01
You choose what represents you
Left alone, a model summarises whatever page it happened to land on. An llms.txt file is the one chance you get to say which pages define your business.
Free tool
llms.txt Generator: give AI models a map of your site
Paste your URL, get a clean llms.txt file in markdown, and upload it to your site root. It takes about a minute, costs nothing, and tells ChatGPT, Claude, Perplexity and the rest which pages on your site are actually worth reading.
60s
To a finished file
Free
No signup, no email
Markdown
Plain text out
Any CMS
WordPress, Shopify, custom
The basics
What an llms.txt file actually is
llms.txt is a plain markdown file you put at the root of your domain — yourdomain.com/llms.txt — that lists the pages on your site worth reading, each with a one-line description. It was proposed by Jeremy Howard of Answer.AI in September 2024 and documented at llmstxt.org.
The problem it tries to solve is real. When a language model reaches your page it gets the whole thing: navigation, cookie banner, three calls to action, a footer with your VAT number. The useful part — what you do, who for, what it costs — is buried in markup written for browsers, not readers. An llms.txt file skips that. It says: here are the twelve pages that matter, here is what each one covers, read these.
It is not robots.txt. robots.txt tells crawlers what they may not touch; llms.txt makes no rules and blocks nothing. It is closer to a sitemap written for a reader rather than a crawler — fewer URLs, in a sensible order, each one explained.
There is usually a companion file too. llms-full.txt holds the actual content of those pages as markdown, for models that want the text rather than the links. Start with llms.txt; add the full version only if your documentation justifies it.
Fair warning, because you will read the opposite elsewhere
No AI company has publicly committed to reading llms.txt. OpenAI, Anthropic and Google have not said they use it, and nobody can show you traffic that came from one. It is a proposal with growing adoption, not a standard with proven returns. We recommend it because it costs ten minutes and cannot hurt you — not because it will move your rankings next week. Anyone promising otherwise is guessing.
How to use it
From URL to a published file in four steps
The generator does the first half. The half that decides whether the file is any good is yours.
Generate the draft
Paste your homepage address. The generator reads your homepage and sitemap, picks the pages that look like the ones a buyer would read, and writes the markdown.
Cut it down, hard
The draft will be too long. Keep the pages you would send a serious prospect — services, pricing, a case study or two, the real documentation. Delete tag archives, thin blog posts, anything you would not email a client.
Rewrite the descriptions
One line per link, written for a reader who has never heard of you. "SEO services in Nepal, with monthly plans from NPR 35,000" beats "Our SEO Services" every time — a model picks the one that explains itself.
Upload and check
Put the file at the root so it answers at yourdomain.com/llms.txt — same level as robots.txt, not in a folder. Load that URL in a browser; you should see plain text, not a 404 and not your theme.
Why bother
What you actually get out of it
02
It costs almost nothing
Ten minutes once, five minutes a quarter. Compared with anything else in AI visibility work, it is the cheapest thing on the list.
03
It forces a useful audit
Writing one line per page makes it obvious which pages cannot explain themselves. Most people fix two or three pages as a direct result of writing the file.
04
It is ready if adoption comes
If one of the big labs starts reading these files, sites that already have a good one are in the answer that day. Sites that do not are writing theirs that week.
05
It helps your own tooling
Internal assistants, support bots and retrieval tools you build later can use the same file as a reading list. We use clients' llms.txt files when we build their chatbots.
06
It cannot hurt you
It does not block crawlers, change your HTML or affect how Google renders your pages. The worst case is that nothing reads it.
What goes in
What belongs in the file, and what does not
A short, honest file beats a long one. Think of it as the twelve links you would send someone who asked what you do.
| Section | What good looks like | What to leave out |
|---|---|---|
Title and summary | Your business name and one sentence a stranger would understand. “Makura Creations — SEO, web development and AI visibility work, Lalitpur, Nepal, since 2013.” | Taglines. “You dream, we weave” is lovely branding and tells a model nothing. |
Core pages | Services, pricing, about, contact. The pages you would send a buyer. | Privacy policy, terms, cookie pages. Nobody is asking an assistant about those. |
Proof | Case studies with numbers in them, documentation, anything that shows the work. | Logo walls and testimonial sliders — no text for a model to read. |
Writing | The handful of articles you would still be proud of in two years. | Every blog post you have ever published. Volume dilutes the file. |
Descriptions | One line, specific, written in plain words. Mention the price, the place, the outcome. | Repeating the page title. The description is the only new information you add. |
Optional section | Mark genuinely secondary links under an “Optional” heading, as the spec allows. | Using it as a dumping ground so the file feels complete. |
Watch out
How people get this wrong
✕
Listing all 300 pages
A file with everything in it says nothing about what matters. If the list would not fit on one screen, it is too long.
✕
Putting it in a folder
/files/llms.txt or /wp-content/llms.txt does not count. It has to answer at the domain root, exactly like robots.txt.
✕
Serving it as HTML
If your site wraps it in the theme, it is not a text file any more. Load the URL and check you see raw markdown.
✕
Writing descriptions for Google
Keyword-stuffed lines read as spam to a model the same way they read as spam to a person. Write them for a human who is in a hurry.
✕
Generating once and forgetting
A file listing two services you dropped last year is worse than no file. Re-check it whenever your navigation changes.
✕
Expecting rankings to move
This is housekeeping, not a growth tactic. The visibility work that actually moves AI answers is content, entities and citations — the file just makes you easier to read.
FAQ
llms.txt questions, answered
A markdown file at the root of your domain that lists the pages on your site most worth reading, each with a short description, so a language model can find the useful parts without wading through your markup. The proposal and format are documented at llmstxt.org.
No AI company has publicly confirmed that it does. Adoption so far is on the publishing side — documentation sites, SaaS products and agencies adding the file — rather than on the consuming side. Treat it as cheap insurance, not a channel.
robots.txt is a set of permissions: it tells crawlers which paths they may not fetch. llms.txt grants and blocks nothing — it is a curated reading list with descriptions. You want both, and they do not overlap.
At the web root, next to robots.txt, so it answers at yourdomain.com/llms.txt. Upload it over FTP or your host’s file manager. Do not paste it into a page — WordPress would wrap it in your theme and it would stop being a text file.
An optional companion that contains the actual content of your key pages as markdown, rather than links to them. It suits documentation-heavy sites. For a typical business site, llms.txt on its own is enough.
No. Google has not said it reads the file, and nothing about it changes how your pages are crawled or rendered. If someone sells you llms.txt as an SEO tactic, ask them for the data.
Whenever your navigation or services change, and otherwise once a quarter. A stale file that lists services you no longer sell is worse than not having one.
No. It is a static text file that nothing else reads. It does not affect crawling, indexing or rendering. The realistic downside is that it sits there unread.
Yes — it is part of the AEO and AI visibility work, alongside schema, entity cleanup and tracking what the assistants actually say about you. The file is the easy part; being worth citing is the work.
Beyond the file
A tidy llms.txt will not make AI assistants recommend you
Being named in an AI answer comes from the same things that always mattered: pages that explain themselves, entities a model can connect, and other sites citing you. We do that work — and we will tell you honestly whether you are close or nowhere near.
Keep going
