What is Generative Engine Optimisation (GEO)?
Generative Engine Optimisation is the work of being present, correct and attributable inside answers that a model composes rather than retrieves. The measurable outcome is not a ranking. It is whether the assistant describes you, and describes you accurately.
An answer engine retrieves passages. A generative engine composes. Ask an assistant which firms do a particular kind of work in the Benelux and it will not hand you a ranked list it looked up. It will write a short account from whatever it has read, and that account will include four or five names.
Being one of those names is a different problem from ranking. There is no position to hold and no link to earn. The question is whether the model, drawing on everything it has absorbed, reaches for you — and whether what it then says about you is true.
Where does a model's picture of your company come from?
Two places, and they behave completely differently.
Slow, broad, and impossible to edit. This is where a model's general sense of who you are lives. You influence it over years by being written about consistently, and by being described the same way everywhere you appear.
Fast, narrow, and very much editable. Most assistants now run a live search before answering anything current. This is where a page published last week can end up in an answer given today.
Almost all practical GEO work happens in the second place. You cannot retrain a model. You can absolutely make sure that when it goes looking, the clearest, most specific, most recently confirmed page on the subject belongs to you.
You cannot edit a model's memory. You can decide what it finds when it looks.
What makes a model reach for one source over another?
No provider publishes the weighting, and anyone who gives you a formula has made it up. What is observable, by reading a few hundred assistant answers and looking at what they picked, is a consistent shape:
- Specificity beats polish. A page that names conditions, figures and exceptions gets used over a page that is better written and vaguer.
- First-hand beats aggregated. A company explaining its own method is preferred to a listicle explaining twelve companies' methods.
- Recency matters where the subject moves. For pricing, regulation and anything dated, a page with a visible and recent revision date wins against an undated one.
- Agreement matters. Where three independent sources say the same thing, the model states it confidently and cites one of them. A lone contrary source is usually dropped rather than argued with.
- Reachability is binary. A page behind a consent wall, a login or an unrendered client-side app is, for this purpose, unpublished.
Being described correctly is half the work
The failure mode that surprises companies is not absence. It is a confident, fluent, slightly wrong description: a service you stopped offering, a market you left, a founder who moved on, a price from three years ago. The model is not being careless. It is repeating the most consistent account of you it could find, and that account is your old website, a stale directory listing and a conference bio nobody has updated.
The remedy is unglamorous. Publish a current, plain, first-person account of what you do, where, for whom, and at what sort of price, on your own domain, and keep the same facts consistent in every other place a crawler can reach — your profiles, your listings, your entries in industry registers.
How is this different from AEO?
| AEO | GEO | |
|---|---|---|
| The engine is doing | Retrieving a passage | Composing an account |
| You want | Your passage to be the one used | Your company to be named and described correctly |
| Main lever | How a page is written and structured | Consistency and specificity across everything about you |
| Feedback arrives | As a citation under an answer | As a mention inside the text, often unlinked |
| Timescale | Weeks | Months |
In practice nobody does one without the other, because the same pages serve both. A well-made answer hub is simultaneously a set of retrievable passages and a consistent, specific, first-hand account of a company. That is why the two acronyms keep appearing together.
Questions this settles
Not in the organic answer itself. Providers have experimented with advertising formats alongside answers, but the sources a model reaches for while composing are not purchasable, and any service offering that is describing something else.
Ask it, in each of the languages your market uses, with the phrasings a buyer would use rather than your brand name. Record the answers and repeat monthly. The change over time is the useful part.
Yes, because most of the lever is the live retrieval step rather than the model's memory. A small company with twenty sharp, specific, current pages on one subject is regularly quoted over a large one with none.