What Lighthouse Agentic Readiness Is and Why It Matters
Lighthouse agentic readiness is a new category in Google’s Lighthouse tooling that checks whether your website’s structure, tools, and metadata are prepared for AI agents to discover, understand, and use it effectively. Instead of measuring only human-focused web performance, it tests how well automated agents can interpret your HTML, accessibility data, and machine-readable instructions so they can carry out tasks on behalf of users. This matters because agents rely on predictable structures and clear rules to fill forms, trigger actions, and fetch accurate information. As AI agent usage grows, websites that support these capabilities will be better positioned to appear in agent-driven recommendations and automations. Google’s update brings AI agent website testing directly into Lighthouse, making it a standard part of modern web development instead of an experimental add‑on or separate tool.
How to Run the Lighthouse Agentic Readiness Report in Chrome Canary
You currently need Chrome Canary to use the Lighthouse agentic readiness checks, because the feature is not yet available in the standard Chrome build. After installing Chrome Canary, open your website, then right‑click anywhere on the page and select “Inspect” to open DevTools. At the top of DevTools, select the Lighthouse tab. In the list of Lighthouse categories, you will see a new option called “Agentic Browsing.” Enable this category, choose the device type you want to test (mobile or desktop), and click “Analyze page load.” Lighthouse will generate a dedicated report focused on agentic readiness, separate from performance, accessibility, and SEO. According to Marie Haynes, this built‑in report means “you do not need external software to run it” because everything runs from within your Chrome browser, keeping AI agent website testing close to your normal workflow.
Reading the Report: Ratios, Checks, and Common Issues
Unlike traditional Lighthouse audits, the Chrome Canary Lighthouse agentic readiness report does not return a score out of 100. Instead, it shows a ratio of passed checks to total checks so you can see at a glance how many agentic readiness requirements your site meets. Each item includes a short explanation and links to more documentation where available. When Marie Haynes ran the report on Google’s own documentation about agentic browsing, the page still showed issues that could hinder agents, which underlines how new these standards are for everyone. Focus first on failing checks that affect discoverability and interaction, such as missing or confusing markup for key actions, incomplete accessibility structures, or lack of declared tools. Treat the ratio as a progress indicator rather than a ranking: your goal is to methodically move more items into the “passed” column as you adjust your site.
Improving AI Accessibility and the Accessibility Tree
The report highlights how well agents can use your accessibility tree, which has become essential for web performance AI agents that need to understand page structure. Agents can examine your site in three ways: screenshots (vision), HTML, and the accessibility tree. The tree was designed for screen readers, but it effectively maps where buttons, links, and important elements live. If labels, roles, and hierarchy are missing or misleading, agents will struggle to complete actions such as submitting forms or pressing key buttons. Start by ensuring that interactive elements use semantic HTML (button, a, form) with clear text labels or aria‑labels, and avoid attaching critical actions only to generic divs or spans. Run the Lighthouse agentic readiness report again after fixes to verify that your accessibility tree is well‑formed. Over time, agent‑friendly accessibility is likely to influence which pages agents choose to recommend or use.
Using WebMCP and LLMs.txt to Guide AI Agents
Two concepts in the new report are WebMCP and LLMs.txt, both aimed at giving agents clearer instructions. WebMCP is a proposed web standard that exposes structured tools for AI agents; it teaches them how to use your site’s functionality. Declarative WebMCP wraps existing forms in simple code, while imperative WebMCP supports more advanced back‑and‑forth interactions between agents and your server. The report also looks for an LLMs.txt file, which is similar to robots.txt but for agents using your website. It can contain markdown guidance telling agents what they are allowed to do and where to find important sections. LLMs.txt is not required for Search ranking and is most useful if you host tools or workflows that agents will execute. As Marie Haynes notes, this report is about “agents using your website,” so treat WebMCP and LLMs.txt as opt‑in upgrades for agent‑heavy use cases.
