Short answer
For most websites, llms.txt does nothing yet. Google says its Search systems, including AI Overviews and AI Mode, ignore the file. No major AI company has stated that its search crawler reads llms.txt across the public web. It has a real use for developer documentation read by coding agents. That is a narrow case.
llms.txt is a proposed Markdown file that points AI tools at a site’s most useful content. It sits at the root of a domain, like robots.txt.
The idea is reasonable. The question is whether anything reads it, and for search and citation the answer is currently no.
Who actually reads it
| Platform | Position |
|---|---|
| Google Search | Ignores it, including for AI features. Neither helps nor harms rankings |
| OpenAI | Publishes one for its own docs. No stated use in ChatGPT search |
| Anthropic | Publishes one for its own docs. Crawler documentation does not mention it |
| Perplexity | Publishes one for its own docs. No stated crawler use |
Notice the pattern. The AI companies publish llms.txt for their own developer documentation. None has said its search product reads yours.
What Google said
Google’s 15 May 2026 guide to generative AI search listed llms.txt among things site owners do not need.
Its later guidance went further, stating that Google Search ignores these files, so creating one will neither improve nor damage your visibility. That covers AI Overviews and AI Mode, not just classic results.
Google’s John Mueller has compared it to the keywords meta tag, and described it as a stopgap for AI coding tools reading developer documentation rather than a search signal.
The confusing signal
Chrome’s Lighthouse added an experimental llms.txt check shortly after Google’s search team said the file was unnecessary.
That looks contradictory. It is not. Lighthouse checking a file is not Google Search using it. Different teams, different purposes.
Fetching is not using either. Some AI crawlers have been seen requesting llms.txt files. A bot downloading a file tells you nothing about whether it influences what gets cited.
When it is genuinely worth adding
- You run developer documentation that coding agents and AI assistants read. This is the use it was built for.
- You have a large API or product reference where pointing tools at the right pages saves them effort.
- AI agents are a meaningful source of users for your product.
Outside these, skip it. A service business, blog or store gains nothing measurable from it today.
I have not added one to this site, for exactly that reason. It is a content and consulting site, not developer documentation.
What to do instead
The things that do affect AI visibility are less novel and more effective.
- Let the search crawlers in. robots.txt and CDN settings decide whether ChatGPT, Claude and Perplexity can cite you at all. Covered in how to get cited by ChatGPT, Perplexity and Claude.
- Rank for the question. Google’s AI features draw on its normal index.
- Answer directly and specifically in the HTML, not behind a script.
The same applies to schema. Google says no special markup is needed for AI features either, covered in schema markup and AI search.
Could this change?
Yes. The proposal is now at version 2 and the format is cheap to adopt.
If a major AI search product announces that it reads llms.txt from the public web, the calculation changes overnight. Until one does, it is a speculative bet, and the effort is better spent on things that are confirmed to matter.
If you build one anyway
Some sites have a genuine reason, usually documentation that coding tools read. Keep it simple.
- A title and a one paragraph summary of what the site is.
- A short list of your most useful pages, each with a one line description, written in Markdown.
- Links to existing pages, not a second copy of your content written for machines.
- An owner and a review date, because an out of date map is worse than none.
Do not create separate thin pages just for AI crawlers. Google’s John Mueller has described that kind of duplication as a poor use of time, and it adds maintenance with no documented benefit.
What would change this advice
One signal would change it: a major AI search product stating in its crawler documentation that it reads llms.txt from public websites. Until that appears, treat any claim that the file improves citations as unproven, however confidently it is made.
For most sites, the time is better spent elsewhere. Crawler access, clear answers and accurate content are all confirmed to matter, and none of them depend on a file that no search product has promised to read.
More claims like this one, each checked against what Google has actually said, are collected in Google SEO myths.
What to do this month instead
- Confirm AI search crawlers can reach your key pages, including through your CDN.
- Put a direct answer near the top of each important page.
- Check the facts on your most visited pages are current.
- Earn mentions on sites AI tools already trust in your field.
Each of these is confirmed to matter, which is more than can be said for the file.
Revisit the question once a year, or sooner if a major platform changes its documentation.
Chasing AI visibility tactics that may not work?
Most new AI tactics are unconfirmed, and the confirmed ones are unglamorous. Send your Search Console access and within 48 hours I will tell you which ones would actually move your site.
Get free recovery audit