Generative engine optimisation — GEO, sometimes AEO or “AI SEO” — is the practice of being visible in answers generated by AI tools rather than in a list of links. The term is young, the acronyms multiply, and a lot of what is written about it is either a definition with no method or a method that is really a sales pitch. It is worth being clear about what is genuinely new.

The honest summary: GEO is SEO done in a way that also survives being read by a machine that answers a question instead of ranking ten pages. The audience did not change. The way you are found, and the way you are judged, did.

What actually changed

Three things, and only three, are meaningfully different from ordinary search work.

  1. The unit of competition is the answer, not the result. A search engine returns a page. An assistant returns a sentence with a few names or facts in it. To be in that sentence you have to be a source the assistant will actually cite, not merely a page that ranks. In my baseline of 60 answers, no domain dominated: the most-cited source appeared six times, and small agency blogs were cited alongside Webflow and HubSpot. The bar is being useful in a form the model can lift, not being famous.
  2. Your content is read, not scanned. The model does not see your layout, your hero image or your navigation. It reads text, and it reads it literally. A page that answers the question in its first paragraph is worth more than a beautiful page that arrives at the answer in the fourth section. The words are the interface.
  3. Corroboration carries more weight. A model answering “who should I use for X” is looking for agreement between sources. If your site says you are excellent and a directory, a review site and a trade body say the same, that is corroboration. If only your site says it, that is an assertion. This is why being mentioned elsewhere matters at least as much as what your own pages say.

What did not change

Almost everything else. The foundations GEO rests on are the same ones that made a site work in Google: answering a specific question, in the words buyers use, on a page built for that question, with the facts stated plainly. If that sounds like the research-led approach, that is because it is.

  • Know what your market actually asks. Without it, you write the wrong pages faster.
  • One question, one page, with the answer up front. This is what gets quoted.
  • State your facts in structured data, so name, services, location and reviews are unambiguous.
  • Keep pages current and dated. A pricing page from 2022 is read as stale by a model too.
  • Make sure the content is server-rendered, because many AI crawlers do not run JavaScript.

None of those is a GEO trick. They are the boring fundamentals, and they happen to be exactly what an assistant needs in order to quote you.

Where GEO differs in practice

The differences show up in emphasis, not in kind. Where classic SEO might obsess over a title tag and an internal link, GEO leans harder on three things:

Writing to be quoted

A model lifts a sentence. So a page that contains a clean, self-contained answer — “the average cost is X because Y” — is more quotable than one that makes the reader work for it. Plain declarative sentences, concrete numbers, and a visible last-reviewed date all help.

Being present in the sources that get cited

Assistants draw on directories, comparisons, review sites, forums and lists. A business that is absent from all of them is harder to corroborate, however good its own site is. This is the same work as reputation and listings, viewed through a new lens.

Measuring citations rather than positions

There is no rank tracker for an AI answer, but there is a count. Ask the same fifteen questions across four assistants and record every mention and citation of your brand.1 That count is your before-and-after. It is crude, and it is far more honest than most reporting.

The acronym trap

GEO, AEO, AI SEO, LLMO — the vocabulary is unstable because the products are. The risk for a small business is paying a premium for a new label on the same work, or being sold “GEO” as a separate programme that quietly duplicates the SEO you already need. If a provider cannot explain what they will change on your site and how they will measure it, the acronym is doing the work the substance should.

The test is simple. Ask which questions you want to be in the answer to, which pages currently answer them, which sources the assistants cite for them, and how the provider will show the count moving. If those four answers exist, you have a method. If not, you have a word.

Where to start

For a small business the sensible first move is not to chase every surface at once. It is to pick the handful of questions that actually lead to work, check who currently owns them, and make sure you have a page that answers each one and a profile that corroborates it. That is research first, then build, then the ongoing work of staying cited.

The main guide in this series, how AI assistants choose which businesses to recommend, sets out the mechanism in full. If you would rather have the starting point mapped, GroundWork produces the Demand Map that says which questions are worth answering, and Coverage is the ongoing part that keeps the count moving.

  1. KOODOS AI-visibility baseline, 17 September 2026: 15 questions × Perplexity, ChatGPT, Gemini and Google AI Mode, 60 completed calls. Raw data in docs/research/ai-visibility-2026-09/. ↩