llms.txt: What It Is, Do You Need It?

If you've read somewhere that your site "must have" an llms.txt file, the right question isn't how to write it. It's whether you need it. This page answers both, and the answer to the second one is uncomfortable.

llms.txt is a proposed standard: a text file placed at the root of your site, like robots.txt, that tells AI models which content matters and where to find it. The spec lives at llmstxt.org and has only been open for a short while.

What it is, and why it's not robots.txt

Robots.txt and llms.txt answer two different questions, and mixing them up is the most common mistake. Robots.txt says what can be read: it's a permission, and whoever ignores it is breaking a convention. llms.txt says what's worth reading: it's a suggestion, and no one is required to follow it.

The format is different as a result. Robots.txt is a list of rules for crawlers. llms.txt is written in Markdown: the business name, a one-line description, and a list of links to the pages that matter, each with a sentence explaining what's there. It's meant to be read by a machine that has little space and needs to understand quickly.

68 of 88 Italian sites we measured don't have it

The number comes from our own measurements, not an estimate: we used our engine to analyze 88 Italian sites across five categories, including law firms, restaurants, physiotherapists, real estate agencies, and beauty centers. Of these, 68 don't expose any llms.txt.

Of the remaining twenty, we can't say for certain that they have one: the check may not have run, for instance if the site responded poorly at that moment. So the honest figure is this: at least 77% of the sites we measured don't have llms.txt. The real share is likely higher.

What it promises, and what no one has proven yet

Here's the thing almost no article on this topic says. There's no official documentation from OpenAI, Google, or Anthropic confirming they read llms.txt, and no public studies measuring its effect on citations. It's a community proposal, not a standard adopted by search or AI engines.

This doesn't mean it's useless. It means anyone promising you a result is promising something they can't actually know. The plausible benefit is orientation: if a model reads the file, it finds a clean map of the site instead of having to work it out from the HTML. The proven benefit, today, is zero.

How to write one, in practice

A minimal llms.txt has a title with the business name, a paragraph saying what you do and where, and a section of links to the main pages, each with a one-line explanation. The links must be real, reachable addresses: a file pointing to pages that don't exist is worse than no file at all.

Save it as llms.txt and place it at the root, meaning at the address yoursite.com/llms.txt. There's no need to declare it anywhere; like robots.txt, it's found because it's there.

A real example, ours

Guides on this topic almost always show a made-up example. This is the actual file we serve at trovavio.com/llms.txt, shortened only in the final lines:

# Trovavio

> SEO and GEO audits for professionals and Italian agencies: analyze your site
> in 30 seconds and find out if it's visible on Google and AI engines like ChatGPT.

Trovavio is an online analysis tool built for professionals and agencies
in Italy. In a few seconds it generates a complete audit of your website,
explained in plain Italian and tailored to your industry.

## Main pages

- [Trovavio, SEO and GEO Audit](https://trovavio.com/): the analysis tool,
  pricing, and frequently asked questions
- [Blog: SEO and GEO guides](https://trovavio.com/blog): how to get found on
  Google and get cited by AI assistants

Three things worth noting. The title is the business name, not a slogan. The line starting with the greater-than sign is the one-sentence description, and it's the part a model reads first. Each link has an explanation after the colon, because a bare list of addresses doesn't tell anyone which one to open.

What not to put in it

It's not a place for keywords. Filling it with repeated phrases hoping a model will weigh them is the 2026 version of white text on a white background: it's never been shown to work, and it's exactly the kind of thing engines learn to ignore first.

And don't put in pages you don't want read. A file listing the members area or a draft page is flagging it, not hiding it: that's what robots.txt is for, which is the other file doing the other job.

Is it worth doing? The honest answer

It's worth it if you treat it for what it is: half an hour of work, once, on a low-cost bet. It's not worth it if someone sells it to you as the reason you don't show up on ChatGPT, because that reason, in the sites we've measured, is almost always something else: content that doesn't answer any real question, no structured data, or AI bots blocked by robots.txt without the owner knowing.

Put another way: llms.txt is the last of the steps, not the first. If your site has bigger problems, that file won't make up for them.