What is llms.txt? It is a plain Markdown file at a website’s /llms.txt address that gives AI agents a short, curated map of the site: a name, a one-line summary and links to the pages worth reading. Jeremy Howard of Answer.AI proposed it in 2024. It is still a proposal, not an official web standard.
The file had a busy 2026. Howard published a revised v2 of the proposal on August 10, 2026. Chrome’s Lighthouse started auditing llms.txt in May 2026, while Google Search’s guide to generative AI features, last updated in July 2026, says Search ignores the file. Some SEO tools now flag sites without one, so site owners want to know whether the file matters.
What is llms.txt? The format, with an example
An llms.txt file is a Markdown file served at /llms.txt that summarizes a website for large language models and links to clean versions of its key pages. Jeremy Howard of Answer.AI proposed llms.txt on September 3, 2024. The only required element is an H1 with the name of the site or project. Everything else in an llms.txt file is optional.
Howard’s original Answer.AI post argued that site authors know best which pages an AI model should use. The llms.txt specification sets a fixed order for the file.
- An H1 with the name of the project or site (required).
- A blockquote with a short summary that explains the rest of the file.
- Optional paragraphs or lists with more detail, but no headings.
- Sections under H2 headings that hold “file lists”: Markdown links in the form [name](url), each with an optional colon and a short note.
A minimal file for a fictional software company could look like the one below.
# Example Co
> Example Co makes invoicing software for small agencies.
## Docs
- [Quick start](https://example.com/docs/quick-start.md): set up an account and send a first invoice
- [API reference](https://example.com/docs/api.md): endpoints, authentication and rate limits
## Optional
- [Changelog](https://example.com/changelog.md)
The spec also suggests a clean Markdown copy of each linked page at the same address with .md added, such as page.html.md or, since v2, page.md. By convention, an H2 section titled “Optional” holds links an agent can skip when it needs a shorter context.
What changed in v2
The v2 revision, dated August 10, 2026, makes three practical changes. A file at a subpath such as /docs/llms.txt now covers the pages under that path, and when several files apply, agents should use the most specific one. Pages can point to their covering file with rel=”describedby” and to their Markdown copy with rel=”alternate” type=”text/markdown”, in an HTML link element or an HTTP Link header. Markdown copies can now use either form, page.html.md or page.md.
The spec keeps the file out of /.well-known/ on purpose: many authors control only a folder on a shared host, not the root of the domain.
llms.txt vs llms-full.txt
The llms-full.txt file is not part of the llms.txt specification. Mintlify, a documentation platform, described it in a November 2024 post as an auto-generated file that compiles all documentation text into a single Markdown document. On October 22, 2026, the llms-full.txt on Anthropic’s developer-docs site measured about 46 MB.
Who uses llms.txt: our check of the Forbes AI 50
Of the 50 companies on the Forbes AI 50 2026 list, 30 (60%) served a valid llms.txt file on their main domain in a BuySI Domains check on October 22, 2026. Seven of the 50 also served llms-full.txt. Of the 30 valid files, 27 passed Lighthouse’s three content rules, and only 2 homepages sent the v2 rel=”describedby” Link header.
Our script requested /llms.txt and /llms-full.txt on each company’s main domain and followed redirects. We counted a file as valid only if it returned HTTP 200, was not an HTML page and began with “# “. We left out developer-docs subdomains, and results can change from day to day.
| Measure (Forbes AI 50 2026 primary domains) | Result |
|---|---|
| Serve a valid /llms.txt | 30 of 50 |
| Also serve a valid /llms-full.txt | 7 of 50 |
| Valid files that pass Lighthouse’s content rules | 27 of 30 |
| Valid files served as text/markdown | 0 of 30 (all text/plain) |
| Homepage sends a rel=”describedby” Link header | 2 of 50 |
Source: checked by BuySI Domains on October 22, 2026.

The three files that failed Lighthouse’s rules listed bare URLs or had no Markdown links at all. Several files read like documentation indexes rather than company summaries. That matches the spec’s own note that llms.txt sees its heaviest use in software documentation, where coding agents follow it to API references and tutorials.
The best-known labs follow a pattern of their own. Anthropic, OpenAI and Perplexity had no llms.txt on their main domains, but each serves one on its developer-docs site. Publishing a file for developers is a different thing from reading other sites’ files, and we found no public documentation that any of the three does the second.
What Google and Chrome Lighthouse say about llms.txt
Google Search ignores llms.txt. Google’s guide to optimizing for generative AI features, last updated July 10, 2026, says creating the file “will neither harm nor help your site’s visibility or rankings in Google Search.”
Google’s John Mueller voiced doubts earlier. Search Engine Journal reported on April 17, 2025, that he called the file “comparable to the keywords meta tag” in a Reddit comment. Like that tag, llms.txt is the site owner’s own claim about what the site covers.
Chrome, a separate Google product, takes a more practical line. Lighthouse added an experimental Agentic Browsing category in version 13.2.0 on May 1, 2026, and turned it on by default in version 13.3.0 on May 7, 2026. The Chrome documentation calls the category experimental and based on proposed standards, and says testing it requires Chrome 150 or later.
What the Lighthouse llms.txt audit checks
The Lighthouse llms.txt audit fetches /llms.txt and passes the file only if it has an H1 line, at least one Markdown link in the form [text](url), and at least 50 characters. A 5xx server error fails the audit.
In Lighthouse 13.5.0, the latest release as of October 22, 2026, a 404 or any other 4xx response counts as not applicable, because the file is optional. That includes an HTTP 429 rate-limit response, although a change merged on October 9, 2026, and not yet released, makes a 429 fail as well.
The audit does not check the blockquote, the H2 sections, the content type or whether the links work, so a file can pass and still be a poor guide for an agent. The category also reports a ratio of passed checks rather than a 0–100 score. Google has not tied it to Search rankings.
Does llms.txt work? What the evidence shows
We found no published study showing that llms.txt improves search rankings or AI citations. An SE Ranking study of nearly 300,000 domains, dated October 15, 2026, found that 10.13% of them had an llms.txt file. It also showed no correlation between having one and being cited by AI systems. Removing the llms.txt variable from SE Ranking’s model even made its predictions more accurate.
A Search Atlas study of 13,911 domains across six AI platforms, last updated June 19, 2026, reached a similar result: its data does not support a causal claim that llms.txt raises visibility. Both are vendor studies of correlation, so neither proves the file can never help. Tracking whether assistants mention a brand at all is a separate job, covered in our guide to AI visibility.
As of October 22, 2026, we found no statement from an AI company that its crawlers or assistants use other sites’ llms.txt files to decide what to read or cite. Site owners have reported bot requests for the file in their logs, but a request does not show that a model used it. Howard’s v2 page says coding agents use llms.txt files reliably; that is the author’s view, not a measured result.
The clearest use today is narrower than search. The spec describes llms.txt as a file an agent reads on demand while it helps a user, mainly at inference time rather than for training. A developer can give a coding agent a docs site’s llms.txt, and the agent gets a clean list of the right pages.
How to write and publish an llms.txt file
To write an llms.txt file, pick the key pages, describe them in Markdown under an H1 and H2 sections, publish the file at /llms.txt and test it. The file stays short, and Google says it will neither harm nor help a site in Google Search. The four steps below cover most sites.
- Pick the pages. Choose the pages an agent needs to answer common questions: product, docs, policies and contact. Curate; a sitemap already lists everything.
- Write the file. Start with “# Name”, add a short blockquote summary, then H2 sections with [name](url) links and brief notes after a colon.
- Publish it at the root. Serve it at /llms.txt with HTTP 200 as plain text or Markdown. A docs section can carry its own file at /docs/llms.txt.
- Test it. Run the Lighthouse Agentic Browsing audit. Then follow the spec’s advice: ask an AI agent questions about your content and give it only your llms.txt as a starting point.

Many platforms can generate the file for you. The llmstxt.org integrations list includes Yoast SEO and AIOSEO for WordPress, Wix, and documentation tools such as Mintlify and GitBook. Review any generated file before you rely on it: three of the 30 AI 50 files we checked failed Lighthouse’s rules.
An llms.txt file does not block or allow anything. To control which AI crawlers may visit a site, use robots.txt, which the spec itself treats as the access file. A clean llms.txt is one small part of LLM optimization.
llms.txt on .si sites
.si is Slovenia’s country-code domain. For AI companies it reads as SuperIntelligence. That reading is our framing, not an official meaning, and our guide to the .si domain covers the registry facts. For a brand that leans on that AI reading, being easy for AI agents to read fits the name.
None of the large .si websites we checked on October 22, 2026, served an llms.txt file. BuySI Domains checked 14 major .si sites across government, media, banking, telecom, retail, energy and education: two blocked automated requests, and none of the other 12 served llms.txt or llms-full.txt.
Our own site publishes one. The buysidomains.si llms.txt has an H1, a blockquote summary and seven H2 sections. They run from a portfolio overview and a “Why .si” section to domain pages, articles, the buying process, an FAQ and contact details. It passes all three Lighthouse rules. Some sections mix prose and a table with the link lists, a small step away from the spec’s file-list format.
A .si name can also describe the work itself. The name llmo.si, short for LLM optimization, is one example built around how AI systems read the web. BuySI Domains lists llmo.si for sale.
Should you add llms.txt to your website?
Add llms.txt if you publish documentation or sell to developers; treat it as optional everywhere else. The file costs little to write and does not affect Google Search either way, but no study we found has linked it to more AI citations.
| Site type | Add llms.txt? | Reason |
|---|---|---|
| Developer docs, APIs, dev tools | Yes | The spec names software docs as the main use |
| Product or SaaS site | Reasonable | Low effort; gives agents a curated map |
| Small brochure site | Optional | No proven effect; harmless in Google Search |
| Goal: higher Google rankings | No effect | Google Search ignores the file |
If you add one, keep it short and accurate, and update it when key pages move. Spend most of your effort on the pages themselves, since generative engine optimization still starts with content that AI systems can find, read and quote.