Is llms.txt worth publishing? An honest answer
What the proposal is, who actually reads it today, what it costs you to publish one, and the work that matters more.
llms.txt is a proposed convention: a markdown file at the root of your site that tells language models what your site is and which pages matter.
The idea is reasonable. The honest status, as of September 2026, is that site owners have adopted it faster than the models have.
What it is
A file at yourdomain.com/llms.txt, in markdown, structured roughly like this:
# Your Company
> One sentence on what you do and who for.
## Core pages
- [Pricing](https://example.com/pricing): Plans, prices and what is in each.
- [How it works](https://example.com/how): The process, step by step.
## Guides
- [Guide title](https://example.com/guides/x): One line on what it covers.
The premise is that a model fetching this gets a clean, curated summary instead of having to infer structure from navigation and markup.
Some sites also publish llms-full.txt, containing the actual content of key pages in markdown, so a model can read the substance in one fetch.
Who reads it
This is the part usually glossed over.
No major AI assistant has committed to treating llms.txt as authoritative. There is no public documentation from the big vendors saying they fetch it and weight it. Some developer tools and smaller products do use it.
So publishing one today is a bet on future adoption, not an action with a measurable effect this quarter. Anyone telling you otherwise is guessing, and you should ask them for their evidence.
Should you publish one
Yes, if: you have an hour, you have already done the work below that does move answers, and you like being early to conventions that might become standard.
Not yet, if: you are choosing between this and getting listed in the five directories that your category's AI answers actually cite. That is not a close call.
The risk is not the file. It is the substitution: publishing one, feeling productive, and skipping the unglamorous work that changes results.
If you do publish one
Four rules to make it worth having.
Keep it curated. Twenty to forty links, not your whole sitemap. The point is editorial judgement about what matters. A dump of every URL is a sitemap with extra steps.
Write real descriptions. One clear line per link, describing what is on the page, not marketing copy about how great it is.
Include the facts. What you do, who for, where you operate, what it costs. If a model reads one file, that file should answer the obvious questions.
Keep it current. A stale llms.txt is worse than none, because it points at pages that have moved. If you will not maintain it, do not publish it.
What matters more, in order
If you have limited time, spend it here instead.
1. Check you are not blocking the crawlers. Read your robots.txt and your firewall settings. A blocked site is invisible regardless of what files it publishes. This is a three minute check that fixes more visibility problems than anything on this page.
2. Get into the sources the assistants read. Ask your buyer questions, read the citations, and go and get listed on the pages that recur. This is the highest return work available and it is the one people keep looking past.
3. Put your facts in plain text on your own pages. What you do, for whom, where, what it costs. Models summarise text. They cannot summarise a hero image or an implication.
4. Be consistent everywhere. Same description across your site, your Google listing, your LinkedIn and every directory.
5. Then llms.txt, if there is time left.
Measure, do not assume
The reason to be sceptical of llms.txt specifically is the reason to be sceptical of everything in this area right now: it is easy to do something that feels technical and never check whether it changed anything.
Take a baseline first. Twenty buyer questions across ChatGPT, Claude, Gemini and Perplexity, recording who gets named. Then make changes. Then re measure on the same questions.
If your answer share moves after publishing llms.txt and nothing else changed, that is evidence and we would genuinely like to see it. If it does not, you have lost an hour and learned something, which is a fair trade.
Questions people ask
Do AI assistants read llms.txt?
No major assistant treats it as authoritative today. Adoption among site owners has grown; adoption among the models has not kept pace.
Is it worth publishing anyway?
It costs an hour and it might matter later. Publish one if you have the hour spare, after the work that demonstrably moves answers today.
What goes in it?
A markdown file at your root with a short description of what your site is, and a curated list of your most important pages with one line each.
Will it hurt anything?
No. It is an extra file that nothing depends on. The only real cost is the time, and the risk of believing it did something it did not.
See it on your own competitors
Figo checks their ads, pages, rankings, reviews and AI answers every week, then tells you what to do in plain words. Set up in two minutes.
Keep reading