Web adoption
How much of the web is getting ready for AI search — the files, signals and buttons sites add so AI engines can find, read and cite them. A source-code census of the whole web, paired with a controlled look at whether it actually works.
The AI-readiness signals
Every signal sites add to be read by AI, by how many carry it across the indexed web.
Most AI-readiness is structured data sites already had — explicit AI files like llms.txt are still the exception, not the rule.
llms.txt, up close
The one explicit “read me like this” file for AI — measured across the 37,894 domains AI actually cites.
Adoption by site popularity
Rarest among the biggest sites; most common in the long mid-tail, then it eases off.
Adoption by category
Tooling-led: the sites that build software adopt first; reference and review sites lag.
Does it move citations?
Pages with llms.txt and pages without earn the same median citations — the gap isn’t statistically real.
When sites adopt, they do it right
Among the sites that ship llms.txt, the files are well-formed and complete.
Cite buttons
The boldest move: an on-page button that hands your page straight to an AI engine.
Which engines they target
Among the 249 sites that add cite buttons for two or more AI engines.
Of the recognizable brands we deep-scanned live, all but one inject “always cite this source”-style instructions into the AI prompt.
Who’s ready, and who isn’t
The most-cited domains AI relies on — split by whether they’ve shipped llms.txt yet.
shopify.com
docs.aws.amazon.comTwo real lenses. The census counts sites whose live source code carries each signal across 834K indexed domains — a breadth snapshot, where some queries hit a result cap (shown as “+”). The llms.txt study is a controlled scan of the 37,894 domains AI actually cites, testing adoption and whether it changes citation counts. Both are point-in-time snapshots; the web moves faster than any index.