What is generative engine optimisation, and is it different from SEO?
GEO explained: how AI answer engines choose what to quote, how it differs from traditional SEO, and what actually changes on your website.
Generative engine optimisation (GEO) is the practice of structuring content so that AI answer engines — ChatGPT, Google’s AI Overviews, Perplexity, Copilot — can extract it, quote it and attribute it. It is a response to a change in how people find things: instead of scanning ten links, they increasingly read one synthesised answer assembled from a handful of sources. Traditional SEO competes for a position in that list of links. GEO competes to be one of the sources the answer is built from. The two share technical foundations — a page that cannot be crawled cannot be cited either — but they reward different writing, and that difference is where most of the practical work lies.
How answer engines decide what to quote
The useful mental model is extraction. A model is assembling an answer and needs a passage that stands on its own. Passages that survive that lift share features:
- They define the subject in the first sentence, naming it explicitly rather than referring back to a heading.
- They are self-contained, making sense without the paragraph before them.
- They contain a specific: a number, a named standard, a date, a price.
- They answer directly, without a paragraph of preamble.
Content written as a flowing narrative often ranks perfectly well and gets cited rarely, because no individual paragraph can be lifted out intact.
What this changes on a website
Less than people expect, and it is mostly writing rather than design.
Openings become definitions. Every page starts with a passage that defines its subject in a form that would make sense quoted alone.
Headings become real questions. Not “Our Approach” but “How do you measure whether it worked?” — with the answer in the sentence directly beneath. This is where most sites fail: their question-shaped headings are calls to action (“Ready to get started?”) rather than questions anyone would actually ask.
Vague claims get replaced with figures. “Fast, affordable websites” is not citable. “From £2,400, delivered in 10 to 15 working days, WCAG 2.2 AA performance” is.
Structured data gets added, so machines can read the organisation, its services and its prices without inferring them from prose.
What GEO cannot do
It cannot guarantee citation. Answer engines are not queryable indexes; their outputs vary between runs, between users and between model versions. Anyone selling guaranteed AI visibility is describing something they do not control. Measurement is correspondingly coarser than rank tracking: you test the questions your customers would actually ask, across several engines, on a dated schedule, and record whether you appear.
Is it worth doing yet?
That depends on whether your customers are already asking assistants for recommendations in your category. If they are, the work pays now. If they are not, the same structural changes still make pages clearer for human readers and stronger for conventional search — which is why it is rarely wasted effort, and also why it should not be sold to you as an emergency.