Google now ships a tool that audits for a file another part of Google says is unnecessary. Chrome’s Lighthouse 13.3 added an experimental Agentic Browsing category that checks whether a site provides an llms.txt file, the structured summary intended to help AI agents parse a site. Search Engine Land reported the change this week. It lands days after Google Search told site owners that llms.txt does nothing for AI search visibility.

The contradiction is real, and it traces to two teams optimizing for two different things. Lighthouse 13.3’s Agentic Browsing audits assess how well a site is built for machine interaction. The category includes four checks: WebMCP integration, accessibility-tree integrity, layout stability measured through Cumulative Layout Shift, and the presence of an llms.txt file. Google’s framing is that without llms.txt, an agent may spend extra time crawling a site just to understand its structure and primary content.

Google Search’s position is the opposite. Its recent guide on optimizing for generative AI features lists llms.txt in a section of things site owners do not need. John Mueller of Google has compared the file to the old keywords meta tag, noting that no AI search service actually consumes it. He has also explained that when Google publishes llms.txt files on its own developer properties, that is to help AI coding assistants parse documentation, not to influence Search.

Both statements are accurate because they describe different systems. llms.txt provides no ranking benefit inside Google Search’s generative answers. The same file may help a browser-based agent navigate a site more efficiently during an autonomous task. One claim is about search visibility. The other is about agent execution. The file sits in the gap between them, useful for one job and inert for the other.

That gap is the problem for a technical SEO team, because a Lighthouse audit reads as an instruction. A red or flagged item in Lighthouse has historically meant fix this. An item that is flagged but explicitly disconnected from search ranking breaks that convention. A team that treats every Lighthouse flag as a task will spend effort generating and maintaining an llms.txt file for a benefit that does not touch organic search at all.

The honest read is to treat the llms.txt audit as informational, not directive, for now. The audit is experimental, the file is optional, and Lighthouse marks a missing file as not applicable rather than as a failure. There is no ranking penalty for skipping it. The only argument for adding one is a deliberate bet that autonomous browser agents will become a meaningful traffic or task channel for the specific site, and that bet should be made on its own merits.

There is a more durable signal buried in the same release. Three of the four Agentic Browsing checks, accessibility-tree integrity, layout stability, and machine-readable structure, are not speculative. A clean accessibility tree and a stable layout help screen readers, conventional crawlers, and agents alike. Those improvements pay off across every consumer of a page, which makes them worth doing regardless of whether the agent era arrives on schedule.

For search teams, the move over the next quarter is to separate the two. Fund the accessibility and layout-stability work surfaced by the Agentic Browsing category, because it compounds. Treat the llms.txt line as optional, and revisit it only if agent traffic shows up in the logs.

ATTRIBUTION: Reported by Search Engine Land on 2026-05-20.