Strategy

What is llms.txt, and should your B2B site have one?

Published 1 August 2026 · 7 min read · By Ripe Leads

The short answer

llms.txt is a plain markdown file published at the root of a domain, at /llms.txt, that gives a large language model a short curated map of the site: who you are, what you sell, and which URLs hold the answers worth reading. It controls nothing and guarantees nothing. Adoption by AI crawlers is still uneven, so treat it as half an hour of cheap insurance, not a traffic channel.

A proposal from late 2024 has quietly become a file that plenty of B2B sites now ship without being able to explain what it does. Here is the honest version: what llms.txt is, how to write one, and what it will and will not do for you in 2026.

What is llms.txt?

llms.txt is a plain text file, written in markdown, that lives at the root of a domain at /llms.txt. It hands a language model a compact, curated summary of the site: a paragraph describing what the company does, then grouped lists of the URLs that matter, each with a line of context.

Jeremy Howard of Answer.AI proposed the format in September 2024. The reasoning was practical. Models read with a limited context window, and a normal HTML page is mostly navigation, scripts, cookie banners and markup wrapped around a small amount of actual content. A markdown index strips all of that away and says, in a few hundred words, what a site is for and where the substance sits.

Think of it as a README for your domain, written for a machine that has one shot at understanding you.

What llms.txt is not

Three files get confused with each other constantly, so it is worth separating them.

robots.txt controls access. It tells crawlers, including AI crawlers such as GPTBot and ClaudeBot, which paths they may fetch. That is a permission file. llms.txt has no permission semantics at all.
sitemap.xml enumerates. It lists every indexable URL for search engines, with no commentary and no priority beyond a few weak hints. It is complete, not curated.
llms.txt curates. It is editorial. You choose the twenty or fifty pages that answer real questions and you say what each one contains.

You want all three, and none of them replaces another. Publishing llms.txt while blocking AI crawlers in robots.txt is a contradiction that costs you both ways.

How the file is structured

The proposed format is deliberately small. There is no new syntax to learn, and it is all valid markdown:

An H1 with the site or company name.
A blockquote holding a one-paragraph summary of what the site is.
Optional prose paragraphs for anything that needs context.
H2 sections, each containing a bullet list of links in the form: link title, URL, then a colon and one line of description.
An optional final section headed "Optional", holding links a model can skip when its context is tight.

That is the entire specification. A companion convention, llms-full.txt, holds the full text of the linked pages in one file rather than just the index. It gets large fast and it is not required.

Does anything actually read it today?

This is where most articles on the subject stop being honest, so here is the position as it stands. No major AI provider has publicly committed to using llms.txt as a retrieval source. Google's search team has said publicly that it is not used for search. Server logs at plenty of sites do show AI-related user agents fetching the file, but a fetch is not evidence that anything downstream used it.

What has adopted it firmly is documentation tooling. Several developer documentation platforms now generate llms.txt automatically, which is why a large share of the files in the wild belong to software docs rather than company sites.

So the honest verdict: nobody can promise you that llms.txt lifts your citations in ChatGPT or Perplexity. Anyone selling you that promise is guessing. What the file does reliably is force you to write one clean, consistent, machine-readable statement of what your company does, which is the same discipline that underpins generative engine optimization generally. Half an hour of work, no ranking risk, a real benefit if something fetches it. That is cheap insurance, not a channel.

30 minRoughly what it costs to write a good llms.txt for a B2B site. The value is mostly in being forced to state your facts once, consistently.

What should a B2B site put in one?

Start with the identity paragraph, and make it do real work. Name the company, say what it sells, say where it operates, state how it prices if pricing is public, and state what it refuses to promise. A model summarising you will lean heavily on this paragraph, so every clause you leave vague is a clause it will fill in from somewhere less reliable.

Then group the links into sections a buyer would recognise rather than sections your CMS invented:

Services. One line per offer, in the buyer's language, not internal product names.
Markets. Country and region pages, if you sell geographically.
Industries. Vertical pages, each with the specific problem it addresses.
Comparisons. Your honest comparison and alternatives pages. These are disproportionately useful, because comparison questions are exactly what buyers ask an AI assistant.
Guides. Link the hub, not four hundred articles. A model does not need your full sitemap twice.
About and contact. Founder, company registration, email, booking link.

A worked example

Ripe Leads ships one at /llms.txt. The summary blockquote states the whole business in a paragraph: done-for-you B2B outbound run from Vilnius for clients across Europe, campaigns in Lithuanian, English, German and Russian, GDPR-native, flat pricing of EUR 3,750 in the first month for setup and launch and EUR 2,850 a month after that, cancel anytime, and an explicit line saying we never promise a fixed number of meetings because nobody can promise that honestly.

That last clause is there on purpose. If a model is going to describe us to a buyer, we would rather it repeats our refusal to guarantee meeting counts than invents a number. The same figures appear on the pricing section of the site, word for word. Consistency across the site and the file is the entire point.

Below the summary the file runs sections for services, markets, industries, comparison pages, the Academy hub, about and contact. One line each, factual, absolute URLs.

The rules that make it useful

One factual line per link. Marketing adjectives are noise to a model. State what the page contains.
Absolute canonical URLs. Relative paths and redirect chains both cost you.
Match the site exactly. If the file says one price and a landing page says another, you have taught the model that your numbers are unreliable.
Keep it curated. A few dozen strong links beat a dump of every URL you own.
Update it when you publish. A file listing pages that 404 is worse than no file.
Serve it as plain text. text/plain or text/markdown, HTTP 200, no login, no JavaScript rendering required.

Common mistakes

Auto-generating it from the sitemap. The output is a URL list with no curation and no context, which throws away the only thing the format adds.
Writing sales copy in it. Nobody reads this file for persuasion. Claims without specifics are the first thing a model discards.
Treating it as an SEO tactic. It is not a ranking signal in any search engine. If your organic traffic is the problem, that work sits in SEO for B2B lead generation, not here.
Shipping it and stopping. The file points at pages. If those pages are thin, you have built a clean index of weak content.
Blocking the crawlers you are writing for. Check robots.txt before you celebrate.

Should your B2B site have one?

Yes, if you can write it in an afternoon and keep it current. The cost is trivial, the downside is zero, and the exercise of compressing your company into one honest paragraph is worth doing whether or not a crawler ever reads it.

No, if it becomes a substitute for the work that actually earns citations. Being quoted by an AI assistant comes from being the clearest available answer to a specific question, publishing numbers nobody else has, and being mentioned on sites you do not control. That is the real programme, covered in how to get cited by ChatGPT and Perplexity. llms.txt is a tidy front door on a building that still has to be worth entering.

Write the file, keep it accurate, then go back to the pages.

Frequently asked

What is llms.txt?

llms.txt is a plain markdown file published at the root of a domain, at /llms.txt, that gives a large language model a short curated map of the site: who you are, what you sell, and which URLs hold the answers worth reading. The proposed format is an H1 with the site name, a blockquote summary, then H2 sections containing lists of links with one line of context each. It is a summary for machines, not an access control file.

Does llms.txt actually work in 2026?

Adoption is uneven and no major AI provider has publicly committed to using llms.txt as a retrieval source, so nobody can honestly promise it lifts your citations. What it does reliably is force you to write one clean, consistent, machine-readable statement of what your company does. The file takes about half an hour to write, carries no ranking risk and helps a model that does fetch it, which makes it cheap insurance rather than a channel.

What is the difference between llms.txt and robots.txt?

robots.txt controls access: it tells crawlers, including AI crawlers, which paths they may or may not fetch. llms.txt controls nothing. It is an editorial summary that points a model at the pages you consider most useful and explains what they contain. A sitemap.xml sits in a third category again, listing every indexable URL for search engines with no commentary. You want all three, and they do different jobs.

What should a B2B llms.txt file include?

Start with a one-paragraph identity statement naming the company, what it sells, where it operates, how it prices and what it refuses to promise. Then group links into sections a buyer would recognise: services, markets, industries, comparison pages, your guides hub, about and contact. Give every link one factual line of context, use absolute canonical URLs, and keep every fact identical to what the same page says on the site.

Rather not build this yourself?

We run the targeting, data, copy and follow-up as a done-for-you service, and send the interested replies straight to your inbox. You bring the close.

Book a strategy call