Field notes from
the answer layer.
Everything we publish comes out of the same place: a scanner that reads roughly three thousand sites, an Observer that queries four AI assistants on a schedule, and the gap between what those two things say. Some posts are essays about where discovery is heading. Others are the exact three changes that moved one customer's citation rate. All of them carry the numbers behind the claim.
There are four kinds of post here and they are worth reading differently. Essays argue a position about where search is going and are the ones to read when you are trying to convince someone else — a founder, a board, a sceptical head of content — that the AI answer layer is worth a budget line. Guides are procedural and assume you have already been convinced; read those with a browser tab open on your own site. Case studies are the honest middle: one company, one starting position, the specific interventions, and the measured result including the parts that did not work.
Technical posts are the smallest category and the densest. They cover formats and specifications — the llms.txt proposal, schema shapes, the mechanics of how retrieval layers pick chunks — and they are the ones that age fastest, because the specs themselves are still moving. If you only have time for one thing, start with the featured post below; it is the closest thing we have to a single-page summary of the whole thesis, and every other post assumes it.
Essays
1 postArguments about where discovery is going, written to be forwarded. Read these when you need to make the case to someone who controls a budget.
Guides
1 postProcedural, tested, and written to be executed rather than admired. Open your own site alongside them.
Case Studies
1 postOne company at a time, with the starting position, the interventions in order, and the measured outcome — including the changes that did nothing.
Technical
1 postFormats, specifications and mechanics. The densest posts here, and the ones most likely to need revision as the specs move.
How often do you publish?
Roughly every two weeks, and only when there is a result worth writing up. We would rather publish four posts a quarter that each carry original data than a weekly post that restates the category consensus. Freshness matters for AI retrieval, but a thin post published on schedule does more harm than good — it dilutes the pages that are actually working.
Where do the numbers in these posts come from?
From the AIRank scan cohort and the Observer query panel. The scanner grades sites against a 47-point rubric; the Observer queries ChatGPT, Claude, Perplexity and Gemini on a fixed schedule with a panel of buyer-intent prompts and logs every citation. When a post cites a figure it names the cohort and the sample size, so you can judge how much weight it deserves.
Can I republish or quote these posts?
Quote freely with a link back — that is the whole point of writing in extractable chunks. For full republication, get in touch first so we can agree a canonical tag; syndicating a duplicate without one splits the signal across two URLs and both of us end up worse off.
What should I read if I am completely new to GEO?
Read the featured essay on the shift from SEO to GEO for the frame, then the practical guide to getting cited by ChatGPT for the first set of changes. After that the glossary explains any vocabulary the posts assume, and the how-to guides turn the arguments into step-by-step work you can ship the same afternoon.
Read the argument here, then ship the change. The how-to guides are the executable version.
Browse the how-to guides