What Is GEO? Generative Engine Optimisation Explained for UK Businesses

What Is GEO, and What Does It Cost?

Generative Engine Optimisation (GEO) is the practice of structuring your website and business information so that AI assistants like ChatGPT, Perplexity, Claude and Google AI Overviews can find, trust and cite your business when someone asks a relevant question. What a GEO engagement costs depends on how many sites or locations are in scope, how deep the technical and schema gaps run, and whether it is a one-off audit or an ongoing programme — and an ongoing programme is worth running over a three-month minimum, because AI citation behaviour takes time to shift and measure.

GEO sits alongside SEO rather than replacing it — most of the technical groundwork (fast, crawlable pages, clear structured data, genuinely useful content) helps both. What is different is the target: you are no longer optimising purely for a ranking position, you are optimising to be the fact an AI model reaches for when it answers a question about your service category.

GEO vs SEO

Neither replaces the other. A page that ranks poorly in Google rarely gets cited by an AI assistant either, since most assistants' real-time retrieval still leans on a conventional search index behind the scenes.

How AI Assistants Actually Retrieve and Cite — What We Know and What We Are Inferring

This is the part of GEO where honesty matters most, because a lot of what gets written about it is guesswork dressed up as certainty. Here is what is actually documented by the vendors, versus what is inferred from testing and third-party analysis.

Documented: ChatGPT only cites and links to sources when it actively searches the web (via ChatGPT Search or a browsing-enabled response); a plain answer from its training data carries no citation. Perplexity is built around live retrieval by default — it searches, ranks and reads a set of pages, then writes an answer with inline numbered citations, and it always cites something when it answers from the web. Claude's base model answers from training data with a fixed cutoff unless a user turns on web search, in which case it retrieves and can cite live pages, similarly to the others.

Inferred from third-party testing, not officially confirmed: independent audits reported by outlets tracking AI search behaviour have found that these systems retrieve considerably more pages than they ultimately cite — one widely cited analysis of Perplexity's Sonar system found it visits around ten relevant pages per query but names only three or four in the final answer, and similar gaps have been reported for ChatGPT. In other words, being crawled and even being read is not the same as being cited — the model is also deciding, in a way none of the vendors document publicly, which of the pages it read are worth naming.

The most solid piece of independent research in this space is the Generative Engine Optimization paper from researchers at Princeton, Georgia Tech, IIT Delhi and the Allen Institute for AI, presented at the ACM KDD conference in 2024. It tested nine content-level optimisation strategies across roughly 10,000 real queries and found that adding clear statistics, quotations and citable facts to a page improved its visibility in generative answers by 22–41% depending on the domain. That is the closest thing GEO has to a controlled study, and it is the basis for most of the practical advice below.

The Practical Work Involved

Answer-first content. Pages that state a direct answer, a price or a fact in the first few sentences give an assistant something short and quotable to lift, rather than making it infer an answer from paragraphs of scene-setting.

Structured data and schema markup. JSON-LD schema — Organization, LocalBusiness, Service, FAQPage and similar types from the schema.org vocabulary — gives a machine-readable version of who you are, what you do, where you are and what you charge, checkable with Google's own Rich Results Test. It is not confirmed to be a direct AI citation signal on its own, but it removes ambiguity that would otherwise have to be guessed at from prose.

Consistent business facts. Your business name, address, phone number, services and pricing should read identically everywhere they appear — your website, your Google Business Profile, directories, and any press mentions. Assistants that cross-reference sources are more likely to trust a fact that appears the same way in several places than one that only appears once.

The llms.txt convention. llms.txt is a proposed plain-text file, placed at yourdomain.com/llms.txt, that gives AI systems a structured summary of your site — key pages, services and facts in one place. It is worth being precise about its status: no major AI provider, including OpenAI, Google or Anthropic, has publicly confirmed that it reads third-party llms.txt files as a ranking or citation signal, and Google's own search team has said it does not use it. OpenAI and Anthropic maintain llms.txt files for their own documentation sites, which shows the convention has some internal traction, but that is different from confirming they consume other companies' files. We add it because it costs almost nothing and causes no harm, not because it is a proven lever.

Citations on sources assistants already trust. Because these systems weight authority, a mention on a source they already crawl and trust — UK trade press, recognised industry directories, your own well-structured site — carries more weight than a mention nowhere else. This is closer to traditional digital PR than to classic link-building.

Technical crawlability. None of the above matters if the relevant crawlers cannot reach your pages in the first place.

Which Crawlers Matter for Citation vs Training

Different AI companies operate separate crawlers for training their models versus fetching content in real time to answer a query, and you can allow one while blocking the other in robots.txt.

Many UK business websites block all bots matching "GPT" or "AI" out of caution, without realising they have also blocked the search crawlers that could cite them. Checking your robots.txt against this distinction is one of the fastest wins in a GEO audit.

Measuring GEO and a Realistic Timeline

Rank trackers do not work for GEO, because there is no fixed position to track. The practical method is prompt testing: build a list of 15–30 realistic questions a buyer would ask an assistant — "who builds AI chatbots for law firms in Manchester", "best web development agency in the UK for a small business" — and run them regularly across ChatGPT, Perplexity and Google AI Overviews, recording whether your business is named, how, and alongside which competitors. Repeated monthly, this becomes your actual measurement, alongside directional signals like referral traffic tagged from AI assistant sources in your analytics.

On timeline: technical fixes — schema, robots.txt, crawlability — can take effect within a couple of weeks once the crawlers next visit. Content and citation-building work realistically takes two to four months to show up consistently across prompt tests, because it depends on the assistants' own crawl and retraining cycles as much as on your changes.

What Decides Which GEO Tier You Need

Book a free GEO audit call at /contact for a fixed quote once we know your site count and current schema coverage.

Who Should Bother With GEO, and Who Should Not

GEO is worth it for a UK business whose buyers plausibly type a category question into an assistant before they search Google directly — professional services, agencies, anything with a "best X for Y" comparison shape to the buying decision. It is a poor investment for a business that lives entirely on repeat, name-based custom, or one that has not yet fixed the basics of a fast, well-structured, genuinely useful website — GEO is not a shortcut past a weak site, it is a layer on top of a good one. Our own web development work at /services/web-app-development builds that foundation in first; see also our post on getting recommended by ChatGPT at /blog/how-to-get-your-business-recommended-by-chatgpt-uk, and how we build AI-ready sites for Manchester businesses at /blog/web-development-manchester-ai-ready-websites-2025. For e-commerce specifically, see /industries/ecommerce.

Read more about the full service at /ai-seo, or book a free GEO audit call at /contact to see where your business currently stands in the assistants your buyers are already using.