Peak Answer ran the llms.txt checker on the 887 sites ChatGPT and Google cited in the crawler access study, once each in November 2026. 513 of 887 (58%) had a real llms.txt, and 17 (2%) returned a web page at /llms.txt instead. The study is descriptive only.
Hosts checked
887
Sites ChatGPT and Google cited in the crawler access study, each checked once in November 2026.
Had a real llms.txt
513 of 887
58% of hosts in November 2026.
Returned a web page at /llms.txt
17 of 887
2% of hosts answered with an OK status and a web page, which a plain fetch cannot tell from a file.
Real files that pass every check
62 of 513
12% of the real files had no error and no warning.
Had a /llms-full.txt
125 of 887
14% of the hosts checked.
Pointed to the file with rel="describedby"
32 of 513
6% of the real files had a home page that points to them.
In November 2026, 513 of 887 hosts (58%) had a real llms.txt. 17 (2%) returned a web page at /llms.txt with an OK status, 265 (30%) returned not found, 52 (6%) returned another error or no connection, 28 (3%) sent the request to an address that was not a file, 10 (1%) returned other text and 2 (0%) timed out.
Of the 513 real files, 62 (12%) passed every check of the llms.txt checker. 16 (3%) had at least one error and 451 (88%) had at least one warning.
The most common warnings were List items are links (56% of real files), Every H2 is a link list (50%) and Links go straight to the page (32%).
164 of the 448 real files with links (37%) had at least one broken link among the links checked. 171 of 421 real files (41%) had linked pages with a Markdown version.
125 of 887 hosts (14%) had a /llms-full.txt. 32 of 513 real files (6%) had a home page that points to the file with rel="describedby".
Among giant sites, 229 of 385 (59%) had a real llms.txt. Among small providers' own sites, it was 277 of 466 (59%).
By citing engine, 245 of 403 ChatGPT-cited hosts (61%) had a real file, against 290 of 524 cited in Google's organic results (55%) and 32 of 54 cited in Google AI Overviews (59%). A host cited by more than one engine counts under each.
Bars show the share of hosts in each group that had a real llms.txt and the share that returned a web page at /llms.txt.
Share of hosts with a real llms.txt and share returning a web page at /llms.txt, by group.
| Group | Real llms.txt | Web page at /llms.txt | Observations |
|---|---|---|---|
| All hosts | 513/887 (58%) | 17/887 (2%) | 887 / 887 |
| Small provider own site | 277/466 (59%) | 9/466 (2%) | 466 / 466 |
| Giant | 229/385 (59%) | 4/385 (1%) | 385 / 385 |
| Cited by ChatGPT | 245/403 (61%) | 7/403 (2%) | 403 / 403 |
| Cited by Google organic results | 290/524 (55%) | 10/524 (2%) | 524 / 524 |
| Cited by Google AI Overviews | 32/54 (59%) | 2/54 (4%) | 54 / 54 |
The numbers behind the chart, with how many observations each group has.
| Result | Hosts |
|---|---|
| A real llms.txt | 513/887 (58%) |
| A web page at /llms.txt (200, HTML or soft 404) | 17/887 (2%) |
| Other text at /llms.txt that is not a llms.txt (200) | 10/887 (1%) |
| Not found (404 or empty) | 265/887 (30%) |
| Another error or no connection | 52/887 (6%) |
| Sent to another address, and not a file there | 28/887 (3%) |
| Timed out | 2/887 (0%) |
What each host returned at /llms.txt.
| Measure | Result |
|---|---|
| Real files that pass every check | 62/513 (12%) |
| Real files with at least one error | 16/513 (3%) |
| Real files with at least one warning | 451/513 (88%) |
| Real files with links | 448/513 (87%) |
| Median links in a file with links | 51.5 |
| Real files with a broken link among the links checked | 164/448 (37%) |
| Real files whose linked pages have a Markdown version | 171/421 (41%) |
| Hosts with /llms-full.txt | 125/887 (14%) |
| Real files whose home page points to them with rel="describedby" | 32/513 (6%) |
How the real files did on the checker's checks.
| Level | Check | Real files |
|---|---|---|
| Error | H1 comes first | 16/513 (3%) |
| Warning | List items are links | 289/513 (56%) |
| Warning | Every H2 is a link list | 259/513 (50%) |
| Warning | Links go straight to the page | 166/513 (32%) |
| Warning | Links work | 164/513 (32%) |
| Warning | No headings below H2 | 103/513 (20%) |
| Warning | Summary | 87/513 (17%) |
| Warning | File size | 37/513 (7%) |
| Warning | Only one H1 | 20/513 (4%) |
The errors and warnings that came up in the most real files.
Each count is the number of real files with that error or warning, out of 513.
| Host type | Hosts | Real llms.txt | Web page at /llms.txt | Real files that pass every check |
|---|---|---|---|---|
| Small provider own site | 466 | 277/466 (59%) | 9/466 (2%) | 36/277 (13%) |
| Giant | 385 | 229/385 (59%) | 4/385 (1%) | 26/229 (11%) |
| Directory or review site | 10 | 5/10 (50%) | 0/10 (0%) | 0/5 (0%) * |
| Forum or user-generated | 8 | 1/8 (13%) * | 2/8 (25%) * | 0/1 (0%) * |
| Social | 7 | 0/7 (0%) * | 2/7 (29%) * | n/a (0 files) |
| Government or education | 6 | 0/6 (0%) * | 0/6 (0%) * | n/a (0 files) |
| Publisher listicle or guide | 4 | 1/4 (25%) * | 0/4 (0%) * | 0/1 (0%) * |
| Google Business Profile or maps | 1 | 0/1 (0%) * | 0/1 (0%) * | n/a (0 files) |
The same results by the type of host.
* Fewer than 10 observations, so the cell is thin.
A host's type is the type of most of its cited pages in the crawler access study.
| Engine | Hosts | Real llms.txt | Web page at /llms.txt | Real files that pass every check |
|---|---|---|---|---|
| ChatGPT | 403 | 245/403 (61%) | 7/403 (2%) | 26/245 (11%) |
| Google organic results | 524 | 290/524 (55%) | 10/524 (2%) | 37/290 (13%) |
| Google AI Overviews | 54 | 32/54 (59%) | 2/54 (4%) | 3/32 (9%) |
The same results by the engine that cited the host.
Peak Answer checked the 887 hosts from the crawler access study, the sites ChatGPT and Google cited for the questions in that study. The host list is the one that study used, with no host added or removed.
Each host went through the llms.txt checker's own analysis, the same code that runs when someone uses the checker on the site. It used the checker's normal user agent, at most 5 requests at a time, a 10 second timeout and one retry. No paid services were used.
A real file is a Markdown file with a top-level heading or link lists. A web page, a soft not-found page or other text at /llms.txt is not. A real file passes when it has no error and no warning on any check. For real files, the analysis follows up to 50 of the links and looks for Markdown versions of up to 25 linked pages.
/llms-full.txt was requested on every host, next to the file for real files and as its own request for the others. A response counts when it is a successful response with a body that is not a web page. Host type and citing engine come from the crawler access data.
The CSV files hold every host and the figures behind the tables. Units are in the column names.
Peak Answer, "llms.txt Adoption: How Many Sites Have One (2026)", https://peakanswer.com/research/llms-txt-adoption.
The llms.txt checker runs the same checks on your own site.
The crawler access study is where the host list comes from.