Lighthouse Agentic Browsing: How to rank in chatbots

Most teams already know Lighthouse Performance as a 0–100 lab grade behind PageSpeed Insights and CI. Chrome now documents an experimental Agentic Browsing category: how a page behaves when software agents discover and use the interface, not only when a human scrolls.

Scoring works differently. Instead of another weighted 0–100 blend, the category reports a fractional pass ratio across agentic readiness checks, plus pass/fail and warnings on audits such as WebMCP tool registration, accessibility-tree integrity for machine navigation, CLS (agents miss shifting targets), and llms.txt discoverability. A small movement in that ratio is coverage of checks, not a universal quality score you can paste next to last week's Performance number.

What we treat as actionable while the category is still experimental:

• Keep CLS budgets on interactive templates; late ads and embeds break agent coordinates the same way they break tap targets.
• Treat accessibility tree hygiene (names, roles, visibility) as a machine-navigation requirement, not only a human a11y pass.
• If you expose WebMCP tools, pin Lighthouse version and timing so imperative registration does not flicker between runs.
• Decide whether llms.txt belongs on the domain as a discoverability signal; do not confuse it with Core Web Vitals field data.
• Keep scheduled Performance and CrUX monitoring for user pain; use Agentic Browsing as a lab readiness layer beside that stack.

We still run Watcher schedules on the URLs that matter for retainers. This category is an extra lens when clients ask what chatbot-facing readiness looks like in Lighthouse, not a replacement for portfolio budgets and alerts.

Read more: Lighthouse Agentic Browsing: how to rank in chatbots