A Practical Way to Use Lighthouse to Test Your Website for Agentic Readiness

The way we interact with the web is shifting. For decades, we have optimized our sites for humans clicking links and for search engines indexing keywords. Now, we are entering an era where AI agents act on behalf of users. These agents do not just read content, they perform tasks. If your website is a black box to these agents, you are not just losing a ranking, you are losing the ability to be useful in an automated ecosystem.

Google has recently provided a way to see how your site stacks up in this new environment. Through a new report in Lighthouse, you can now check if your site is discoverable for AI agents and if you have implemented the necessary standards to let them operate. It is a quiet but significant shift in how we think about technical readiness.

Accessing the Agentic Web Report

You will not find this report in the standard version of Chrome that most of us use daily. Because this is still in the experimental phase, you need to use Chrome Canary, which is the beta version of the browser where Google tests upcoming features.

Once you have Canary installed, the process is straightforward. You simply navigate to the page you want to test, right click anywhere on the screen, and select Inspect Page. From there, you move to the Lighthouse tab at the top of the developer tools. In the category selection, you will see a new option called Agentic Browsing.

From a practical standpoint, the fact that this is tucked away in Canary tells us that Google is still refining the standards. The tradeoff here is stability versus insight. Canary can be buggier than the stable build, but it gives you a window into how Google intends for the web to function in the next few years. The decision for most site owners is whether the early warning is worth the slight instability of a beta browser. In my view, it is. Knowing where you stand before these checks become standard is a significant advantage.

Understanding the Agentic Readiness Results

When you run the report, the output is different from the traditional Lighthouse scores we are used to. You will not see a score from zero to one hundred. Instead, the report provides a ratio. It shows you exactly how many agentic readiness checks your site passed compared to the total number of checks performed.

It is worth noting that even Google is not perfect here. When running this report on Google's own documentation regarding agentic browsing, the results showed that their own pages have issues that could hinder AI agents. This is a helpful reminder that these standards are new and evolving. If the creators of the tool are still refining their own implementation, there is no reason to panic if your first report is not perfect.

The expert interpretation here is to look at the ratio as a roadmap rather than a grade. The goal is not a perfect score for the sake of a metric, but to identify specific friction points where an agent might get stuck. The decision you need to make is whether to fix everything at once or prioritize the elements that are critical to your business logic, such as checkout flows or contact forms.

Critical Factors for Agentic Readiness

There are three primary areas that the report focuses on. These are the same areas I have been discussing with people trying to navigate this shift. They represent a move away from visual presentation and toward structural utility.

AI Accessibility and the Accessibility Tree

Most of us think of accessibility in terms of screen readers for visually impaired users. However, the accessibility tree is now a primary way for AI agents to understand a page. While a human sees a styled button, an agent looks at the accessibility tree to determine where the buttons are and which elements are actually important.

If your accessibility tree is poorly formed, an agent will struggle to navigate your site. It is essentially the map the agent uses to interact with your interface. I believe that as agents become the primary way people access the web, being agent friendly will eventually become a ranking factor. If an agent cannot easily use your site to complete a task, it simply will not recommend your page to the user.

The tradeoff here is between aesthetic complexity and semantic clarity. Many modern design trends rely on complex JavaScript components that look great but break the accessibility tree. The decision for developers is to move back toward semantic HTML. You must decide if a custom styled div that looks like a button is worth the risk of an AI agent ignoring it entirely.

The Role of WebMCP

WebMCP is a proposed web standard designed to help site owners expose structured tools to AI agents. Think of it as a way to teach an agent how to use the specific functionality of your website without the agent having to guess based on the visual layout.

There are two main types of WebMCP implementations. The first is declarative, which involves wrapping simple code around a form to tell the agent what the form does. The second is imperative, which is more complex and allows the agent to engage in a back and forth interaction with your website to achieve a goal.

If your website provides tools, calculators, or booking systems that you expect agents to use, WebMCP is going to be essential. The expert interpretation is that WebMCP turns your website into a set of APIs that agents can call. The tradeoff is the effort required to implement these standards versus the potential for increased conversion from agent driven traffic. You should inspect which specific tools on your site are most valuable and start there.

Implementing the LLMs.txt File

There is some conflicting information regarding the LLMs.txt file. Google has stated in their search ranking documentation that you do not need an LLMs.txt file to rank in AI search features. However, it is important to distinguish between Search and Agentic use. Search is about being found, while agents are about being used.

The proposal for LLMs.txt is similar to the robots.txt file. It provides markdown information that helps agents understand your site at inference time. This file allows you to give specific instructions to agents, telling them what they are allowed to do and where the most important information is located.

For the average blog or informational site, an LLMs.txt file is likely unnecessary. However, if you have complex elements specifically designed for agent interaction, this file becomes a critical control layer. It is a way to provide a manual for the AI.

The tradeoff here is transparency versus control. By providing an LLMs.txt file, you are guiding the AI, but you are also revealing the internal logic of how you want your site to be used. The decision to implement this should be based on whether your site has a high degree of functional complexity that an agent might misinterpret without a guide.

Checking your site in Chrome Canary now is a smart move. While most of us do not need to overhaul our entire technical stack today, the transition to an agentic web is happening. Being aware of these tools now ensures that when the shift accelerates, your website is already speaking the language of the machines.

Practical next steps

The useful part is not only the idea itself, but the operating habit behind it. Use it as a checklist for decisions: what deserves attention now, what should be monitored, what needs a stronger evidence base, and what can wait until the system has more scale.