What is generative engine optimization (GEO)?
Making your content likely to be retrieved, understood and represented accurately by generative AI. It starts with permission, not with writing — and most sites fail at the first step without knowing.
By Sapun Lamichhane · Arcetis
A growing share of people never see a list of links. They ask an assistant, read a paragraph, and act. Whether that paragraph mentions your business, and whether what it says is correct, is decided by a set of factors that no traditional SEO audit examines.
Generative engine optimisation is the name for managing those factors. The term is new and somewhat contested, but the underlying problem is concrete and testable, and it divides cleanly into three layers.
Layer one: permission
Nothing else matters until this is right, and it is where most sites fail without ever discovering it.
Every AI operator crawls under its own user-agent token, and each is governed independently by robots.txt. A file that cheerfully allows Googlebot says precisely nothing about GPTBot, ClaudeBot or Google-Extended. Worse, the blanket AI blocks that many sites added around 2023 are still in force, still excluding them from products those rules predate, and produce no symptom at all — rankings are unaffected, because rankings are Googlebot's department.
There is a second trap in the mechanics. robots.txt groups are not additive: a crawler obeys the single most specific group matching its name and ignores every other group in the file, including the wildcard. So adding a permissive group for one AI crawler does not grant it the site-wide rules plus permission — it releases that crawler from the site-wide rules entirely. Sites regularly expose paths this way while believing they have done the opposite.
Layer two: retrievability
Permitted is not the same as readable. The single largest technical determinant of GEO is whether your content exists in the HTML your server returns.
Googlebot runs a full rendering pipeline and will execute your JavaScript. AI crawlers, as a class, should be assumed not to. A single-page application that ranks perfectly well can be an empty shell to every AI crawler that fetches it — and again, nothing anywhere reports this. Your analytics look normal. Your rankings hold. You are simply never cited.
- Server-render or statically generate anything you want represented.
- Keep the important content out of tabs, accordions and modals that require interaction to populate.
- Return a fast, complete response. Crawlers give up, and a slow page is a page that sometimes does not get read.
- Do not gate the substance behind a cookie wall or interstitial the crawler cannot dismiss.
Layer three: representation
Once a model can fetch and read you, the remaining question is whether it can use you accurately — and, ideally, attribute it.
This is where GEO overlaps most with answer engine optimisation. A model assembling a response needs to identify a claim, bound it, and decide it stands on its own. Prose that builds an argument across five paragraphs is hard to use. A heading that states a question, followed immediately by a complete answer, is easy.
- State facts explicitly rather than implying them. A model cannot infer your opening hours from a photograph of a sign.
- Add structured data that matches the page — Organization, Product, Article, LocalBusiness — so the entity is machine-readable rather than inferred from prose.
- Keep the same facts consistent everywhere they appear. Contradicting yourself across pages is how a model ends up confidently stating the wrong price.
- Publish a named author and a visible date, which is what makes attribution possible at all.
How to measure it
GEO resists the dashboard-shaped measurement SEO has trained everyone to expect. There is no rank to track and no impression count. What you can do is test the inputs, all of which are deterministic:
- 1Resolve your robots.txt per AI user-agent token and confirm each result is what you intended rather than what you assumed.
- 2Fetch your own page without JavaScript and read what comes back. If your content is not in it, no AI crawler is seeing it either.
- 3Check that your structured data parses and describes the page accurately.
- 4Ask the assistants directly about your business, periodically, and record what they say. It is unscientific and it is also the only true end-to-end test available.
Our AI crawler checker covers the first, and reports what each token is actually permitted to fetch — which is frequently not what reading the file suggests.