llms.txt generator and validator for the v2 spec

What is llms.txt?

llms.txt is a Markdown file at the root of a website (/llms.txt) that gives AI agents a short, curated map of the site: its name, a one-paragraph summary and lists of links to the pages worth reading. Jeremy Howard of Answer.AI proposed it in September 2024 and published version 2 of the proposal on 10 August 2026.

An agent helping someone, such as a coding assistant looking up an API, reads the file first and fetches only the pages it needs, ideally as clean Markdown rather than HTML wrapped in navigation and scripts. The proposal says the file is meant for this on-demand use, not for training models.

What does the llms.txt format require?

Only an H1 with the site or project name is required. The other parts are optional but come in a fixed order, so a program can parse the file as well as a model can read it.

Scroll the table sideways to see every column.

A byte order mark may precede the H1. A file can also sit in a subfolder, such as /docs/llms.txt: it covers the pages under its path, and where two files apply, agents should use the more specific one.

What changed in llms.txt v2?

The file format did not change, so a valid v1 file is still valid. The v2 changes are about everything around the file:

• Discovery. A page can point to its Markdown version with rel="alternate" type="text/markdown" and to the llms.txt that covers it with rel="describedby", as an HTML link element or an HTTP Link header. The generator writes both.
• Markdown URLs. Both page.html.md and page.md are allowed; a URL without a file name uses index.html.md or index.md.
• Scope. A file covers everything under its path, so a project that controls only a folder, such as a GitHub Pages site, can publish its own.
• Use. Agents are expected to read or search the file and follow its links, so links should point at LLM-friendly content. The context-expansion tool from v1 is no longer part of the proposal, and the Optional section is now only a convention.

How to create an llms.txt file: a worked example

1. Copy your sitemap (usually yoursite.com/sitemap.xml) into Import links, or use Paste an example sitemap.
2. Choose Import links. The six example URLs become three sections: Pages (home and pricing), Docs and Blog. A page alone in its folder joins Pages.
3. Prune. Agents need the pages that answer questions, not every post, so delete the rest and move background reading to an Optional section.
4. Add a short note after each link on what the page answers.
5. If your site serves Markdown copies, tick the .md option before importing: /docs/api becomes /docs/api.md.
6. Download the file, upload it to your site root, then paste the live file into the validator to check your server returns it, not an error page.

# Example Analytics

> A hosted dashboard for small online shops.

## Docs

- [Quick start](https://example.com/docs/quick-start.md): Connect a store
- [API reference](https://example.com/docs/api.md): Endpoints and limits

What is llms-full.txt?

llms-full.txt is not part of the llms.txt proposal. It is a convention from documentation platforms: Mintlify, for example, generates one that combines a whole documentation site into a single file, each page with its title, URL, description and full Markdown. It suits a small doc set an agent can load in one request; a large site soon outgrows a context window.

The proposal's own companions are the ones in the v2 list: Markdown versions of pages, two link relations and files in subfolders. It skips /.well-known/ because well-known addresses exist only at a domain's root, while an llms.txt can describe any path.

Does Google or any AI company use llms.txt?

No major search engine has said it uses llms.txt. Google says the opposite: its AI optimization guide, updated 10 July 2026, says you do not need "new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them."

We found no statement from Microsoft Bing, OpenAI, Anthropic or Perplexity that their crawlers or search features read llms.txt (checked 17 October 2026). What does exist:

• Chrome's Lighthouse has an experimental agentic browsing llms.txt audit: it flags a /llms.txt that returns a server error and marks a missing file not applicable.
• OpenAI, Anthropic and Google's Gemini API publish llms.txt files for their developer docs, which is not the same as reading other sites' files.
• The proposal's author reports that coding agents follow these files when pointed at documentation, which is where llms.txt earns its keep.

Treat it as a cheap convenience for agents that come looking, not a ranking factor or an access control. Crawling is governed by robots.txt; our robots.txt checker shows it crawler by crawler.

Common llms.txt mistakes

• Serving HTML. A site that answers /llms.txt with its home page or a 404 page gives agents nothing to parse.
• Two H1s, or text before the H1. The H1 comes first, and there is only one.
• Prose under H2 headings. The format expects link lists there; the proposal's reference parser stops at a paragraph inside a section. Put explanations above the first H2.
• Relative links. A file read away from your site cannot resolve /docs/api; use full URLs.
• Dumping the sitemap. Hundreds of bare links defeat the point.
• Expecting it to block anything. It invites; it does not restrict.

How Olano helps

We publish llms.txt files for olano.ai and olano.sg. Run through this validator on 17 October 2026, the olano.ai file had no errors and nine warnings: five sections held prose or plain bullets, such as the tools our home page offers in-browser agents, instead of links. We moved that prose above the first H2, where the spec allows it, and the file (227 lines, 122 links in four H2 sections) now checks clean. We publish no llms-full.txt, and our robots.txt names 20 search, AI and answer-engine crawlers and allows them all.

The Olano platform is where agents go from reading a business to working in it: AI agents that act across 15+ messaging channels and 450+ connector tools, with approvals before sensitive actions. The documentation shows how it fits together, and plans start at US$25 a month.