GEO, generative engine optimisation, means optimising content so that generative AI systems use and name it as a source in their answers. The term is often used alongside AEO, answer engine optimisation; both describe essentially the same shift.
The difference from classic SEO
Classic SEO targets a position in a results list. The user sees several hits and chooses. GEO targets being part of a generated answer. The user sees an answer and possibly its sources. That changes what matters: it is not the page as a whole that competes but the individual, clearly worded statement a model can take over.
What makes content citable
Four properties make content usable to an AI system. First, self-containment: a paragraph must make sense without the surrounding text, because it is extracted on its own. Second, concreteness: checkable, specific statements get taken up; general marketing phrasing does not. Third, structure: clear headings, defined terms and a recognisable outline ease machine reading. And fourth, an identifiable source — whoever makes a statement must be identifiable as its author.
The citable unit
What works in practice is a short, self-contained answer at the start of a section — around fifty words that answer the core question completely. That passage is the unit that gets extracted. It is also what works as a featured search result, which is why work on classic and generative visibility overlaps at exactly this point.
Structured data and llms.txt
Two technical elements support machine reading. Structured data following Schema.org declares what a piece of content is — a definition, a question with an answer, a how-to, an organisation. And the file llms.txt, an emerging standard, gives AI systems an ordered plain-text overview of a site’s most important content. Neither replaces good content, but both make it easier to find and correctly classify.
What can be measured and what cannot
Measurement is the weakest part of the field. Unlike search results there is no reliable position, and answers vary between systems, users and moments. What is observable is whether and in what context your content appears in answers — by sampling, not exhaustively. Anyone expecting robust metrics will be disappointed; anyone observing visibility qualitatively gains usable signals.
Practical consequence
GEO is not a second channel beside SEO but an additional layer on the same content. The measures — self-contained short answers, concrete checkable statements, clean structure, correct annotation — improve both at once. Optimising separately for AI systems builds effort that does not pay off.