Google just shared more information to help us prepare our websites for agents. There is a new report from Lighthouse that anyone can run. You don’t need any external software to run it. You can do this directly from your Chrome browser.
The report tells you if your website is discoverable by AI agents, if WebMCP integration is configured, and something worth discussing: an assessment of your LLMs.txt file!
How to run the Web Agent report
Right now, if you try to get this report from the standard version of Chrome, you probably won’t find it. You can use it if you have it Chrome Canary. This is the next beta version of Chrome. Once you have that, just right-click and choose Inspect Page, then go to Lighthouse at the top. You will see a new category for “Agentic Navigation”.
Walk through the New Lighthouse Agent Readiness Report with me
I posted this on Google’s own page about the new Lighthouse report for agent navigationand it turns out that Google’s own documentation has some issues that might bother agents! The new report does not give a score out of 100, but rather a ratio indicating the number of agent readiness checks your site performs.
3 things to watch for in agent preparation
Here are the topics I discussed with my clients regarding this new shift.
1. AI Accessibility and Accessibility Tree
Agents can view your pages in three ways:
- Vision (screenshots).
- HTML.
- The accessibility tree.
The Accessibility Tree was originally intended for screen readers, but it actually tells an agent where the buttons are and what things are important. If your accessibility tree isn’t well-formed, agents will have difficulty using your site. I think making our pages agent-friendly will ultimately be a ranking factor in determining whether agents recommend your page.
2. Understand WebMCP
WebMCP is a web standard proposed to help you create and expose structured tools for AI agents. It’s basically a way to teach agents how to use your website’s features.
There are two types: declarative and imperative. Declarative is simple code that you wrap around a form, while imperative allows the agent to interact with your website. If you have tools on your site that people will use with their agents, WebMCP will be very important.
3. The LLMs.txt file
This seems crazy because Google just released documentation for ranking in search AI featuressaying you do not need an LLMs.txt file. But this report is not about research; it is about agents using your website. The proposal is to use a LLMs.txt file (similar to robots.txt) to provide markdown information that helps agents understand your site at inference time. It allows you to give specific instructions to agents about what they are allowed to do and where they can find important information. You probably don’t need an LLMs.txt file unless you have items that will be specifically used by agents.
LLMs.txt is intended for agents using your site and not for research purposes.
I highly recommend setting aside some time to review your own site in Chrome Canary. Most of us don’t need these files right now, but we need to be aware of them as our websites start to mature.
More resources:
Read Marie’s newsletter, AI News You Can Use. Subscribe now.
Featured image: Roman Samborskyi/Shutterstock
