Is llms.txt enough?

Key takeaways

• llms.txt is a proposed convention, not an adopted standard. No major engine has committed to reading it.
• Google has said explicitly that no new file type is needed to appear in AI features.
• Independent monitoring puts AI bot requests for llms.txt at roughly 0.1% of traffic. The file is mostly unread.
• It is cheap hygiene. Publish it, link it, move on. The cost of over-investing is the quarter you lose believing it worked.
• The thing that moves an AI answer is third-party corroboration, which is slower and harder and is why people reach for the file instead.

What the file is meant to do

llms.txt is a proposal: a plain-text file at the root of your domain that gives a language model a clean, structured summary of your site, in the same spirit as robots.txt or a sitemap. The idea is reasonable. A model that has to reconstruct what you do from a marketing homepage full of interstitials and animation will do a worse job than one handed a paragraph of plain facts.

The problem is not the idea. It is that a convention only works when the other side agrees to read it, and the engines have not agreed.

Why it does not do much yet

• No engine has committed to it. It is a community proposal, not a specification any major model provider has adopted.
• Google has said it is unnecessary. Google has stated that no special AI-specific file is required for a page to appear in its AI features. The existing crawl and index path is the path.
• The logs agree. Independent monitoring of AI crawler traffic has found requests for llms.txt at roughly 0.1% of AI bot hits. Almost nothing is reading it.
• It describes you, and engines discount what you say about yourself. This is the structural problem. Self-description is the weakest evidence class in an answer that names three companies.

The reason it is so appealing anyway

It is the only task on the list that is finishable. Everything else that moves an AI answer is slow, external, and partly outside your control: getting into roundups other people write, earning mentions in forums you do not run, correcting a stale description on a review site. Publishing a file is a Tuesday afternoon and a closed ticket.

That is a real pull, and it is worth naming rather than pretending nobody feels it. The risk is not that the file hurts. It is that shipping it feels like progress, and a quarter passes before anyone checks whether anything changed in an actual answer.

What to do instead, in order

If you have an afternoon, spend it here rather than on the file.

1. Find out what the engines currently say about you. Ask ten real buyer questions in ChatGPT, Claude, Perplexity and Google AI Mode. Record who gets named and what is said. This is free and it is the only way to know which problem you have.
2. Check the technical floor. Confirm AI crawlers are not blocked at the edge, that pages render server-side, and that status codes are honest. This fails silently and more often than people expect, and a 403 at your CDN overrides anything your robots.txt permits.
3. Fix what is wrong about you where engines already read. A stale price or an old positioning line on a review site or comparison page becomes the sentence a buyer sees. Correcting the source is higher leverage than publishing a rebuttal.
4. Get into the sources engines already trust. Editorial roundups and comparison lists in your category do a disproportionate amount of the work, because they are third-party and structured as exactly the answer the buyer asked for.
5. Then publish llms.txt. It takes an hour, it might matter later if adoption changes, and it costs nothing to have.

If you do publish one, do it properly

• Keep it factually current. An outdated llms.txt is worse than none, because it is a machine-readable source of wrong facts about you.
• Make sure every URL in it returns 200. Advertising dead links to a parser is a bad look for a file whose entire purpose is machine trust.
• Do not contradict your own site. If the file says one thing and your pricing page says another, you have created a conflict an engine may resolve against you.
• Do not put positioning spin in it. Plain facts read better to a parser and are less likely to be discarded.

The short version

Publish it. Keep it accurate. Then treat it as done and spend the real effort on the parts you cannot finish in an afternoon, because those are the parts that change what an engine says.

If you want to know which of the three problems you actually have, absence, inaccuracy, or a technical block, the free AEO report answers that specifically.

Frequently asked questions

Does llms.txt help with AI search?

There is no evidence it meaningfully does. Google has said no special file is needed for AI features, and AI crawlers request llms.txt on roughly 0.1% of requests. It is cheap enough to publish anyway, but it should not be the centre of a plan.

Do ChatGPT or Claude read llms.txt?

No engine has publicly committed to reading it. It is a community proposal rather than an adopted standard, and crawler logs suggest it is requested very rarely.

Is llms.txt the same as robots.txt?

No. robots.txt controls crawler access and is universally honoured by major crawlers. llms.txt is a proposed content summary with no comparable adoption. Getting robots.txt and your CDN rules right matters considerably more.

What actually gets you cited by AI engines?

Third-party corroboration first: roundups, comparison pages and forum threads that you do not control. Then retrievable page structure, entity clarity, and a working technical floor. Self-published description is the weakest signal in the set.

Should we remove our llms.txt file?

No, unless it is out of date or contradicts your site. A stale one is a machine-readable source of wrong facts, which is worse than not having one. Keep it accurate or delete it.