- [ Home ](/) / - [ Blog ](/blog/) / - State of llms.txt 2026 # State of llms.txt adoption, June 2026 No global census of llms.txt exists, so we built a tracker: 219 notable hosts, fetched and parsed every week. Here is what the data actually says, with the raw JSON published alongside. Last updated: August 3, 2026 ## Why we measure it Most articles about llms.txt cite adoption numbers nobody can verify: “tens of thousands of sites”, “exploding adoption”, rarely with a source, never with a methodology. That is a problem in both directions: boosters oversell the convention, and skeptics dismiss it, with the same absence of data. So we measure it. Not the whole web (nobody can), but a fixed, published panel of 219 well-known hosts across eight sectors, re-checked automatically every week. A cohort you can compare run over run. ## How the tracker works The checker requests `https:// /llms.txt` for every panel member, follows redirects, and counts a host as serving only when the response is HTTP 200, plain text rather than an HTML app shell, and starts with a valid Markdown H1, the spec's one hard requirement. Unreachable hosts are excluded from the base rather than counted as misses. The panel, the raw results and the checker itself are all public (links below), so anyone can audit or reproduce the numbers. ## The numbers (2026-08-03) Of 218 hosts tested, **113 (51.8%) serve a llms.txt file**. The median file weighs 14.0 KB, and 90 of the 113 files include the recommended blockquote summary. Sector Serving / tested Rate Developer tools 42 / 61 68.9% SaaS 32 / 48 66.7% E-commerce 7 / 13 53.8% AI & ML 19 / 36 52.8% Fintech 8 / 16 50% Other 4 / 23 17.4% Docs platforms 1 / 7 14.3% Media 0 / 14 0% ## How to read them The spread between sectors is the real story. Developer-facing companies adopt heavily: their users ask AI assistants about APIs and SDKs all day, so a curated map of the docs has an obvious audience. Media sites have almost entirely ignored the convention, which is coherent with their posture toward AI crawlers generally: many block them in robots.txt rather than guide them. Two honest caveats. First, the panel skews toward technology companies by construction, so the headline rate is a cohort number, not a web-wide estimate; the value is in the trend and the sector contrast. Second, serving the file says nothing about whether AI systems consume it: [Google has said it does not use llms.txt](/does-google-use-llms-txt/), and no major provider has confirmed inference-time use. The case for the file rests on [narrower, documented effects](/does-llms-txt-work/), not on a confirmed pipeline. ## Takeaways - Among notable developer-facing hosts, llms.txt is now common enough that *not* serving one makes you the exception in that cohort. - The median file is small (14.0 KB): adopters treat it as a curated map, as the spec intends. - Outside tech, adoption remains marginal. Anyone selling llms.txt as a universal standard is ahead of the data. ✓ Check your own file The validator applies the same rules as this tracker. If your file passes, you can embed the validated badge. The full verified list lives in the [directory](/examples/), re-checked every Monday. These numbers update automatically each week; this page always shows the latest run. ## Sources - [ Raw stats (JSON, regenerated weekly) ](https://llmtxt.info/adoption-stats.json) - [ Measurement panel (JSON) ](https://llmtxt.info/adoption-panel.json) - [ llmstxt.org, official spec ](https://llmstxt.org/) On this page - [ Why we measure it ](#why) - [ How the tracker works ](#method) - [ The numbers ](#results) - [ How to read them ](#reading) - [ Takeaways ](#takeaways)