GEO

What it is

GEO, generative engine optimization, is preparing a page for systems that do not show a list of results but write an answer and name a few sources under it. ChatGPT, Perplexity, Gemini and the AI overviews in Google all work that way. Classic SEO competes for a place on the list; GEO competes for being one of the sources, and for the model naming the business.

The name comes from a late-2023 paper by researchers at Princeton, who measured which changes to a text raise its chance of being cited in a generated answer. Since then the market has attached several more acronyms to the same field: AEO, LLMO, AIO. In practice they mean one list of changes to a website.

How GEO differs from SEO

By who reads the result and what counts as success. In SEO a person reads it, scanning a list, and success is a position and a click. In GEO a model reads it, assembling an answer from a few fetched pages, and success is the business being named and the page being linked.

GEO does not replace SEO, because it stands on it. Most AI systems do not answer from memory: they search, fetch pages and assemble the answer from what they managed to fetch. A page a search engine does not index, or one that renders its text with a script, does not exist for a model either.

How GEO differs from AEO

By nothing that would change the list of work. AEO grew out of featured snippets and voice assistants, so it stresses the format of an answer: the question in the heading, the answer in two sentences. GEO looks at how generative models choose sources and how they cite them. Both are done on the same page, in the same week.

What actually helps

Four things, in the order they pay off. One page that says plainly what the business is, for whom and where, because a model that has to infer it guesses. Facts stated in one place rather than spread across three paragraphs. Structured data for the organisation and its services. And a page a crawler can fetch without running a browser: light HTML and, where it makes sense, Markdown versions and an llms.txt file. This site has both, so it can be checked by adding .md to the address of any page.

How it is measured

There is no Search Console for AI answers, so the measurement has to be built: a fixed panel of the questions customers ask, run in the same systems before and after the changes, with the date recorded. Two free signals in between are referrals from assistant domains in GA4 and Search Console’s report on results with AI features. Anyone quoting a “position in ChatGPT” or a 0–100 score is inventing the measure, because the answer is not a list.

What GEO is not

It is not a separate discipline with its own budget, and not a reason to rewrite a site. It is not a trick for one prompt either: the models, their data and their retrieval change outside the site owner’s control, so nobody honest guarantees a citation. GEO is sound SEO plus a clarity nobody demanded before, because the reader used to be a person who already knew the business.