09/04/2026
PageSpeed to Test Websites for AI Agents
Google’s Lighthouse testing system, which powers PageSpeed Insights audits, now includes an experimental category called Agentic Browsing. Instead of measuring how quickly a page loads, it checks whether AI agents can understand the page’s structure and reliably use its links, buttons, forms, and other tools. Chrome for Developers
It checks whether buttons, links, and form fields have clear machine-readable names. An AI agent needs those labels to understand what each part of the page does.
It checks whether the page unexpectedly moves while loading. If a button shifts after an agent finds it, the agent could select the wrong thing.
It looks for an llms.txt file, which can give AI systems a structured summary of the website’s important content.
It can test WebMCP, a proposed technology that lets websites expose actions, such as searching content or working with forms, directly to AI agents.
This is not a traditional 0–100 PageSpeed score. It reports how many checks passed.
Google says the category is still experimental, based on proposed standards, and subject to change. WebMCP testing currently requires Chrome 150 or later and an origin trial.
Why It Matters: Websites are beginning to serve two kinds of visitors: people who click through pages themselves and AI agents acting on their behalf. An agent might eventually help someone compare services, find store information, request a quote, or complete an appointment form.
Think of your website like a physical store. Clear signs, labeled aisles, and a checkout counter that stays in one place help customers. The same structure also helps an AI agent understand where to go and what it can safely do.
This does not mean Agentic Browsing is an SEO ranking factor or that every business needs WebMCP today. It’s an early signal of the shift ahead: accessible code, clearly labeled forms, stable pages, and organized information may determine whether future AI assistants can successfully use a business’s website.
