Until recently, whoever arrived at your website was either a person or a search engine crawler. Now a third kind of visitor shows up: an AI agent browsing on a user’s behalf. In January 2026 Google began offering auto browse in Chrome, a Gemini feature that fills in forms and builds shopping carts for the user, starting in the US for paying subscribers. In August it reached Chrome on Android as well. A form that a person completes without thinking can be a dead end for an agent like this.<\/p>\n\n\n\n
According to Google’s guidance on web.dev, agents combine three sources: a screenshot, the page’s HTML structure, and the accessibility tree. The accessibility tree is the same representation of the page that screen readers for blind users rely on. In June 2026 the Chrome blog described a well-formed accessibility tree as the main way agents understand a page. OpenAI also advises site owners to mark up buttons, menus and forms with ARIA, giving every element a role and a label.<\/p>\n\n\n\n
A site built properly for accessibility has already covered most of the ground. A site where a “button” is really a clickable area with no role, or where form fields have no labels, confuses agents too.<\/p>\n\n\n\n
In June 2026 Chrome announced a new Lighthouse category called Agentic Browsing, which works from version 150 of the browser. It does not give a score from 0 to 100. It reports which checks passed, and Chrome explicitly describes it as informational, with no recommended threshold. These are the checks:<\/p>\n\n\n\n
In September a check was added for a file that tells agents which resources a site offers.<\/p>\n\n\n\n
In April 2026 Cloudflare released a check called Agent Readiness, which anyone can run for free at isitagentready.com. It looks at whether an agent can find the content, whether it can get it in a convenient format, how the site sets access rules for bots, and which capabilities it exposes to agents. A scan of the top 200,000 sites found that 78% have a robots.txt file. Only 4% state how their content may be used for AI, and fewer than 4% can serve an agent a Markdown version of a page when it asks for one.<\/p>\n\n\n\n
Keep in mind that these scores are diagnostic tools. Google makes clear that no outside tool has access to its ranking systems, and a high score does not guarantee a place in the results.<\/p>\n\n\n\n
WebMCP is a proposal written by engineers from Google and Microsoft. It lets a site define tools for an agent, meaning defined actions such as “book a meeting” or “request a quote”, with a description for each field. In Chrome you can try it from version 149 as part of an origin trial. The proposal is not a standard yet and may change.<\/p>\n\n\n\n
For lead generation sites, the interesting part is the declarative version: a few attributes added to an existing form, with a name and a description for the form and for each field. If you leave out the attribute that allows automatic submission, the agent fills in the form and the user clicks “Submit” themselves. That is the right boundary. The agent saves the typing, and the person still makes the decision.<\/p>\n\n\n\n
An llms.txt file is a summary of a site written for language models. In June 2026 Google wrote that it is not needed for Google Search, where it neither helps nor hurts. According to figures published by PPC Land in July, the number of sites with such a file grew 8.8 times in a year, and according to Ahrefs, 97% of the files received no requests at all. There is no reason to remove an existing file, but no reason to start with one either.<\/p>\n\n\n\n
Most of these items are accessibility requirements that already apply: in Israel under standard IS 5568, and in Europe under WCAG. Investing in accessibility improves the site in one go for people with disabilities, for agents and for everyone else.<\/p>\n\n\n\n
Google added a check to Lighthouse that shows whether AI agents can work with a website, and Cloudflare built a check of its own. What gets tested, what the numbers show, and what to fix on a lead generation or ecommerce site.<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[12],"tags":[],"class_list":["post-102","post","type-post","status-publish","format-standard","hentry","category-ai-development"],"_links":{"self":[{"href":"https:\/\/www.assemblyso.com\/en\/wp-json\/wp\/v2\/posts\/102","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.assemblyso.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.assemblyso.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.assemblyso.com\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.assemblyso.com\/en\/wp-json\/wp\/v2\/comments?post=102"}],"version-history":[{"count":1,"href":"https:\/\/www.assemblyso.com\/en\/wp-json\/wp\/v2\/posts\/102\/revisions"}],"predecessor-version":[{"id":104,"href":"https:\/\/www.assemblyso.com\/en\/wp-json\/wp\/v2\/posts\/102\/revisions\/104"}],"wp:attachment":[{"href":"https:\/\/www.assemblyso.com\/en\/wp-json\/wp\/v2\/media?parent=102"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.assemblyso.com\/en\/wp-json\/wp\/v2\/categories?post=102"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.assemblyso.com\/en\/wp-json\/wp\/v2\/tags?post=102"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}