Study · 14 November 2026
How many large websites have an llms.txt?
About four in ten, and almost all of them sell to developers. Of 48 well-known sites that gave a clear answer, 19 have an llms.txt: 10 of 14 software companies and 6 of 9 developer tools, against 0 of 9 news sites. Only 5 of the 19 follow the published format on every point, and 3 are over 100 KB.
- have an llms.txt
- 19 of 48
- follow the format on every point
- 5 of 19
- list more than 100 links
- 9 of 19
- links in the median file
- 90
By type of site
Share of sites with an llms.txt, by type
- Software companies (10 of 14)71%
- Developer tools (6 of 9)67%
- Shopping (2 of 9)22%
- Reference and community (1 of 7)14%
- News (0 of 9)0%
Show as a table
| Software companies (10 of 14) | 71% |
| Developer tools (6 of 9) | 67% |
| Shopping (2 of 9) | 22% |
| Reference and community (1 of 7) | 14% |
| News (0 of 9) | 0% |
| Type | Clear answers | Have an llms.txt |
|---|---|---|
| News | 9 | 0 |
| Software companies | 14 | 10 |
| Shopping | 9 | 2 |
| Reference and community | 7 | 1 |
| Developer tools | 9 | 6 |
The file is a habit of companies whose customers are developers. That fits what it is useful for: giving a coding agent a clean map of documentation. Nothing here suggests publishers or shops see a reason to have one.
The files themselves
| Site | Size | Links | Differs from the format in |
|---|---|---|---|
| stripe.com | 68 KB | 307 | list lines that do not start with the link; more than 100 links |
| shopify.com | 12 KB | 74 | Nothing |
| hubspot.com | 35 KB | 225 | list lines that do not start with the link; more than 100 links |
| salesforce.com | 781 KB | 2,857 | no single title line; no summary line; no link lists under sections; more than 100 links |
| notion.so | 7 KB | 48 | no single title line |
| slack.com | 40 KB | 165 | more than 100 links |
| atlassian.com | 26 KB | 105 | more than 100 links |
| mailchimp.com | 515 KB | 984 | no link lists under sections; list lines that do not start with the link; more than 100 links |
| zendesk.com | 724 KB | 3,133 | no single title line; list lines that do not start with the link; more than 100 links |
| intercom.com | 10 KB | 67 | list lines that do not start with the link; few notes on links |
| etsy.com | 26 KB | 76 | no link lists under sections; list lines that do not start with the link |
| target.com | 6 KB | 54 | Nothing |
| substack.com | 19 KB | 72 | list lines that do not start with the link |
| github.com | 28 KB | 117 | more than 100 links |
| vercel.com | 5 KB | 28 | Nothing |
| cloudflare.com | 17 KB | 90 | Nothing |
| digitalocean.com | 12 KB | 102 | few notes on links; more than 100 links |
| netlify.com | 3 KB | 28 | Nothing |
| supabase.com | 3 KB | 44 | no summary line; few notes on links |
Three things stand out.
Few follow the format. 5 of 19 pass every check in our llms.txt checker. The format asks for one title line, a summary, and sections of links; most files drop or bend at least one of those. Nobody is enforcing it, so a "valid" file is a courtesy.
Some have lost the point. The proposal exists so a tool with limited room can read a short list in place of a whole site. 9 of the 19 list more than 100 links, and 3 are over 100 KB. Those are sitemaps in Markdown.
Some are not lists at all. Two files are written as prose and headings: briefings for an assistant, with links inside sentences. They fail the format and may serve their purpose better than a tidy list.
How we ran this
- When and where. 2026-11-14, from a connection in Germany.
- What was fetched. One request for llms.txt per site, the www address first and then the bare domain. Links inside the files were not followed.
- What counts as having one. A 200 response that is text and not a web page. A 404, or an ordinary web page served at that address, counts as not having one.
- What was left out. 2 sites did not answer (washingtonpost.com, npr.org) and 10 turned the request away: reuters.com (401), wsj.com (403), bloomberg.com (403), theatlantic.com (403), canva.com (403), homedepot.com (403), medium.com (403), imdb.com (202), tripadvisor.com (403), npmjs.com (403). A refusal says nothing about whether the file exists, so those 12 are not in the percentages.
- How the files were checked. Against the format at llmstxt.org, with the same code as the checker. "More than 100 links" is our own threshold, not part of the format.
What this does not show
Whether any assistant reads these files. Having one and its being used are separate questions, and the only measurement we know of on the second, by Otterly, found AI crawlers fetched the file 84 times out of more than 62,100 visits to its test site.
The list is 60 large sites we chose, so the share describes this list. Frase found a higher figure in a different group: 58 of 93 B2B software vendors, checked on 9 October 2026, which fits our software row.
What to do with it
If your customers are developers, generate one and keep it short. If they are not, read this first: there are better uses of the same ten minutes.
Use this data
The site-by-site results are published as JSON. Quote any figure with a link to this page. To cite it:
Content Autopilot, “How many large websites have an llms.txt?”, 14 November 2026. https://apexmention.com/studies/llms-txt-adoption