model describes you is the third and the one that costs money. A model that names you while hedging about whether you are established is doing you harm at the exact moment a buyer is deciding.
Worth separating two things that get muddled: being mentioned means the answer names you. Being cited means your page shaped the answer, with or without your name attached. Citations without mentions usually mean your content is good and your entity data is thin, which is a fixable problem and a different one from being invisible.
Doing it without buying anything
Write down ten to fifteen questions a customer would actually type. Not your keywords. The questions. Run them once a month in ChatGPT, Perplexity and Google AI Mode, in the same words each time, and paste the answers into a document with the date. That is the whole method.
Two rules make it useful. Keep the wording identical, because rephrasing changes the answer and you will fool yourself. And never judge from a single run: models vary day to day, so a change is a change only if it survives two months. Tools exist that automate this across models and track it daily; if you are running AEO for clients, one of them will pay for itself. For your own business, a document and a monthly reminder gets you most of the way.
If you’d rather have this done than read about it, this checklist, entity data, structured pages, trust signals, is part of how we build every site, and it is what our AI search optimisation work covers end to end. Send a short brief through start a project and we’ll show you what it looks like applied to your business.