When someone asks ChatGPT, Gemini or Perplexity to recommend a firm like yours, your brand is either named in the answer or it is invisible. Getting named comes down to three things: writing a machine can quote cleanly, your brand being talked about on sites these tools already trust, and the basic technical plumbing that lets them read you. Here is how a UK business actually does it.
A few weeks ago I asked Perplexity to recommend a PR agency in Leeds. It gave me three names, confidently, in a neat little paragraph. One of them closed down last year.
That is the part nobody warns you about with AI search. It sounds certain even when it is wrong, and it reaches its verdict without ever showing you a list of blue links to argue with. The customer just gets an answer, and your brand is either in it or it is not.
If your search strategy still ends at ranking on the first page of Google, you are polishing a shop window on a street people are starting to avoid. The machines now make the introductions, and they decide whether tools like ChatGPT, Gemini and Perplexity put you forward or hand the customer to someone else.
Is this just SEO with a new name?
No, though it sits on top of everything you already do for Google. The shorthand doing the rounds is generative engine optimisation, or GEO, and if you want the full breakdown with our own traffic and leads data, we put that in our GEO masterclass. For this guide I want to stay practical: how does a UK business actually get named.
The useful distinction is this. Traditional SEO is about being the best result to click. AI search is about being the best source to quote. It means a potential client can form a firm opinion of your business having never once visited your website. They asked a machine, the machine answered, and the machine decided who was worth naming.
I had a client last year who started getting enquiries that referenced things their own website did not even say. We worked out an AI had stitched the picture together from an old trade interview and a podcast appearance. The good news is it was flattering. The unsettling bit is they had no idea it was happening, and no hand in what got said.
How AI engines decide who to recommend
An AI engine does not crawl one page and rank it the way Google’s old system did. It builds an answer out of bits and pieces from across the web, and it leans hardest on sources it can read cleanly, verify, and trust.
In practice that rewards a few things. Clear, factual writing it can lift without guessing. Structure that lets it grab the right sentence. Consistency, so you are not saying one thing on your site and something different in an interview. And mentions of your brand on other sites it already rates.
There is also a brutal recency bias. The LLMrefs 2026 GEO guide notes that once a page passes roughly three months old, the rate at which AI tools cite it drops away sharply. So the brilliant explainer you wrote two years ago and never touched again is doing less for you than you think. Freshness is no longer a nice-to-have. It is part of how you get picked. I got access to a LLM tracker yesterday and 80% of everything they were citing was PR earned media and it was from the last 12 months. That is a fact.
Google’s own May 2026 guidance backs the boring half of this up. Its guide to optimising for generative AI features tells site owners to write genuinely useful, non-commodity content with a clear point of view, keep a clean technical structure, and reduce duplicate content. No magic, just the fundamentals done properly.
The three jobs, minus the jargon
Strip the acronyms away, and getting recommended comes down to three jobs.
Be readable to a machine. Write in plain, factual sentences, answer the question in the first line, and structure the page so an engine can pull a clean quote. The waffly marketing intro that says nothing for three paragraphs is worse than useless here, because the machine gives up before it reaches your point.
Be talked about across the web. AI tools weight brand mentions on third-party sites, and the research suggests they count them even when there is no link attached. A casual reference to you in a trade title or a podcast can lift how visible you are to an AI. This is the bit most SEO agencies are weakest at, and it happens to be the thing PR has always done.
Be technically legible. Schema markup, a clear author byline, consistent business details. The plumbing. Worth saying, Google has now confirmed you do not need an llms.txt file or any special AI markup, so do not let anyone sell you that. The basics will not save weak content, but they help a machine trust and place the good stuff.
Where PR fits, and why earned mentions beat backlinks
Here is the part the SEO crowd tends to skip. For fifteen years, getting found online meant chasing good old backlinks. You wanted other sites to link to yours, and a whole industry grew up to game that, which was brilliant.
AI search rewards something older and harder to fake: being talked about by credible people. The trade interview. The quote in PRWeek or the byline in Marketing Week. The hour you spent on someone’s podcast. An engine reading the web sees your name come up, in context, on sites it trusts, and that shapes whether it recommends you.
That is earned media. It is what a decent PR team has produced all along. The difference now is that the payoff has widened. A piece of coverage used to reach the people who read that title. Today it also teaches the machines who you are.
Innocent: the same voice everywhere it appears
Take Innocent Drinks. What makes them clever is not the smoothies, it is that they sound identical everywhere. Same dry, daft tone on the bottle, on social, in every press interview they have ever given.
What they did was refuse to switch register for the corporate channels. The wonky jokes that work on Twitter are the same ones in their national press coverage and their trade write-ups.
The PR result is years of national pickup, from the broadsheets to the Mail, plus a steady run of marketing trade coverage that uses them as the textbook example of brand voice. The earned attention compounds.
The lesson is the useful bit. When a brand says the same clear thing about itself everywhere, an AI can form a confident, repeatable picture and hand it to a user without hedging. Inconsistency is what makes a machine vague about you. I cannot show you Innocent’s citation numbers inside ChatGPT, and I would not trust anyone who claimed they could, but the principle holds: clarity and consistency are what get you quoted.
How to tell whether any of this is working
You measure it by asking. There is a clumsy term doing the rounds, share of answer, and underneath the jargon it is simple. Write down the ten questions your customers actually type when they are looking for a firm like yours. Ask each one to ChatGPT, Gemini and Perplexity. Note who gets named.
Do it again in a month. If your name starts appearing where it did not before, something is working. If a competitor owns every answer, you know where the gap is. There are tools arriving that promise to automate this, and some are decent, but the manual version costs you an afternoon and tells you most of what you need.
It is rougher than a Google ranking report, and the picture wobbles because the engines change their minds. Treat it as a direction of travel, not a precise score.
A 90-day way in
You do not need to do everything at once. You can do it in three months over three careful moves.
First month, find out where you stand. Run the share-of-answer check above. Then fix your most important pages so they answer the question in the opening line, and add basic schema and a clear author byline. This is the unglamorous groundwork, and it is the cheapest win available.
Second month, feed the machine something fresh. Publish or properly update two or three strong pages, because of that recency bias. Then go and earn two pieces of third-party coverage, a trade comment, a podcast slot, a guest piece. Real PR, pointed at the topics you want to be known for.
Third month, measure and repeat. Re-run the check, see what moved, and keep going. This is not a launch, it is a habit. The brands that win at it are the ones still doing it in month nine.
Can a smaller UK business really compete?
Yes, and being British is quietly an advantage. AI tools treat locally written, UK-specific content as more trustworthy for UK queries. A focused Leeds or Manchester firm that writes clearly for a UK audience can get named ahead of a bigger, blander national player that publishes generic copy for everyone and no one.
The trick is to be specific. Name your city, your sector, the kind of client you serve. When someone in Yorkshire asks an AI for a recommendation, a page that plainly says who you help and where beats a vague national homepage every time. Small and clear travels further here than big and bland.
What next
If you want to know whether AI search engines are recommending you, ignoring you, or quietly quoting your competitors, that is exactly what we look at in an AEO audit. Drop me a line and I will send over what is involved.
And if you want the deeper version, with our own HubSpot data on what AI search did to our traffic and leads, read the GEO masterclass.
Frequently asked questions
Is getting recommended by AI the same as GEO?
Roughly, yes. Generative engine optimisation, or GEO, is the umbrella term for getting your brand into AI answers. This guide is the practical, UK-focused version. For the full breakdown and our own data, see our GEO masterclass.
How do AI search engines find my brand?
They read your site, but they also learn about you from everywhere else: trade press, podcasts, directories, social, other people’s articles. Consistent, credible mentions on sites they trust are a big part of how they decide to recommend you.
How long does it take to get recommended by AI?
Longer than you would like and shorter than traditional SEO used to. Because the engines favour fresh content and update often, changes can show up in weeks rather than months. The catch is you have to keep feeding it, or your visibility fades as your pages age.
Do I need technical schema, or is good content enough?
Content does most of the heavy lifting. Schema and the technical plumbing help a machine trust and place that content, so they are worth doing, but they will not rescue a page that says nothing. Get the writing right first. And ignore anyone pushing llms.txt files, Google has confirmed you do not need them.
About The Author
