What GEO is, and who gets cited.

For twenty-five years the job was to appear on a page of ten results and win the click. Increasingly the answer arrives written, with a handful of sources named underneath it, and most readers never click anything.

That changes what winning means. You are no longer competing for a position. You are competing to be one of the few sources an assistant draws on while composing an answer — and, ideally, the one it names.

The overlap is large and worth saying plainly: a page an assistant cannot crawl, cannot parse or cannot load is out of the running, exactly as it would be in search. Most of what has always been good practice is still the price of entry.

What changed is the unit. Search ranks pages; an assistant retrieves passages. A page that ranks well because of its links, its domain and its overall relevance can still be useless to an assistant if no individual paragraph of it answers anything on its own. The question stopped being is this page authoritative on the topic and became can this paragraph be lifted out and still be true.

Self-contained claims. A sentence that depends on the three paragraphs above it cannot be extracted. One that carries its own subject can.

Specifics that can be checked. Numbers, dates, named places, actual constraints. Assistants are built to prefer sources that commit to something, because a claim that can be verified is safer to repeat than a claim that cannot.

The answer near the question. A page that spends six paragraphs establishing why the topic matters before answering it gives a retrieval system nothing to grab. State the answer, then explain it.

Plain structure. Headings that describe what is under them, short paragraphs, lists where a list is genuinely the shape of the thing. This is not a trick — it is the same reason those pages are easier for people to read.

And a position. Answers that say something distinctive get cited more than answers that summarise everyone else, for the simple reason that there is no point citing a source that says exactly what four other sources said.

There is no published list of ranking factors for AI answers, and anyone selling you one is guessing. Different assistants retrieve differently, their behaviour changes without announcement, and the same question asked twice can produce different sources.

So the honest version of this work is: make the page genuinely worth citing, make it trivially easy to retrieve, and accept that the rest is not a dial you can turn. Anything sold as a guaranteed placement in an AI answer is being sold by someone who cannot deliver it.

Search Console will not tell you. It reports clicks from Google search, not mentions inside an assistant's answer, so the usual dashboard shows nothing while this is working or failing.

What does work is unglamorous: ask the assistants the questions your buyers ask, monthly, and write down whether you appear and what they say about you. Watch your referral traffic for the assistant domains that do pass a link. And watch for the pattern where traffic falls while enquiries hold steady — that is often not a loss, it is people arriving already informed.