Last updated:
llms.txt Checker and Validator
This llms.txt checker reads a site's llms.txt and checks it against the proposed format, line by line.
Our server fetches the site's /llms.txt (or the llms.txt address you give) as SchemaBizBot, then it's checked here in your browser.
How to use the llms.txt checker
"Fetch llms.txt" takes a site's address and fetches /llms.txt at its root, or the file itself when the address ends in llms.txt. Our server fetches it as SchemaBizBot, as plain text or Markdown, and the file is checked in your browser. A 404 means the site has no llms.txt at that address. "Paste llms.txt" checks a draft in your browser only. A link such as /ai/llms-txt-validator/?url=https://example.com/ runs the check when it opens.
"Try an example llms.txt" loads a small shop's file with three problems: on line 10 the size guide's link is a path rather than a full URL, line 11 is a list item with no link, and line 16 links to the trail shoe page a second time. The first two are warnings and the duplicate is a note.
What the results show
- Problems, each with its line: no H1 or a second one, an H1 that isn't first, headings other than H1 and H2, list items without a link or with a link in the wrong form, links without a name, relative or invalid addresses, duplicate links, empty sections, and an Optional section that isn't last. A missing summary or a file without sections is a note: both are optional.
- An outline: the name, the summary and each section with its number of links, marking the Optional one.
- The file with line numbers, to find each problem.
The llms.txt format
Jeremy Howard proposed llms.txt in September 2024 at llmstxt.org; our guide to llms.txt covers its background and support. A file has these parts, in order:
- An H1 with the name of the site or project. It's the only required part.
- A blockquote with a short summary: the key information an agent needs to read the rest.
- Any other Markdown except headings: paragraphs or lists with more detail.
- H2 sections of links. Each is a Markdown list, one link per item, written
[name](url)and optionally followed by a colon and notes about the page. - An "Optional" section, by convention: links an agent can skip when it needs a shorter context.
The proposal also suggests a clean Markdown copy of each page an agent may need, at the page's address with .md added. That's a separate step; the checker reads only the llms.txt file. This file passes with no problems:
# Example Shop
> Trail running shoes and gear, sold online and shipped from Denver, Colorado.
We answer email within one business day. Prices are in US dollars.
## Products
- [Trail running shoes](https://www.example-shop.com/shoes/trail/): Every model, with sizes and weights
- [Size guide](https://www.example-shop.com/help/sizes/): How our sizes compare with other brands
## Help
- [Shipping and returns](https://www.example-shop.com/help/shipping/): Delivery times and costs; returns are free within 60 days
## Optional
- [Blog](https://www.example-shop.com/blog/): Running guides and product news What llms.txt can and can't do
It's a proposal, not a standard, and support from the major AI providers is limited or unconfirmed: we haven't found one that documents reading it. It doesn't control access the way robots.txt does: a crawler that wants to fetch a page doesn't consult it. What it can do is give a person or an agent a short, curated map of a site: which pages matter and what each one covers. Writing one takes minutes, so it's a reasonable optional step once the parts that are documented are in order.
Those parts come first. AI crawlers must be allowed to fetch your pages, which the robots.txt checker shows for each documented AI crawler. Pages need clear text and valid structured data, which the AI Search Optimizer scores. And a sitemap tells crawlers which pages exist: check it with the sitemap checker.
Common llms.txt mistakes
- Relative links. An agent may read the file away from your site, so
/docs/means nothing; writehttps://example.com/docs/. - Prose inside link sections. Sections hold lists of links. Put descriptions in each link's notes, or in the free text above the first H2.
- Extra headings. An H3 inside a section, or a second H1, breaks the simple structure parsers expect.
- An HTML page at /llms.txt. A site that answers every address with its home page, or a 404 page with a 200 status, serves HTML; the checker says so.
- Too much. The file is a curated map, not a sitemap. Link the pages that answer the questions people ask about you, with notes that say what's on each.
Frequently asked questions
FAQ
Do ChatGPT, Claude, Gemini or Perplexity read llms.txt?
We haven't found documentation from the major AI providers or search engines saying their crawlers or assistants read llms.txt (checked November 2026). The checker tells you whether a file follows the format; it can't tell you who reads it. Treat llms.txt as a cheap, optional step, not a ranking factor.
Where does llms.txt go?
At the site's root, as https://example.com/llms.txt, served as plain text or Markdown. The proposal also allows one at a path, covering the pages under it. Give the checker a site's address and it fetches /llms.txt there; give it the address of a file ending in llms.txt and it fetches that.
Is llms-full.txt part of the format?
No. Some sites publish an llms-full.txt with the full text of their documentation in one file, but llmstxt.org doesn't define it. Paste one into the checker and it's read with the same rules, which may not fit it.
What does the Optional section do?
Links under an H2 named "Optional" are ones an agent can skip when it needs a shorter context: background, older material, nice-to-haves. By convention it comes last, and the checker notes one that doesn't.
Does the checker store my llms.txt?
A pasted file is checked in your browser and never leaves it. For a site address, our server fetches the file as SchemaBizBot and sends it back to your browser, where it's checked. Like every tool on the site, the checker records what kind of check ran: counts of sections, links and problems, and the problem codes, never the file or its links. A fetched file's address is recorded without its query string. See the privacy policy.