What generative engine optimisation actually means

Generative engine optimisation is the work of becoming the source an AI assistant's answer is built from, rather than only a result on a results page. That is the whole definition. Everything else people attach to the term is either a restatement of that or a claim worth being skeptical of.

It is worth being precise here because the term gets used loosely. Some vendors use GEO to mean "SEO, but we added AI to the pitch deck." Others use it to mean something closer to prompt injection: stuffing pages with text meant to manipulate a model rather than inform a reader. Neither is what the discipline actually is.

What changed

Search used to mean a person typed a query, saw ten blue links, and picked one. Increasingly, a person asks an assistant a question and acts on the answer without ever seeing a results page. Google's AI Overviews and AI Mode, ChatGPT's web search, Perplexity, and Copilot are all doing the same underlying thing: fetching pages, extracting an answer, and citing a small number of sources.

That last part is the entire opportunity and the entire risk. A page that is easy to extract an answer from gets cited. A page that is not, does not exist as far as that surface is concerned, no matter how good the content actually is.

The three things that actually matter

**AI crawler accessibility.** If GPTBot, ClaudeBot, PerplexityBot, or Google-Extended cannot fetch your page, or if your content only exists after client-side JavaScript runs, none of the rest of this matters. This is the precondition, not an optimisation. We have written separately about which crawlers do which job, because "block all AI bots" and "allow all AI bots" are both decisions people make without reading what they are actually choosing.

**Passage-level citability.** An assistant does not usually cite a whole page. It lifts a passage, a paragraph or two, self-contained enough to stand alone as an answer. A page written as one long argument that only makes sense read start to finish is hard to extract from. A page with clear, direct answers to specific questions, each one complete in a few sentences, is easy to extract from. This is most of why the FAQ blocks on this site's own service pages exist: not to pad word count, to give something quotable.

**Entity and brand consistency.** A model forms a picture of who you are from every source that mentions you, not just your own site. If your business name, description, and claims are inconsistent across your site, your Google Business Profile, and wherever else you are mentioned, you are making that picture blurrier, which makes a model less confident citing you at all.

What does not matter as much as it sounds like it should

An `llms.txt` file is a genuinely useful, informal convention for pointing an assistant at your best pages in one place. It is not a ranking signal, it is not read by Google Search, and a site with a beautifully written llms.txt and thin, unreadable pages behind it has not done the work, it has written a table of contents for a book with no chapters.

The same goes for schema markup on its own. Structured data helps a system understand a page it can already read. It does not compensate for content that gives an assistant nothing to extract.

How to tell whether a site is actually doing this

Fetch your own page with JavaScript disabled, or from a terminal, and read what comes back. If the answer to your page's core question is not in that raw response, an AI crawler does not see it either. The free site check on this site runs this exact test, along with the crawler-access and structured-data checks, against any public URL.

Then read your own content the way an assistant would: pick a specific question your page answers, and see if you can quote a self-contained two-or-three-sentence answer to it. If you cannot, neither can the system trying to cite you.

None of this replaces classic SEO. It is the same fundamentals, applied with a different extraction mechanism in mind. Our AI search visibility work is this checklist, run properly, on a real site, not a new discipline invented to justify a new invoice line.