Does llms.txt Matter? What the Evidence Says in 2026
Short answer: as of November 2026, I have found no evidence that adding an llms.txt file makes major AI engines cite you more, and the strongest statements from Google say it is not needed. It costs about an hour, so it is not harmful to add. It is just a poor use of that hour compared with the things that do move citations. Please verify the current status yourself before relying on this, because it could change.
- Google has said no AI system uses llms.txt and that its AI features need no special files.
- No major AI company we checked documents llms.txt as a ranking or retrieval input.
- Claims that “ChatGPT, Claude and Perplexity all read it” mostly trace to SEO blogs, not to the companies.
- Where it plausibly helps: developer documentation consumed by coding agents.
- If you add one, treat it as hygiene, not strategy.
What llms.txt is
llms.txt is a proposed convention: a markdown file at the root of your site (yourdomain.com/llms.txt) that gives a language model a short summary of the site and a curated list of links to the most useful pages, ideally in a clean format. The idea is that a model with limited context can read one small file instead of crawling a cluttered site.
It looks like a sitemap for models, though a sitemap lists everything and llms.txt is meant to be curated. A related idea is an extended file with the full text of key pages. The proposal came from the community, not from a standards body, and no search engine or AI platform is obliged to honor it. That last point is the whole story.
What the platforms have said
I checked the primary sources I could reach, and here is what they say, with the date I checked: November 2026.
Google. In a June 2025 post on Bluesky, John Mueller wrote that no AI system currently uses llms.txt, noting that server logs show consumer chatbots fetching pages for training and grounding but not requesting that file. Google’s own documentation on AI features in Search, which I read directly for this post, says there are no additional technical requirements for appearing in AI Overviews and AI Mode beyond being indexed and eligible to show a snippet, and that you do not need to create new machine-readable files, AI text files or markup. The page does not name llms.txt, but “AI text files” is hard to read any other way. Several SEO publications also report that Google staff said at a July 2025 event that Google does not support llms.txt. I could not verify that one from a primary source, so I am flagging it as reported.
OpenAI. OpenAI’s documentation for its crawlers describes three: OAI-SearchBot for surfacing sites in ChatGPT search, GPTBot for content that may be used for training, and ChatGPT-User for actions a user triggers. The documentation discusses robots.txt controls for them. I did not find llms.txt mentioned. Absence from documentation is not proof that nothing reads the file, but it means there is no documented lever.
Perplexity. Perplexity’s bot documentation describes PerplexityBot and Perplexity-User and robots.txt guidance. It does not mention llms.txt either.
Anthropic. I did not find Anthropic’s own crawler documentation in my searches, so I will not characterize its position. Third-party guides discuss Anthropic’s bots, but that is secondary.
If a statement turns up from one of these companies after this post’s date, it overrides my summary. Check before you decide.
Where the claims come from
You will see articles saying some or all of the major engines read llms.txt. When I traced them, the claims fell into three groups:
- Observations that a bot requested the file. Some site owners report seeing a request in their logs. A request shows a crawler fetched it. It does not show the contents changed any answer.
- Platforms that publish their own llms.txt. Some AI and developer companies host one on their documentation sites. That tells you they like the convention, not that their crawlers use yours.
- Plain assertion. Vendor blogs that sell llms.txt generators or audits often state it as fact with no source.
I have not seen a controlled test showing that adding the file raised citation rates. If you know of a credible one, I would like to read it. Until then, “unproven” is the accurate label, which is different from “useless.”
Why it probably does not move citations
Think about how a citation happens. For a recommendation prompt, an engine searches, picks pages that match the question and look credible, and writes an answer. Several things in that chain matter, and a curated file at your root addresses none of them:
- Can the crawler access the page? That is robots.txt and your firewall.
- Does the page match the question? That is your content.
- Is the source trusted for this kind of question? That depends heavily on independent mentions.
- Is there corroboration elsewhere? Third-party coverage and discussion.
In ReddGrow’s studies of what AI engines cite on Reddit, LinkedIn and YouTube, the strongest consistent signal was how closely the content matched the prompt that surfaced it. For Reddit threads, prompt words in the title and the opening of the post separated most-cited from once-cited threads better than popularity did. Nothing in those studies involved a machine-readable file, because there was none to compare. The point is that the levers we can measure are about the content and the source, not about a manifest.
Where llms.txt can make sense
I am not telling you to avoid it. A few cases where it is reasonable:
Developer documentation. If you ship an API or SDK, coding assistants and agents sometimes fetch docs on demand. A clean index of your docs, or a flat text version, can make that easier for tools that choose to use it. Several documentation platforms can generate the file for you. The gain is convenience for agents, not a ranking boost.
Sites with messy navigation. If your key pages are buried behind scripts and menus, a curated list is a reasonable fallback, though fixing the site is the better answer.
Low cost, low risk. If someone on your team can generate it in an hour and keep it current, there is little downside.
What to avoid is treating it as an AI SEO strategy, reporting it to leadership as progress, or paying a vendor a meaningful fee for it.
What to do with the hour instead
If you have an hour that would have gone to llms.txt, these have a better evidence base:
- Check crawler access. Confirm OAI-SearchBot and PerplexityBot are not blocked in robots.txt or by your firewall, if you want to appear in those engines’ search results. OpenAI’s documentation says opted-out sites will not appear in ChatGPT search answers.
- Fix the facts. One clear description of what you do, a real pricing page and honest comparison pages. See how to get your brand mentioned in ChatGPT answers.
- Run a baseline. Write 30 prompts a buyer would ask and see who gets cited. That gives you a target, as shown in how to track brand mentions in AI search.
- Go where the engines already look. Look at the sources cited for your category and earn accurate presence there, including the Reddit threads that match your prompts. Our Reddit AEO guide explains the approach.
- Run through a checklist. The GEO checklist covers 25 items, none of which need llms.txt.
How to verify the status yourself
Because this area changes, here is the quick way to re-check before you act on anything, including this post:
- Read Google’s AI features documentation in Search Central and look for any mention of AI text files.
- Read each AI company’s crawler documentation: OpenAI’s bots page, Perplexity’s bots guide and Anthropic’s equivalent.
- Check your own server logs for requests to /llms.txt and note which user agents make them.
- Search for the engine’s name plus “llms.txt” and prefer statements from the company, not from vendors.
- If you add the file, run your baseline prompts before and after and see whether anything moves. With answer noise, you will need many prompts and repeats to see a real effect.
My position
If a client asked me whether to spend a day on llms.txt, I would say no, and then say yes if the cost is under an hour and they publish developer docs. The evidence does not support a benefit, the platform that matters most for search says it is not needed, and the work that does show effects is unglamorous: access, accurate facts, matching content and independent mentions. If that changes, I will update this post.
Frequently Asked Questions
- Does Google use llms.txt?
- Google has said it does not. John Mueller wrote in June 2025 that no AI system currently uses llms.txt, and Google's own AI features documentation says you do not need new machine-readable files, AI text files or markup to appear in AI Overviews or AI Mode. Check Google's documentation for any later change.
- Does ChatGPT read llms.txt?
- OpenAI's published crawler documentation covers OAI-SearchBot, GPTBot and ChatGPT-User and discusses robots.txt, and we did not find it mentioning llms.txt. Some third-party blogs claim ChatGPT reads the file, but we found no primary source confirming it. Treat that claim as unproven.
- Can llms.txt hurt my site?
- It is very unlikely to cause harm. It is a plain text file. The risks are wasted time and a false sense of progress, and publishing a stale file that points to outdated pages. If you add one, keep it accurate or remove it.
- Is llms.txt useful for developer documentation?
- That is the strongest use case. Developer tools and coding assistants that fetch documentation can benefit from a clean, flat list of pages, and documentation platforms have added support for generating it. The benefit there is convenience for agents that choose to use it, not a ranking effect.
- What should I do instead of llms.txt?
- Make sure search crawlers can access your pages, state clearly what you do and what you cost, and earn accurate mentions on third-party sources that engines already cite for your category. Those have documented effects on AI answers, unlike the file.