What Lighthouse Agentic Readiness Is and Why It Matters
Lighthouse agentic readiness is a new website readiness report in Chrome that evaluates whether your site’s structure, semantics, and integrations allow AI agents to understand, explore, and use your pages as intended. It focuses on how well agents can discover content, trigger key actions, and follow your rules, so developers can fix structural or accessibility problems before automated tools misinterpret their sites. Google has added this AI agent testing flow directly into Lighthouse, so you can run it without installing extra tools or extensions. According to Marie Haynes, the new report “tells you whether your website is discoverable for AI agents, whether you have WebMCP integration set up, and an evaluation of your LLMs.txt file.” In practice, that means the same panel you use for performance and SEO can now help you future‑proof your site for agentic browsing.
How to Run the Agentic Readiness Report in Chrome Canary
To use this website readiness report, you need Chrome Canary, the upcoming beta version of Chrome where new Lighthouse features appear first. After installing Canary, open your site, right‑click anywhere on the page, and choose Inspect. In the DevTools panel that opens, select the Lighthouse tab at the top. You will see a new category called “Agentic Browsing.” Enable that category, keep other categories on or off as you prefer, and start the Lighthouse run. Lighthouse will crawl the current page and generate an AI agent testing report in the same panel. Instead of a score out of 100, Lighthouse agentic readiness displays a ratio that shows how many checks your page passes. Marie Haynes notes that even Google’s own documentation about this feature can fail some of the checks, so expect to see issues and treat them as a to‑do list.
Reading the Results: Discoverability, Structure, and Semantics
Once the report finishes, scan the overview that lists passed and failed checks. The first group concerns how discoverable your site is for agents, which depends on clear HTML structure, meaningful headings, and links that describe their targets. If essential actions hide behind unclear labels or custom widgets, an agent may not know how to reach them. The report also inspects how your semantics map to real interactions: Are primary buttons properly marked? Are forms grouped and labeled? Does the document outline help an agent decide what is important? These are the same foundations that help human users and screen readers. Each failed item expands to show details and recommended fixes, so you can iterate quickly. Use this as an opportunity to clean up markup, remove ambiguous elements, and ensure critical flows are represented in a way AI agents can follow.
Check the Accessibility Tree: Making Pages Understandable for Agents
A core part of Lighthouse agentic readiness is how your accessibility tree describes the page. Agents can examine your site via screenshots, raw HTML, and the accessibility tree, which was designed for screen readers but now also guides automated tools. If the tree is incomplete or misleading, an agent may not find the right buttons, forms, or navigation paths. In Chrome Canary DevTools, inspect elements and switch to accessibility views to see what an agent would perceive. Look for missing roles, vague names like “button” without context, or landmarks that are not set. Fix these in your HTML with ARIA roles, labels, and semantic tags so the accessibility tree matches the real UI. Improving this area tends to help both human users and AI tools, reducing friction for people who rely on assistive technology and for agents that want to execute tasks on your site.
WebMCP and LLMs.txt: When You Need Them and What to Do
The new Lighthouse agentic readiness report also checks whether your site offers WebMCP integration and, if present, evaluates an LLMs.txt file. WebMCP is a proposed web standard that lets you expose structured tools to AI agents so they know how to use your site’s functionality. There are two flavors: declarative, which wraps existing forms with simple code, and imperative, which supports back‑and‑forth interaction. The report will alert you if no WebMCP integration exists, but you only need to act if you expect agents to run tools on your site, such as calculators, search filters, or booking flows. It will also look for LLMs.txt, a proposal similar to robots.txt that gives markdown instructions to agents about what they are allowed to do and where to find important information. Marie Haynes explains that LLMs.txt “is for agents using your site, and not for Search reasons,” so treat it as an opt‑in control layer for agentic behavior.
