Google Lighthouse has expanded beyond traditional website performance analysis by adding a new category focused on agent-based browsing. While Lighthouse has long evaluated areas such as speed, accessibility, search optimization, and general web standards, this new section examines how effectively AI-driven tools can understand and interact with a webpage.

The agentic browsing category evaluates several aspects of a site, including the structure of its accessibility information, visual stability after loading, implementation of WebMCP features, and compliance with suggested llms.txt practices.

At the moment, this feature is still considered experimental. The industry is still figuring out the best ways to prepare websites for AI-powered visitors, so the recommendations may continue to change as these technologies develop.

A website does not need to add special AI functions to receive a positive result. A simple, well-structured site can still achieve a strong score without any advanced integrations.

Checking your agentic browsing results

You can review this category either through online website testing services or by running Lighthouse directly on your own machine.

For an online test, enter your website address into a compatible performance analysis tool and open the Lighthouse results section. Some popular browser-based testing environments may not yet include the newest Lighthouse version, but support is expected to expand over time.

If you prefer testing locally, install the newest Lighthouse package through npm and run an analysis from your terminal. This allows you to inspect the latest available audits directly from your computer.

Creating websites that work better with AI agents

As AI assistants become more common, websites need to be understandable not only to people but also to automated systems. AI agents may interpret pages through different representations, such as screenshots, raw HTML, or accessibility information.

A major point from recent guidance around agent-friendly websites is that clear structure and predictable behavior are becoming increasingly important. The same principles that improve accessibility and usability often help AI systems navigate pages more reliably.

What does the new category measure?

The new Lighthouse section combines existing website quality data with additional checks designed around AI interaction.

Poor accessibility structure

The accessibility tree provides a simplified view of a webpage’s content and organization. It is used by assistive technologies and can also help AI agents understand page elements without processing every part of the HTML.

The audit checks whether this structure is properly created and whether important information can be interpreted correctly.

WebMCP compatibility

WebMCP allows websites to expose specific actions that AI agents can perform directly within a browser session. It is designed for situations where an agent needs to interact with page elements rather than simply read information.

The related audit reviews whether interactive elements, such as forms, contain the correct descriptions and follow the expected structure. It can also identify available tools registered on a page through WebMCP methods.

llms.txt quality checks

The llms.txt concept is intended to give AI systems a clearer overview of website content. Although adoption is still limited, Lighthouse can check whether such a file exists and whether it follows basic formatting recommendations.

The review looks for common issues, such as missing headings, insufficient content, or a lack of useful references.

Unexpected layout movement

A stable page layout is important for both humans and AI systems. When elements move after loading, users may struggle to interact with the site, and automated agents that rely on screenshots or visual analysis may become confused.

This check is connected to the existing layout stability measurements used by Lighthouse. However, it is especially relevant for AI agents because they may act faster than human visitors and encounter movement during loading more frequently.

Overall, the new Lighthouse category shows a shift toward evaluating websites not only for human visitors but also for emerging AI-based browsing experiences.