Web adoption
how many sites get ready for AI, from a census of public codeStructured data is everywhere. AI files are just starting.
Sites whose public code contains each signal. Log scale: each gridline is ten times the last.
The most-cited sites are the least likely to have llms.txt
Share of sites with an llms.txt file, by citation rank. Scan of 37,894 AI-cited sites, 14 Mar 2026.
Developer tools lead. Reference sites have none.
Share of sites with llms.txt, by category.
No sign it gets a site cited more
Sites with and without llms.txt, across 337,362 AI citations.
Among the 5,000 most-cited sites the difference is too small to tell from chance (p = 0.85).
This compares sites, it does not test cause. 882 citation snapshots, 14 Mar 2026.
Ask-AI buttons point to Claude most
Of the 249 sites whose buttons name two or more AIs, how many name each.
38,303 sites talk about getting found by AI
Sites whose pages use each phrase. A lead indicator of what comes next.
A site can use more than one phrase, so rows add up to more than 38,303.
How we measure
On 30 Mar 2026 we searched PublicWWW, an index of public website source code, for 13 signals that a site is getting ready for AI. A site counts when its live HTML, CSS or JavaScript contains the signal. Separately, on 14 Mar 2026 we checked 37,894 AI-cited sites for an llms.txt file and compared how often AI cited them.
Two searches hit a result cap (FAQPage at 100,000, Speakable at 1,000), so those counts are floors. Sites only count if PublicWWW indexed them, and the 833,791 distinct sites seen across all searches is not the size of the web, so we never show shares of it. This is a snapshot and does not update.
How many sites have an llms.txt file?
In the PublicWWW census of 30 Mar 2026, 33,743 sites named an llms.txt file in their public code and 24,943 an llms-full.txt. In a separate scan of 37,894 AI-cited sites on 14 Mar 2026, 13.3% had one, and 3 of the 50 most-cited did. The census counts sites PublicWWW indexed, not the whole web.
Is the web getting ready for AI search?
Some of it, unevenly. Structured data is common: HowTo schema appears on 89,624 sites and FAQPage on at least 100,000. AI-specific signals are rarer: 6,318 sites name GPTBot in their code and 7,029 carry an ask-AI button. These are counts of sites in the code index on 30 Mar 2026, not shares of the web.
Does adding llms.txt increase AI citations?
We see no sign that it does. In the 14 Mar 2026 scan, sites with llms.txt averaged 6.8 AI citations and sites without 6.7, with 3 for the middle site either way. Among the 5,000 most-cited sites the difference is too small to tell from chance (p = 0.85). The scan compares sites; it does not test cause.