Four layers, and the order matters
Anyone wanting to show up in AI search quickly finds a list of measures. What gets lost: three of them do nothing as long as the first is missing.
The question almost always arrives in the same shape: “What do we have to do so that ChatGPT recommends us?” And the answer you find online is usually a list. Structured data here, a file there, an entry in the robots protocol.
Lists have a problem: they suggest the items are of equal weight and can be worked through in any order. For findability by assistants, that isn't so. There are four layers, they build on one another, and a skipped layer renders the ones above it useless.
Layer 1 — Being found
An assistant has to get hold of your site as a source in the first place. This is the layer of signposts: a clean sitemap, an RSS feed, the files under .well-known with which a domain describes itself.
This layer is unspectacular, which is why it gets skipped. It is also the only one without which nothing else counts: what isn't found isn't read, however well it is marked up.
Layer 2 — Being readable
A person skims a page and finds the passage that answers their question. A language model gets excerpts. Whether the excerpt that reaches it still makes sense on its own is decided by your content model.
In practice that means: sections that stand alone. Definitions that are definitions and not subordinate clauses. Questions as headings, when the text answers questions. None of this is new, and none of it is a trick — it is the same structure that helps a reader in a hurry.
Layer 3 — Permitting, or not
Who may read your content is your decision, and it breaks down into several: may a model be trained on it? May your page be quoted in an answer? May an assistant fetch it at runtime?
Those are three different questions with three different answers, and there is now a notation for them. Block everything wholesale and you also disappear from the answers you would like to appear in. Regulate nothing and you have decided as well.
Layer 4 — Offering something
The top layer is the one everyone thinks of first: interfaces through which an assistant doesn't just read but does something. Check availability, book an appointment, search a catalogue.
It is also the most expensive. That is why it sits up here and not at the top of the list: an interface nobody finds, because layer 1 is missing, is an interface that cost money and changed nothing.
What that means for the first step
You can work through the four layers from the bottom up, and that is also the only order that pays off. The finding of where a site stands today is therefore less a score than a position: it holds up to here, and no further.