Getting found by AI, explained simply.

Generative Engine Optimization — making a business findable inside answers written by AI assistants rather than inside a list of links. The term comes from a 2024 KDD paper that was the first controlled study showing content can be deliberately optimised for AI-generated answers.

Search engine optimization (SEO) optimises to be ranked in a list of links, and assumes the person will choose from that list. Generative engine optimization (GEO) deals with systems that read several sources and write a new answer from them, so the goal shifts from holding a position to being the material the answer is built from.

No. Search engine optimization still matters, because search indexes are how much of the web is discovered, including by assistants that search before answering, so being crawlable and accurate now does double duty. What changed is that ranking and traffic have come apart — ranking is necessary but no longer sufficient.

No, and treat anyone who promises it accordingly. Assistants are run by other companies that decide what to retrieve and what to write, and their behaviour changes without notice. What can be done is to make sure accurate, well-structured information about the business exists where those systems look.

By asking. Put the questions a customer would ask to the assistants on a regular schedule and record what comes back — whether you are named, what is said, and who is named instead. Assistants send no analytics and typically no referral traffic, so measurement is the only feedback loop available.

Answer Engine Optimization is about shape: a question with its answer directly underneath, marked up so a machine knows which sentence answers what. GEO is the broader problem of being the source a generative model reaches for at all. They overlap heavily and are often used interchangeably — doing AEO well is most of doing GEO well.

Only for that conversation. Nothing you say in a chat is written to the model, and the next conversation — yours or anyone else's — starts with no knowledge of it. We tested this directly: a model that had just agreed to recommend a company described it as an entirely different business two seconds later in a fresh call. The only thing that carries across is a page on the public web.

For a business that wants to be recommended, allow them. GPTBot and OAI-SearchBot are OpenAI's, ClaudeBot is Anthropic's, and GoogleOther is Google's crawler for work other than web search. Blocking a retrieval crawler removes you from the answers it feeds. A publisher whose content is the product may reasonably decide otherwise — and can allow OAI-SearchBot while blocking GPTBot, since OpenAI documents the first as not used for training.

No. A sitemap tells a crawler a page exists; a link tells it the page is worth fetching. Our own company pages sat in a sitemap for weeks with zero fetches from any AI crawler, and were fetched four times within hours of adding one link from a page those crawlers already read.

Days to weeks, and it depends on the assistant. The page has to be published, linked from somewhere already crawled, and then fetched — after which it can be used in an answer. There is no way to make a crawl happen on demand, which is why the honest measure is to keep asking the assistants and record what changes.

No, and nobody should claim it does. What it does is remove ambiguity about which text answers which question, which improves the odds a passage is retrieved and quoted correctly. Markup on a page nothing fetches does nothing, and markup that contradicts the visible text is worse than none.

Because a number is impossible to argue with, which is exactly what is wrong with it. A sentence tells you what was said, whether it was accurate, and which competitor came up instead — so you can see what to change. A score tells you nothing about what to do next.

Not necessarily. Publishing accurate answers to the questions your customers actually ask, in a form machines can quote, is the substance of it. Articles help when you have something specific to say that nobody else has said; volume for its own sake does not.