A plain-text file at a site's root (yourdomain.com/llms.txt) that gives AI systems a concise, purpose-built index of the site's content — a summary, key facts, and links grouped by topic with descriptions — analogous to robots.txt or sitemap.xml but written to be read and understood rather than mechanically parsed.

Is llms.txt an official web standard?

Not yet — it's a proposed, increasingly-adopted convention rather than a formally registered standard. It follows the same logic as robots.txt and sitemap.xml: a simple, explicit file that's easier for a machine to consume than inferring structure from rendered pages.

What's the difference between llms.txt and llms-full.txt?

llms.txt is a concise index — meant to be read in one fetch. llms-full.txt is the complete text content of every substantive page, concatenated into one file, for an AI system that wants the full detail without crawling each page individually.

Should llms.txt include FAQ answers or just questions?

Answers, ideally. A system that only fetches the concise index still gets the citable payload — questions alone force a second round-trip to get anything useful.

How do I keep llms.txt from going stale?

Generate it programmatically from the same content that renders your actual pages, rather than hand-writing a parallel file. That way it's structurally impossible for it to describe a page that no longer exists or omit one that does.

Concepts

What is llms.txt? A practical guide (with a working example)

TL;DR

llms.txt is a proposed convention — not yet a formal standard — for giving AI systems a direct, plain-text index of a website's content, the same way robots.txt directs crawlers and sitemap.xml lists pages. Where it differs from a sitemap is intent: llms.txt is written to be read and understood, with a short summary of what the site is, links grouped by topic, and (in a fuller version) FAQ answers and full page content inlined as plain text. This is a practical build guide, illustrated with a live example.

What llms.txt actually is

A plain-text file at your domain root (yourdomain.com/llms.txt) formatted roughly as Markdown: an H1 with your site name, a one-line summary, then H2-grouped sections of links with short descriptions. It's meant to be short enough for an AI system to read in one fetch and understand the shape of your site immediately — a concise map, not a data dump.

A companion convention, llms-full.txt, goes further: the full text content of every page, concatenated into one file, so an AI system that wants the complete detail doesn't have to crawl dozens of individual pages.

What to put in llms.txt

Section	Contents
Title + summary	Your site name and a one-paragraph description of what it is
Key facts	A handful of hand-written, high-value facts an AI system would otherwise have to infer
Content links, by category	Every substantive page, grouped logically, each with a one-line description — not just a URL
FAQ	Your most important questions, ideally WITH answers inline, not just the question text
Links	Your homepage, pricing, and a pointer to llms-full.txt if you have one

A live, working example

This site's own /llms.txt is a real, currently-deployed implementation — inspect it directly rather than taking a description on faith. It's generated from the same content objects that render the human-facing pages, so it can never drift out of sync with what's actually on the site. A few implementation choices worth copying:

Generate it from your existing content, don't hand-write a parallel copy. A hand-maintained file drifts the moment you update a page and forget the second file.
Include FAQ answers, not just questions. An AI system fetching only the concise index still gets the citable payload without a second round-trip.
Strip markdown link syntax from the plain-text output. A raw `label` token is noise to a system extracting meaning, not signal — convert links to plain text or keep the URL separately.
Serve /llms-full.txt for the complete picture — every doc, guide and post's full body text and FAQ, concatenated, so an AI system that wants the whole knowledge base gets it in one fetch.

Common mistakes

Treating it as another sitemap. A sitemap is a flat list of URLs for crawl discovery; llms.txt is meant to be read and understood, with actual descriptions.
Forgetting to keep it updated. If it's not generated from your real content, it will eventually describe pages you removed and omit ones you added.
Making it too long. The point of the concise /llms.txt is that it's fast to fetch and read in full — put the exhaustive detail in llms-full.txt instead.
Not linking to it anywhere. Since it's not yet a registered file type crawlers automatically check, consider referencing it from your normal pages so it's discoverable.

Is this part of a bigger strategy?

llms.txt is one concrete technique inside the broader practice of GEO (Generative Engine Optimization) — see what GEO is and the full technique checklist for structured data, answer-first writing and the rest.

Frequently asked questions

What is llms.txt?: A plain-text file at a site's root (yourdomain.com/llms.txt) that gives AI systems a concise, purpose-built index of the site's content — a summary, key facts, and links grouped by topic with descriptions — analogous to robots.txt or sitemap.xml but written to be read and understood rather than mechanically parsed.
Is llms.txt an official web standard?: Not yet — it's a proposed, increasingly-adopted convention rather than a formally registered standard. It follows the same logic as robots.txt and sitemap.xml: a simple, explicit file that's easier for a machine to consume than inferring structure from rendered pages.
What's the difference between llms.txt and llms-full.txt?: llms.txt is a concise index — meant to be read in one fetch. llms-full.txt is the complete text content of every substantive page, concatenated into one file, for an AI system that wants the full detail without crawling each page individually.
Should llms.txt include FAQ answers or just questions?: Answers, ideally. A system that only fetches the concise index still gets the citable payload — questions alone force a second round-trip to get anything useful.
How do I keep llms.txt from going stale?: Generate it programmatically from the same content that renders your actual pages, rather than hand-writing a parallel file. That way it's structurally impossible for it to describe a page that no longer exists or omit one that does.