For two decades being findable meant one thing. Now it means two, and the two pull in different directions.
A search engine wants to send someone to your page. It rewards pages that look like the best destination for a click.
A language model rarely sends anyone anywhere. It reads, summarises, and sometimes cites. It rewards pages that are easy to extract a defensible statement from.
What changes on the page
For search, the opening paragraph is a promise that keeps someone reading.
For extraction, the opening paragraph is the answer, stated plainly, before any framing.
Those two are not compatible in the same first sentence. Choosing between them is a real editorial decision, not a technical setting.
What both reward
Specificity that can be checked. A model repeating a claim is exposed if the claim is wrong, so vague copy is safer to ignore than to cite. Dated, sourced, concrete statements travel.
Structure that survives being cut apart. Headings that describe their section rather than tease it. List items that still make sense with the list removed.
The uncomfortable part
You cannot measure the second one properly yet. Referral data from AI answers is thin and inconsistent across providers.
Anyone selling you a confident GEO dashboard is selling you a proxy. Write for extraction because it is a reasonable bet about where reading is going, not because you can prove the return this quarter.
