Google says skip llms.txt for search, Lighthouse audits it for agents

Google has split llms.txt into two different conversations, and SEO agencies now have to treat it that way. In Search, Google says there are no additional requirements or special optimizations needed to appear in AI Overviews or AI Mode. In Chrome’s Lighthouse, the same file now shows up in an experimental Agentic Browsing audit, where it can be checked as part of a site’s readiness for browser agents.

That contradiction matters because the products are solving different problems. Google Search’s AI features use a query fan-out technique across related searches and data sources, which means visibility depends on the same fundamentals that have always mattered: crawlability, content structure, and clear pages that answer real queries. Google Search Central’s guidance is explicit that llms.txt is not a shortcut into those surfaces. Chrome’s Lighthouse, by contrast, treats llms.txt as an emerging convention, a machine-readable summary meant to help LLMs and AI agents understand a website faster. If the file is missing and the server returns a 404, the audit is marked N/A, which makes the file optional rather than mandatory.

The technical split became sharper in Lighthouse 13.2.0, which added the agentic browsing category to the default config on April 30, 2026. Chrome’s documentation says that category, and its WebMCP support, are experimental and based on proposed standards. The llms.txt check was part of that same release train, and Lighthouse’s scoring for this area does not behave like a traditional weighted 0-to-100 grade. It returns specific audit results and a pass ratio, which makes it more of a readiness signal than a performance score.

For agencies, the safest decision is to separate search work from agent-readiness work. llms.txt should not take budget away from technical SEO, internal linking, entity clarity, or improving the pages that humans and crawlers actually rely on. It can make sense on large content systems, developer documentation sites, and products that expect browser agents to parse site structure quickly, but it is not a visibility lever and not an AI citation trick. John Mueller has already framed the file skeptically, comparing it in April 2025 to the old keywords meta tag, and Google Search Central Deep Dive APAC coverage later said Gary Illyes and Amir Taboul were not pursuing llms.txt.

The clean client message is simple: Google Search is not asking for llms.txt to rank or appear in AI features, but Chrome is experimenting with it as a signal for agents. That makes the file a documentation choice, not an SEO silver bullet, and the real work remains the same as ever: build pages that are useful, well-structured, and easy for both people and machines to understand.