llms.txt and Content-Signal: the new files for AI visibility

Why AI sees your site differently

The Google bot crawls your pages over time and builds an index it can search later. AI answer systems work differently: they fetch individual pages at runtime, read the text and summarise it on the spot. Nested menus, a lot of JavaScript and unstructured pages make that hard for them. The clearer and flatter your content is, the more likely it gets understood correctly and quoted. Two small files help you send that signal, and neither needs a single line of programming.

The two files at a glance

Different jobs, both tiny. Here is what each one does.

Neither file moves you up the rankings on its own. What they do is remove friction: they hand an AI system the map and the permission slip, so the good content you already have is easy to find and clearly allowed to be used. On a young standard that most sites still ignore, that small head start is cheap to take.

What llms.txt is

llms.txt is a plain text file in the root of your domain (so your-domain.com/llms.txt). It works like a short table of contents for AI systems: who you are, what you offer and which pages are the important ones. It is written in Markdown, easy to read, with no ballast. A minimal version looks like this:

# Sample Roofing Ltd
> Master roofing business in Lüneburg and the surrounding area.

## Services
- [Roof renovation](/services/roof-renovation)
- [Maintenance and repair](/services/maintenance)

## Contact
- [Address, phone, opening hours](/contact)

That way an AI system finds the core information without working through your whole site. Two details matter: start with a single H1 (the # line), and use real Markdown links in the format [label](url), not bare URLs. That is exactly what current validators check for, and it is the difference between a file that merely exists and one that actually gets read.

Content-Signal in robots.txt

You may already know robots.txt, which controls what search engines are allowed to crawl. The Content-Signal directive is newer: with it you tell AI systems how your content may be used. One line is enough.

Content-Signal: ai-train=yes, search=yes, ai-retrieval=yes

This explicitly allows your content for AI training, AI search and AI answer retrieval, which is what you want if you would like to show up in ChatGPT and Perplexity. Set any value to no where you disagree. Just as important: do not block the AI crawlers by accident, for example through a Cloudflare rule, because then even the best llms.txt cannot help.

How to check it actually worked

A file that exists is not the same as a file that passes. Lighthouse now has an “AI agent accessibility” check that flags an llms.txt which is not valid Markdown. The three things it wants: exactly one H1 heading, at least one real Markdown link, and no invisible byte-order mark in front of the first character. It also helps to serve the file as text/markdown rather than plain text. If your CMS generates the file automatically, and some SEO plugins now do, open the live file and check these points, because auto-generated versions often add a byte-order mark or dump raw HTML that quietly breaks the links. A quick manual check: open your-domain.com/llms.txt in the browser, confirm the first visible character is the # of your H1, and that the page links look like [label](url) rather than raw paths.

The order of importance

So the expectation is right: these files are the finishing touch, not the foundation. Work on AI visibility pays off in this order.

1. Citable, clearly structured content with concrete answers, numbers and prices.
2. Clean schema (Organization, LocalBusiness, Person) and consistent contact details.
3. No accidental blocking of AI crawlers.
4. Then llms.txt and Content-Signal as an additional, clean signal.