Google Says Skip llms.txt. Google Also Audits It.
A reader sent me the Google AI-search guide with a one-line note: "So we DO need llms.txt now?" Reasonable read - because at the same time Chrome started flagging sites that don't have it. So I went and read both. The short version: no, and the long version is more useful.
What Google Search actually said
On May 15, 2026 Google published its first official guidance on appearing in generative search. The mythbusting section lists llms.txt first among tactics that don't help. The exact line: "you don't need to create new machine-readable files, AI text files, markup, or Markdown to appear in generative search." Gary Illyes has separately confirmed Google Search does not support the file and has no plans to.
Stay in the loop
Get news and updates about GEO, AI search and new features. Unsubscribe anytime.
So for getting cited in Google AI Overviews and AI Mode, llms.txt does nothing. Flat. That's not my opinion, it's their documentation.
Then why is Chrome auditing for it?
Because a different Google team has a different job. In early May, Chrome's Lighthouse shipped version 13.3 with a new "Agentic Browsing" audit category - and one of its checks is whether your site serves llms.txt and whether it returns cleanly. So:
| Google product | Says about llms.txt | The job it cares about |
|---|---|---|
| Google Search | Myth. Don't bother. | Getting your pages cited in AI answers |
| Chrome Lighthouse | Audits for it. Flags errors. | Letting AI agents navigate your site |
Two different problems wearing the same filename. Getting cited in a generated answer is a content-and-structure problem. Being navigable by an autonomous agent is a wayfinding problem.
"FWIW no AI system currently uses llms.txt."
llms.txt was only ever pitched at the second one - and even there it's a convention, not a ranking factor.
Who actually benefits from the file
Anthropic explicitly recommends llms.txt in its agent-writing guidance, and it earns its keep in a narrow set of cases:
- Developer-documentation sites and API references, where Claude or ChatGPT agents are a real, measurable referrer
- Large content properties where an agent needs a curated map of "the canonical pages, in priority order"
- Anywhere you've watched agent traffic in your logs and confirmed those agents fetch the file
That is not most e-commerce stores. If you sell running shoes, no agent is reading a Markdown index to decide how to navigate you - it's reading your Product JSON-LD.
3 points to follow
- For citations: skip it. Spend the afternoon on complete Product and FAQ schema instead. That's the lever Google's own guide points at, and it's the one our audits keep confirming. See our Product schema guide.
- For agent navigation: ship it only if your logs justify it. If you run docs or an API and see Claude-User or OAI-SearchBot in your access logs, a clean 5-minute llms.txt is cheap insurance. Keep it short - canonical URLs, one line each, no fluff.
- Don't let the Lighthouse flag scare you into the wrong fix. A failing Agentic Browsing audit is not a search problem. Know which job you're solving before you act on the warning, or you'll spend a week chasing a tactic Google Search already told you is a myth.
llms.txt is not dead and it's not magic. It's a navigation convention for agents that some properties need and most stores don't - and it has zero effect on whether AI search cites you. Two Google teams, two jobs, one confusing filename.
Curious whether AI engines can actually read and cite your store as-is? That's the question worth your afternoon. Run GEOlikeaPro's Visibility Vitals checker and start with the schema, not the Markdown.
FAQ
Does llms.txt help me get cited in Google AI Overviews?
No. Google's official AI-search guidance (May 15, 2026) lists llms.txt first among tactics that do not help generative-search visibility, and Gary Illyes confirmed Google Search does not support the file. For citations, complete Product and FAQ schema is the lever that works.
Why does Chrome Lighthouse audit for llms.txt if Google Search ignores it?
Because they solve different problems. Chrome's Lighthouse 13.3 added an Agentic Browsing audit category in early May 2026 that checks whether AI agents can navigate your site - that's a wayfinding job. Google Search cares about citing your content - a structure job. Same filename, two different teams, two different goals.
Who should actually create an llms.txt file?
Mainly developer-documentation sites, API references, and large content properties where AI agents like Claude-User or OAI-SearchBot are a measurable referrer in your access logs. Anthropic recommends it for agent-facing docs. Most e-commerce stores get nothing from it, since shopping agents read Product JSON-LD, not a Markdown index.
Should I worry about a failing Lighthouse 'Agentic Browsing' audit?
Only if agent navigation matters to your site. The audit is about whether autonomous agents can find their way around, not about search visibility. Don't let the warning push you into chasing a citation tactic Google Search has explicitly called a myth - identify which problem you're solving first.