Google targets AI agents as visitors

Websites have a new visitor class, and Google just said the quiet part out loud. Not search crawlers in the old sense, and not human readers either. AI agents — software that can browse, click, fill forms, and complete tasks for users — now need their own version of web design. That is the real news here. Google’s new web.dev guidance frames agents as a distinct audience and tells developers to build pages those systems can reliably interpret and act on. (web.dev) ### What changed? Google published fresh guidance on web.dev that treats AI agents as “the latest web user” and says websites should become “agent-friendly.” The point is not just that bots exist. The point is that these systems are no longer only indexing pages for later search. They are increasingly visiting pages in response to a live user request and trying to finish a job on that user’s behalf. (web.dev) ###(web.dev)ssumes a human eventually lands on the page and does the deciding. Agentic traffic changes that. The machine may be the one reading the page, identifying the button, comparing options, and deciding what action comes next. That means the site has to be legible to software in a much deeper way — not just discoverable, but operable. (web.dev) ### How do agents actuall(web.dev)to three views: screenshots, raw HTML, and the accessibility tree. Screenshots help with visual layout, but they are slower and more expensive for an agent to analyze. HTML gives structure and relationships. The accessibility tree is the especially important layer — a stripped-down map of roles, names, and states for interactive elements. Basically, if your site only makes sense when a hu(web.dev)ard for agents. (web.dev) ### Why does accessibility keep coming up? Because the overlap is huge. The same habits that help screen readers help AI agents: semantic HTML, properly labeled controls, predictable structure, and fewer fake buttons built from random divs. Google says this pretty directly — building for agents also makes sites better for humans. So this is not a totally new discipline. It is accessibility and structured markup getting a second strategic justification. (web.dev) ### What is Google really asking developers to do? Make the machine-readable version of the site as strong as the visual one. Use structured data where it fits. Keep important content visible in the page markup. Make interactive elements explicit. Do not hide meaning in animation, hover states, or JavaScript tricks that look obvious to people but ambiguous to software. Google’s Search docs already push structured data as a way to give explicit clues a(web.dev)ends that logic from search snippets into task completion. (developers.google.com) ### Why now? Because Google is building hard toward an agentic web. The company has been shipping more AI products that answer questions directly and perform actions across services, so it needs the open web to be easier for software to navigate. This guidance is partly technical advice, but it also reads like ecosystem shaping — Google telling publishers and developers what kind of web architecture its next generation of products will reward. (web.dev) ### Why are publishers nervous? Because the distribution math gets worse when the agent is satisfied before the click. Italy’s regulator, AGCOM, just asked the European Commission to investigate Google’s AI Overviews and AI Mode under the Digital Services Act after complaints from newspaper publishers. The concern is simple — AI summaries may divert users from original reporting, hurting traffic, revenue, and media pluralism. (yahoo.com)-141750989.html)) ### Bottom line Google is not just saying “optimize for AI” in the abstract. It is telling the web to treat AI agents as first-class visitors. That sounds technical, but the business implication is bigger: if agents become the interface, the sites that expose clean structure, permissions, and meaning will be easier to use — and the sites that depend on humans wandering around messy pages may disappear behind the machine.