llms.txt explained:
the robots.txt of the AI era
A small file in your site’s root that tells language models what your business is and which pages matter. Here is what it does, what it does not do, and whether you need one.
By Manuel Cilurzo · AI consultant, Belgium · updated August 2026
llms.txt is a plain-text file in your website’s root — yourdomain.com/llms.txt — that tells language models, in readable language, what your organisation does and which pages carry the important facts. It is written in Markdown and meant to be read by a model, not rendered in a browser.
Why it exists
An AI model visiting your site gets the same thing a human does: navigation, cookie banners, sliders, a footer with forty links, and somewhere in between the three sentences that actually matter. That costs context and introduces noise. On long pages, the answer measurably suffers for it.
llms.txt inverts that. Instead of making the model dig, you lay it out: this is what we do, this is what it costs, these are the pages worth reading. It is a table of contents written for a reader that cannot click, cannot scroll and has no patience.
What belongs in it
The format is deliberately simple: one H1 with your name, a blockquote summarising you in a paragraph, then sections of links with short descriptions. In practice this structure works best:
- Who you are — name, what you do, for whom, where, in which languages. One paragraph, unambiguous.
- Your entity — contact details, LinkedIn, related domains. This helps a model connect you to everything else it knows about you.
- Your key concepts — if you work with your own terminology or a named model, define it here. It is the one place where you can dictate to a model how you name something.
- Your offer with prices — concrete figures, not “on request”. A model that finds no price will not name you in an answer where price matters.
- The pages that matter — with a line of explanation each. Not your entire sitemap: that is exactly the problem you were solving.
Keep it compact. A fifty-page llms.txt is a sitemap with extra steps.
What it is not
- Not an official standard. It is a proposal gaining adoption, not something ratified by a standards body. The name is unfortunate too — it resembles robots.txt while doing the opposite job.
- Not a guarantee of being read. No AI provider has formally committed to fetching this file. Some crawlers demonstrably do; others do not.
- Not a replacement for structured data. Schema.org remains the stronger signal, because it is used by both search engines and models and is far more precise. llms.txt sits on top of it.
- Not a way in. You do not get into ChatGPT by publishing this file. Anyone promising that is selling air.
Should you have one?
The trade-off is asymmetric, which makes the decision easy. Writing a good llms.txt takes an hour or two. If it goes unread, you lost two hours and gained an excellent summary of your own business — something a surprising number of companies have written down nowhere. If it does get read, you gain influence over how models describe you, in a place where your competitor has nothing.
When the trade-off is between “two hours” and “might decide whether you get named”, the arithmetic is short.
Three mistakes I see often
1. Pasting in the entire sitemap
A plugin that lists every page automatically turns llms.txt into a second sitemap. The whole value was in the selecting. Pick fifteen pages, not three hundred.
2. Writing marketing language
“We empower businesses with innovative end-to-end solutions” tells a model precisely nothing. Write it the way you would brief a new employee: what you do, for whom, where, what it costs.
3. Letting it go stale
An llms.txt with last year’s prices is worse than none. A model confidently quoting your outdated price makes you look unreliable in a conversation with your own customer.
Where it fits in the bigger picture
llms.txt is one brick in the second layer of what I call the agent ladder: readability. Without the first layer — being retrieved at all — it changes little. With the third layer added, where an agent can actually do something with you, it starts to matter properly.
To see where your business stands across all four layers, an AI readiness audit measures it. To simply see what AI says about you today, the free scan costs nothing.
Frequently asked questions
Where do I put llms.txt?
In your domain root, reachable at https://yourdomain.com/llms.txt and served as plain text. On WordPress that can be a real file in the web root; on closed platforms sometimes via a page plus a rewrite rule.
Does llms.txt replace my sitemap?
No. An XML sitemap is for search crawlers and lists everything. An llms.txt is for language models and deliberately selects. You want both, and they do not conflict.
Is llms.txt an official standard?
No. It is a proposal that has been gaining adoption since 2024, but no AI provider is obliged to read it. Treat it as a cheap opportunity, not a guarantee.
Will llms.txt get me into ChatGPT?
It can help, but on its own it is not enough. Being named in AI answers depends mostly on presence in the sources models consult, clear phrasing, and consistency of your data across those sources. llms.txt makes it easier for a model to understand you correctly once it has already found you.
Want to know what AI says about your business?
I put five real customer questions to four AI assistants and send you the literal answers — including who gets recommended instead of you. Free, within 48 hours.