llms.txt: what it is, an example, and whether it helps SEO

What is llms.txt?

Web pages are built for people. Navigation, scripts and layout get in the way when a language model tries to read them, and a whole site rarely fits in a model's context. The llms.txt proposal, published by Jeremy Howard of Answer.AI on 3 September 2024, offers a fix: one small file, at a predictable address, that tells an AI agent what the site is and where the useful pages are.

It sits alongside robots.txt and sitemap.xml but does a different job. robots.txt says what automated tools may access; a sitemap lists every indexable page; llms.txt is a curated short list, meant to be read on demand when an agent is helping someone.

A second version of the proposal, in August 2026, added that a file can sit at any path and covers the pages beneath it (so /docs/llms.txt covers /docs/), and that pages can offer clean Markdown copies at the same URL with .md added. It recommends standard link relations so agents can find both. The proposal itself says llms.txt is used most heavily for software documentation, where coding agents follow it to API references. OpenAI, Anthropic and Google each publish one for their developer docs.

What goes in an llms.txt file?

The format is ordinary Markdown in a fixed order, so it can be read by a model or parsed by a simple program:

• ·An H1 with the name of the site or project. This is the only required part.
• ·A blockquote with a short summary: the key facts needed to understand everything below.
• ·Optional paragraphs or lists with more detail, but no headings.
• ·Optional H2 sections, each a list of links in the form “- [Page name](URL): a short note on what it covers”.
• ·By convention, a section headed “Optional” for secondary links an agent can skip when it needs less context.

llms.txt example: our own file

Seamwire publishes one at www.seamwire.com/llms.txt. It is generated from the same content as the site, so it cannot drift out of date. A shortened version:

• ·# Seamwire
• ·> Seamwire is a software studio of 50+ people that designs and builds websites, web platforms and custom internal tools, connects business systems like ERP, CRM and accounting, and makes companies visible in AI search (AEO and GEO).
• ·## Services
• ·- [Systems integration](https://www.seamwire.com/services/systems-integration): ERP, CRM, warehouse and accounting systems connected, so data is entered once and nobody re-keys it.
• ·## Guides
• ·- [How to redesign a website without losing your Google rankings](https://www.seamwire.com/guides/website-redesign-without-losing-seo): Before the redesign, record every URL and which pages earn traffic and links.
• ·## Company
• ·- [Contact](https://www.seamwire.com/contact): Tell us where the business slows down.

Does llms.txt help SEO or AI citations?

Not on the evidence available. Google's guidance on AI features says you do not need to create new machine-readable files, AI text files or markup to appear in AI Overviews or AI Mode. In April 2025 Google's John Mueller wrote that, as far as he knew, none of the AI services had said they use llms.txt, and compared it to the keywords meta tag. In June 2026, on Google's Search Off the Record podcast, he said an AI system by design cannot trust such a file to tell websites apart, though it might help an automated system that is already on your site.

The server logs agree. Ahrefs looked at 137,210 domains with traffic in May 2026. About 28% published an llms.txt file, and 97% of those files received no requests at all that month. Of the roughly 22,000 requests the rest received, 96% came from bots, and named AI bots made up 19.5%. GPTBot was the busiest AI fetcher at 4.5% of requests; the AI search crawlers that fetch pages for answers (OAI-SearchBot, PerplexityBot and Claude's search crawler) together made up about 1.1%. A fetch is not proof the file was used, so even these are ceilings.

One small signal points the other way: Chrome's Lighthouse now checks for llms.txt in its agentic browsing audits. It only flags a server error, though, and marks a missing file as not applicable, because the file is optional.

llms.txt best practices

• ·Do the real work first: AI crawler access, server-rendered pages, answer-first content and earned mentions all matter far more.
• ·Keep it short, and link to the pages that answer real questions about your business, not every URL you have.
• ·Write one plain sentence per link that says what the page covers.
• ·Generate it from the same source as your site, so it never says something your pages do not.
• ·Never put hidden instructions or claims aimed at AI in it. Engines cannot trust it for that reason, and it can damage trust in you.
• ·Serve it as plain text with a normal 200 response, so it does not fail the Lighthouse check.
• ·If you publish software documentation, add Markdown copies of key pages; that is where the proposal is most used. Most business sites can skip this step; ours does.
• ·Do not pay for llms.txt as a stand-alone service.

Our view, and how Seamwire approaches it

llms.txt takes little effort to set up and nothing to maintain if it is generated automatically, so we publish one. But it is hygiene, not a ranking lever, and we never sell it as the answer to AI visibility. Our work starts with a free AI visibility check on your own market, then an audit and a programme that fixes what AI engines actually use: crawler access, quotable pages and mentions on the sites they cite.