llms.txt generator and checker: so AI search describes you correctly

What the file is and how we wrote ours

What goes into an llms.txt?

The format was proposed by Jeremy Howard in September 2024 and is described at llmstxt.org (version 2, August 2026). In order:

1. An H1 with the name. The only required part.
2. A blockquote with a short summary.
3. Optional text without headings: paragraphs or lists with what a model needs to read the rest correctly.
4. ## sections with lists of links: - [name](url): notes. A section called Optional is, by convention, for links an agent may skip when it is short on context.

The file sits at the root of the site, /llms.txt. Version 2 also suggests a Markdown copy of each important page (page.html.md or page.md) and a rel="describedby" link pointing to the file. Those are extras; most small companies get the main benefit from the file alone.

What the checker looks for

• Structure: H1 on the first line, only one H1, a summary right after it, no ### headings, at least one ## section with links, no empty sections, Optional last.
• Dead link forms: empty links, # and javascript:, relative paths without a domain, links to localhost, test or staging servers, example.com left from a template, spaces in addresses, unclosed brackets, bare URLs without [name](url).
• Leftovers: unfilled template tokens and HTML tags. Both happen when the file is generated by a build or a CMS plugin and nobody reads the result.
• Length: characters, words and an approximate token count. Over 50,000 characters we suggest moving detail into a second file and linking to it.

What it cannot do: open the links. Whether a page returns 404 you check with your browser or a crawler. That way nothing leaves your machine.

Our own llms.txt, with notes

This is the start of our file, shortened in a few places. It is in English because models everywhere read it; the Slovenian pages are linked on the same line:

# MediaAtlas

> MediaAtlas d.o.o. (Sevnica, Slovenia, founded 2012) is an AI research and development studio. It trains language models on a company's own data and deploys them inside the company's infrastructure, weights included. […] Full technical detail: https://mediaatlas.si/llms-full.txt . […]

## Flagship service: LLM fine-tuning and AI model training

- [Pricing (EN)](https://mediaatlas.si/ai-training-pricing.html) / [Cenik (SL)](https://mediaatlas.si/sl/ai-training-pricing.html): public price list. One-day evaluation on the customer's own documents EUR 1,900 (credited against a pilot); […] EUR, excluding VAT.
[…]
- [What is MediaAtlas?](https://mediaatlas.si/#what-is-mediaatlas) / [SL](https://mediaatlas.si/sl/#kaj-je-mediaatlas): […] not connected to companies called Atlas Media; motion capture is earlier work. […]

Why it is written this way:

• The summary answers who, what, where and since when. A small company is easy to mix up with someone else. Our name is close to companies called Atlas Media, so the file says plainly that we are not connected to them, and that motion capture is earlier work, not what we do today.
• Prices are written out, with currency and VAT. "Contact us for pricing" leaves a model to guess, and it will. The numbers match our price list; check yours against the page every time a price changes.
• Every link is a full https address to a real HTML page. English and Slovenian versions share one line, so a model sees they are the same page.
• Detail goes into a second file. The summary links to llms-full.txt with the technical background. The main file stays something an agent can read in one go.

What llms.txt does not do

• It is not a guarantee of being cited. No AI search engine has promised to quote what the file says.
• Google says it is not needed. Its documentation for AI features in Search states: "You don't need to create new machine readable files, AI text files, or markup to appear in these features." (Google Search Central). What counts there is an indexable page with clear text and structured data that matches what is visible.
• Lighthouse checks only that the file does not break. Chrome's Lighthouse has an llms.txt audit, but it fails only on a server error; a missing file is marked N/A, since the file is optional for now (Chrome for Developers).
• It does not fix wrong pages. If the price list says one thing and the file another, you have made the confusion worse. The file summarises your pages; it does not replace them.

Where it does help: agents and coding assistants that look for the file, and anyone building on your documentation. According to llmstxt.org the AI labs publish llms.txt files for their own developer documentation. And the exercise itself: a company that cannot fill in the summary field in two sentences will not be described well by anyone.

When you do not need us

For llms.txt, almost never. The form above and an hour are enough, and the file costs nothing. Rewrite it whenever your services or prices change, and check it again here.

It gets harder when a site has dozens of pages, prices in several places and two languages, and every change has to land everywhere at once. That is the kind of work our content agent does on our own sites; how it works is on the AI agents page. Building one for your site is an agent pilot, from €6,900 (4–6 weeks), EUR excluding VAT.