What Is GEO? Generative Engine Optimisation

Generative engine optimisation is the work of being cited inside AI-generated answers. What it is, where the 40 per cent figure comes from, and why being cited is not the same as being recommended.

TopicGEO & AI Visibility

Generative engine optimisation, GEO, is the work of getting your business named in the answers that AI engines generate, rather than ranked in a list of links beneath them. It matters because many buying questions now get answered without anyone visiting a website, and a buyer who gets a satisfactory answer naming three suppliers rarely scrolls to find a fourth.

What GEO actually is

Search used to return ten links and let you choose. An answer engine reads the sources itself and writes one response, naming a handful of them. The competition moved from a ranked list to a paragraph, and a paragraph has room for three names rather than ten.

GEO is the practice of being one of those names. It overlaps heavily with search work, because the engines still find pages the way crawlers always have, and Google says its AI features draw on the same quality systems as the rest of search. Two things change. The unit is a passage rather than a page, because the engine lifts a sentence or a table. The test is whether your page is quotable, not just whether it ranks.

Each engine uses its own crawlers, and they are publicly documented. OpenAI publishes its bots and their user agents, Perplexity publishes its own, and blocking one does not block the others. Visibility does not carry between engines either, so being named by ChatGPT is not evidence of being named by Perplexity. You can also rank first on Google and appear nowhere in an AI answer about your own category.

Where the 40 per cent figure comes from, and what it leaves out

Nearly every GEO article quotes a 40 per cent visibility gain. It comes from a paper titled GEO, Generative Engine Optimization, by Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande, accepted to KDD 2024. The abstract does say GEO can boost visibility by up to 40 per cent, and the authors built a benchmark of user queries across several domains to measure it.

The part that gets dropped is in the same abstract. The authors report that the effectiveness of these strategies varies across domains, and say plainly that this points to a need for domain-specific optimisation. So 40 per cent is a ceiling observed under particular conditions, not a number you should expect. Anyone quoting it at you as a forecast has either not read past the headline or is hoping you have not. We summarise the study itself on the page about what the GEO study actually found. If you are starting from zero, the beginner's guide covers the first steps, and you will see the same discipline called LLM SEO in industry writing.

How strong is your own AI marketing? Take the free two minute scorecard and get an honest read plus the next steps that matter most.

Take the scorecard

Guides treat these as one outcome. They are two different things with different causes, and work aimed at the wrong one is wasted.

A citation is the engine using you as evidence. It happens when your page contains a fact, figure or definition that answers part of the question, and it is largely a content engineering problem. Put the answer near the top and show where the facts came from.

A recommendation is the engine naming you as the supplier to use. That is a corroboration problem, not a content problem. An engine will quote your definition of a term and then recommend a competitor to do the work, because the definition only needs to be correct while the recommendation needs your business to be verifiable. If you are cited constantly and never recommended, more content will not fix it. The answer is in how machines verify expertise.

OutcomeWhat it isWhat earns it
CitationBeing used as evidence for part of an answer.Quotable, sourced content on your own pages.
RecommendationBeing put forward as the supplier to hire.Proof of the business from outside your own pages.

How an engine decides what to quote

Three things decide it, in rough order of how much they matter.

The first is whether the answer is findable in the page. An answer spread over several paragraphs sits in no single chunk, so no chunk wins. Content that leads with the answer gets quoted over content that buries it.

The second is whether the claim can be checked. A sourced statistic survives. An unsourced superlative gets skipped. An engine is deciding what it is willing to repeat under its own name, not judging your prose.

The third is whether other sources agree. This part is slow, and most people skip it, because it depends on other people publishing rather than on you writing.

GEO when your business has no product category

Every guide on this subject assumes you sell a product in a category that people write comparison articles about. Optimise for "best CRM", get into the listicles, win the citations. That advice is useless if you are a structural engineer, a haulier or an employment solicitor, because there is no listicle of the best ones and no reviewer publishing a category roundup.

The queries that matter for a service business look different. They describe a problem and a constraint, and often a place. Somebody asks who can handle a particular job, under certain conditions, in their city, and the engine has to assemble an answer from whatever it can verify about local suppliers.

That changes the work. Answer the constrained question rather than the category, because the category page competes with everyone and the constraint page competes with almost nobody. Make the business verifiable, since with no reviews to aggregate the engine falls back on registrations, accreditations and named people.

Local and geographic questions

No guide on the first page of results covers this, and for a service business it is most of the demand. The questions people put to an assistant nearly always mention a place, because they are trying to hire somebody who can turn up.

Geographic answers behave differently from category answers in two ways. They are thinner, since a category question has hundreds of plausible sources and a question about one trade in one city has a handful. That makes them cheaper to win than the broad category terms everyone is fighting over. And they turn on location evidence rather than reputation, because the engine has to satisfy itself that you actually work in the place named, which is a question about addresses, registrations and stated service areas.

A page that names the place, the constraint and the proof gives an engine more to check than a page that only names the service, even if the second page ranks better. Our coverage pages are built that way. Each one shows the real business count for its area, so the claim about the market is checkable.

What GEO is not

It is not a replacement for search work. The engines find pages by crawling, so a site that cannot be crawled or rendered cannot be cited. Our guide on why AI agents ignore JavaScript sites covers where that goes wrong.

It is not a file you add. An llms.txt file is worth having, but it does not by itself make anything cite you. Our llms.txt guide goes into that properly.

It is not hidden instructions to models. Text planted for a crawler to obey is filtered, and it is a genuine reputational risk if anybody finds it.

It is not a rank you can check. There is no position one in a generated answer, and the same question asked twice can produce different sources. Measurement means sampling a fixed prompt set repeatedly rather than reading a number off a dashboard.

Where to start

Take the questions your buyers actually ask, put them to the engines they use, and write down the answers. That is your baseline, and it usually takes an afternoon. Then fix the two things that block everything else. One is whether a crawler can read your pages at all. The other is whether an engine can confirm your business exists and is what it claims.

After that it is content engineering and corroboration, in that order. The complete GEO guide sets out the method, the ninety day sequence and what it costs.

Common questions about GEO

Is GEO different from SEO?

It is an additional layer rather than a replacement. The engines discover pages by crawling, so the technical and content foundations are shared. What differs is that an engine retrieves passages, not whole pages, there is no ranked position to hold, and visibility has to be measured per engine because it does not transfer between them.

How long does GEO take to work?

There are two clocks. Technical and content changes are read on the next crawl, so retrieval-based engines can reflect them within days. Corroboration from third parties is slower because it depends on other people publishing, and runs to months. Any single timeline quoted to you has averaged two things that do not move at the same speed.

Can I just add an llms.txt file?

No, though it is worth having. It tells engines where your canonical content is, but it does not make them cite it. Treat it as plumbing rather than as a tactic.

Does GEO work for local and service businesses?

Yes, and the queries are usually easier to win than product categories, because a constrained local question has far fewer credible answers than a category question. With no reviews or listicles to draw on, an engine leans on whether your business is verifiable, so registrations and named people carry more weight than copy does.

Which engines should I optimise for?

The ones your buyers use, which you should check rather than assume. In the UK that usually means Google AI Overviews, ChatGPT, Perplexity and Copilot, with Bing mattering more than its search share suggests because Copilot draws on it. In the US, add Gemini and Google AI Mode.

Is the 40 per cent visibility gain real?

It is a real published finding and it is a ceiling, not an expectation. The same abstract reports that effectiveness varies by domain and calls for domain-specific optimisation, so treat the number as evidence the work can pay off rather than as a forecast of what yours will return.

Find out what the engines say about you

We run your buyers' questions across four engines and show you the answers before we touch anything. Book a call or run the free scorecard.

Put this to work on your business

We build exactly this, SEO, AI visibility, PPC, email and automation, as one measured system. Talk it through with a real person, no obligation.

Book a free strategy call or take the two minute scorecard