There's a quiet convention spreading across the web that you've probably never heard of. It's called llms.txt, and it's designed to do for AI models what robots.txt did for search crawlers — give them clear, structured guidance about your site.
It matters because AI models are not good at reading websites.
The problem with how AI reads the web
When a model with web access visits your site, it encounters everything: navigation menus, cookie banners, footers, sidebars, JavaScript that may or may not render correctly. Extracting the information that actually matters — what you do, who you serve, what your key pages contain — is imprecise at best.
A model might summarise your homepage as a generic description of your category. It might attribute your services incorrectly. It might miss the thing that makes you different entirely, because that thing was buried in a JavaScript accordion below the fold.
llms.txt is a plain markdown file, placed at the root of your domain (yourdomain.com/llms.txt), that tells AI models exactly what they need to know. Your name, a clear description of what you do, and organised links to your key pages — structured in a way that's both human-readable and easy for a model to process.
How it works
The format is simple. An H1 heading with your site name. A blockquote with a short description. Sections with links to important pages and brief notes on what each contains.
Some implementations go further with an llms-full.txt file, which includes the actual content of key pages rather than just pointers to them. For tighter documentation sites or focused service businesses, this gives a model everything it needs in a single request.
Individual pages can also offer markdown versions — accessible by appending .md to the URL — which strip away layout and give models clean, parseable text.
Should you implement it?
For most sites: yes, and it won't take long. Several CMS platforms and static site generators now generate llms.txt automatically. If you're on a custom stack, it's a single file.
The caveat is worth stating: llms.txt is one signal, not a fix. A model's decision to cite you depends on many factors, and a well-formed llms.txt won't override weak third-party presence or a confusing description of what you do. It removes friction — it makes your content easier to use correctly — and that's worth doing.
For developer documentation, knowledge bases, and content-heavy sites, the case is stronger still. These are exactly the types of content AI models are asked to explain and recommend.
The spec is still evolving and is maintained at llmstxt.org. Worth half an hour of your time to implement.
Want to know how visible your site is to AI models right now — before and after changes like this? Run your free V-Score audit across ChatGPT, Claude, Perplexity, and Gemini.