I publish an llms.txt. The data says nobody reads it.
I publish an llms.txt file. It is 127 lines long and I maintain it carefully. I also think there is a good chance nothing has ever read it.
That sentence is going to annoy people selling llms.txt as a service, so let me put the evidence in front of you and then explain why I still keep mine.
What llms.txt is supposed to be
The idea, proposed in 2024, is a plain-text file at the root of your site — yoursite.com/llms.txt — written in Markdown, that tells an AI system what your site is, what it offers, and where the important pages are. A curated map for machines, by analogy with robots.txt and sitemap.xml.
It is a sensible idea. Web pages are cluttered with navigation, cookie banners and advertising that a model has to wade through. A clean summary written for the machine would save everyone effort. The proposal was well received, a lot of sites adopted it, and an industry of llms.txt audits grew up quickly.
What the evidence shows
Adoption by site owners is not the same as adoption by AI companies, and this is where it falls apart.
No major provider has adopted it. OpenAI's GPTBot honours robots.txt and does not use llms.txt. Google manages AI crawling through robots.txt with no mention of llms.txt support. Meta gives no indication of using it. Anthropic publishes an llms.txt of its own but has not said its crawlers read anyone else's.
The server logs are brutal. An Ahrefs study of 137,000 domains found that 97% of llms.txt files received no requests at all in May 2026. Separate monitoring across more than 500 million AI bot visits over a 90-day window found 408 requests that targeted llms.txt directly. Four hundred and eight, out of five hundred million.
Google has said so out loud. John Mueller: “AFAIK none of the AI services have said they're using LLMs.TXT (and you can tell when you look at your server logs that they don't even check for it).” He compared it to the keywords meta tag — a thing site owners filled in diligently for years while the engines quietly ignored it.
That comparison is the one to sit with. The keywords meta tag was not a scam. It was a reasonable proposal that never got picked up, kept alive by an industry that had already built services around it.
So why do I still publish one
Four reasons, in descending order of how much I believe them.
1. Writing it is a genuinely useful exercise. This is the real reason. To write a good llms.txt you have to state, in plain language, what your business does, what each page is for, and which pages matter. Most businesses cannot do this. The file is a forcing function, and the clarity you are forced into improves the pages themselves — which are read.
2. The cost is asymmetric. An hour to write, minutes to maintain, no risk. If adoption never comes you have lost an hour. If it comes in eighteen months you are already there. I will take that trade.
3. Some smaller tools do read it. Not the major assistants, but various AI developer tools and documentation systems do fetch it. That is a narrow benefit and I will not inflate it into a wide one.
4. For me specifically, it is a demonstration. I advise on this; running a careful one is part of showing the work. That reason applies to me and almost certainly not to you.
Notice what is missing from that list: “it will get you cited by ChatGPT.” I cannot tell you that, because the evidence does not support it, and because you would eventually check.
How to write one, if you want one
Given all of the above, keep it to an hour. Here is the format.
Plain text, Markdown, at yoursite.com/llms.txt. Served as text/plain. Start with an H1 of your business name, then a single blockquote line summarising what you do. Then H2 sections grouping your links, each link followed by a colon and a real description — not a slogan.
| Part | What it is | The mistake to avoid |
|---|---|---|
| H1 | Your business name, exactly as it appears everywhere else | Using a tagline instead of the name |
| Blockquote | One sentence: what you do, for whom, where | Marketing language with no information in it |
| H2 sections | Logical groupings — services, documentation, policies | Mirroring your nav instead of your substance |
| Links | Absolute URLs with a real description after each | A bare list of URLs with no explanation |
| Optional section | Material a machine can skip | Putting the important things here |
The single highest-value habit: write the description after each link as though explaining the page to someone who will never see it. “Pricing” is useless. “What an LLMO engagement costs, the three tiers, and what determines which one a business needs” is a description a machine can do something with.
And make it match your site. An llms.txt that describes services you no longer offer, or uses a different business name from your schema, is actively worse than none — you have manufactured a contradiction about yourself in a file you wrote on purpose.
What to do with the rest of the effort
If you were about to spend a week on llms.txt, spend an hour on it and the rest on things with evidence behind them.
Make your pages readable without JavaScript. AI crawlers largely do not execute it. Content that only exists after a script runs is invisible to them. This is measurable: fetch your own page with JavaScript disabled and read what remains.
Make your facts consistent everywhere. Same business name, address, phone and description across your site, your Business Profile, and every directory. Contradictions are the most common reason a model declines to state something about a business.
Get corroborated by sources you do not control. A model names a business when independent sources agree about it. Your own file, however well written, is you talking about yourself.
That last one is the whole game, and it is why a curated file was never going to be the answer. The problem was never that machines could not find your description of yourself. It is that your description of yourself is the least persuasive evidence available.
Want to know what AI assistants actually say about you? The free audit asks a live assistant real customer questions and reports which sources it used to answer.
Run the free auditCommon questions
Does llms.txt actually work?
There is no evidence that it does, and meaningful evidence that it does not. No major AI provider — OpenAI, Google, Meta — has stated that its crawlers use the file. An Ahrefs study of 137,000 domains found 97% of llms.txt files received no requests at all in May 2026, and monitoring across more than 500 million AI bot visits over 90 days logged only 408 direct requests for one. Google's John Mueller has said that none of the AI services have claimed to use it and that server logs show they do not check for it.
Should I create an llms.txt file anyway?
It is reasonable to spend about an hour on one, for three reasons that do not depend on adoption: writing it forces you to state plainly what your business does and what each page is for, which improves the pages themselves; the cost is small and the downside of being wrong is asymmetric if adoption arrives later; and some smaller AI developer and documentation tools do fetch it. It is not reasonable to spend a week on one, or to buy it as a service sold on the promise of AI citations.
What is the difference between llms.txt and robots.txt?
robots.txt is an established standard that crawlers genuinely honour; it controls which parts of a site a given user agent may access. llms.txt is a 2024 proposal for a curated Markdown summary of a site intended to help AI systems understand it. The practical difference is adoption: robots.txt is read and obeyed, llms.txt is almost never requested.
Can a badly written llms.txt hurt me?
Yes, in one specific way. If it states a business name, service list or description that contradicts your website, your schema or your Business Profile, you have created an inconsistency about yourself in a file you authored deliberately. Contradictory facts are a common reason a model declines to assert anything about a business. An out-of-date llms.txt is worse than none.
What should I do instead to be cited by AI?
Three things with better evidence behind them: make your pages readable without JavaScript, since AI crawlers largely do not execute it; keep your business facts identical across your site, your Business Profile and every directory that lists you; and build corroboration from sources you do not control, because models name businesses that independent sources agree about. A file describing yourself is the least persuasive evidence available about yourself.
Sources
- Ahrefs — What is llms.txt, and should you care about it? (137,000-domain study; provider adoption status; Mueller quote)
- llmstxt.org — the original proposal and format specification
About the author. Siamak Kalhor is an AI marketing and LLMO consultant in Los Angeles with more than forty years of experience in marketing strategy. He advises professionals and e-commerce leaders on becoming the answer AI assistants give. Get in touch, or connect on LinkedIn.