For twenty-five years, the automated visitors to your website were search crawlers — bots that read and indexed your pages so humans could find them. That's changing. In May 2026, Google shipped a new category in Lighthouse (the tool that scores every website on performance, accessibility, SEO, and best practices) called Agentic Browsing. It doesn't ask whether humans can use your site. It asks whether an AI agent can.
AI agents — ChatGPT's browsing mode, Google's own Gemini in Chrome, Anthropic's Claude, Perplexity — increasingly don't just read web pages. They act on them: filling out forms, clicking buttons, completing bookings, finishing checkouts, on a person's behalf. If your site's structure confuses a human-only design, it fails silently for an agent, and that agent (and the task, and potentially the sale) simply moves on.
The Agentic Browsing category checks four things:
- WebMCP implementation — a new proposed standard that lets a site declare its tools and forms directly for AI agents to call, instead of the agent having to guess by screenshot
- Accessibility tree quality — whether buttons, forms, and interactive elements have proper labels and structure a machine can actually parse
- Layout stability (CLS) — if your layout shifts while loading, an agent can click the wrong thing or miss the target entirely
- An llms.txt file — a machine-readable summary at your domain root, similar in spirit to robots.txt, but written for AI systems
It's still marked "experimental" by Google, and it doesn't produce a 0–100 score like other Lighthouse categories — just a pass ratio, like 3/4. That's deliberate; the standards are still forming. But the direction is not ambiguous: agentic browsing went from zero to a default part of every Lighthouse and PageSpeed Insights report in under two months, and it's inherited automatically by every tool built on top of Lighthouse — including the SEO audit tools most agencies already use.
Where WebAcumen fits in
We build and audit websites specifically for this. It's a natural extension of the AI-visibility work we already do under GEO (Generative Engine Optimisation) — getting your brand cited in AI answers is one problem, making sure AI agents can actually operate your site is the next one, and we handle both under the same roof, by the same team, so there's no gap between the two.