Website optimization just picked up a new audience. Alongside human visitors and search crawlers, sites are now being visited by AI agents that browse, click, and complete tasks on a person’s behalf. The Google PageSpeed Update that shipped in Lighthouse 13.3 (May 2026) is Google’s first attempt to measure whether a site is built for that third visitor, through a new category called Agentic Browsing. This guide explains what changed, what the new checks actually test, and how to prioritize fixes without overreacting to an experimental score.
Google PageSpeed Insights runs on an engine called Lighthouse, which has scored Core Web Vitals, Accessibility, Best Practices, and SEO for years. Lighthouse updates constantly, but most releases are minor. Lighthouse 13.3, released May 7, 2026, was different: it added a fifth category, Agentic Browsing, that started appearing in public PageSpeed Insights results around May 21, 2026.
Website speed and Core Web Vitals are part of Google’s page experience signals, but they are one input among many ranking factors, not a standalone guarantee of ranking position. Google has not made the new Agentic Browsing category a ranking factor at all, it’s explicitly labeled experimental.
It’s a Lighthouse audit category that measures whether AI agents, not human visitors or search crawlers, can read, navigate, and act on a page. Rather than providing a traditional 0–100 rating, it displays results as a pass percentage or ratio (such as 2/3) and remains separate from your current Performance, Accessibility, Best Practices, and SEO evaluations.
The category checks a handful of things:
Check |
What it tests |
Typical effort |
|---|---|---|
Accessibility tree integrity
|
Interactive elements have valid names, roles, and parent-child structure
|
Moderate, same work as standard accessibility fixes
|
Cumulative Layout Shift (CLS)
|
Whether the layout stays visually stable while loading
|
Low if Core Web Vitals are already healthy
|
llms.txt
|
A plain-text/markdown file at the domain root summarizing the site for AI tools
|
Low
|
WebMCP integration
|
Whether forms and site actions are registered in a machine-readable schema agents can call directly
|
High, and currently optional, the protocol is experimental
|
Tools like Perplexity Comet, ChatGPT Atlas, and other agentic browsers rely heavily on the accessibility tree, the same structured map of buttons, labels, and roles that screen readers use, rather than reading a page visually. That’s why the audit leans so heavily on accessibility fundamentals: a button with no programmatic label is invisible to a screen reader and to an AI agent alike.
One detail worth flagging honestly: Google’s own Search team has said publicly that llms.txt isn’t used for Search ranking or for surfacing in AI Overviews. At the same time, the Chrome team’s Lighthouse audit checks for that exact file. Both statements are technically consistent, Search ranking and Chrome-based agent browsing are different systems, but it’s easy to see why site owners find the messaging confusing.
Practically: llms.txt costs little to create (a short, accurate markdown summary and links to key pages), so it’s a reasonable low-risk addition, just not a ranking move.
Nothing about Agentic Browsing changes the fundamentals of website speed and page experience. The core metrics still matter for human users and search visibility:
CLS is the one metric that does double duty here: it’s already part of Core Web Vitals, and it’s now also a direct Agentic Browsing check, because agents calculate click coordinates programmatically, a layout that jumps a few pixels can cause an agent to click the wrong element entirely.
Do these first, they’re worth doing for humans, search, and agents alike:
Lower priority for most businesses right now:
Traditional Browsing |
Agentic Browsing | |
|---|---|---|
Primary reader
|
Human eyes, or Googlebot for indexing
|
AI agent acting on a person's behalf
|
Main data source
|
Rendered visual layout
|
Accessibility tree + structured schemas
|
What matters most
|
Visual design, readable content, page speed
|
Semantic markup, stable layout, labeled interactive elements
|
Failure mode
|
Slow load, confusing UX
|
Agent clicks wrong element or can't identify a control
|
Branding isn’t reserved for large restaurant groups; it benefits nearly every type of food business:
Agentic Browsing is a real signal about where the web is heading, even though most of what it checks isn’t urgent today. The businesses best positioned won’t be the ones chasing every experimental checkbox, they’ll be the ones with clean, accessible, fast-loading sites already in place, since that foundation is what both humans and AI agents depend on.
Strong Core Web Vitals, clean technical SEO, and AI-ready site architecture are becoming a real competitive edge. At Sir Marketer, we help businesses improve PageSpeed scores, pass Core Web Vitals, strengthen technical SEO, and prepare websites for the agentic web, without the guesswork.
Lighthouse 13.3 (May 2026) added an experimental Agentic Browsing category to PageSpeed Insights, alongside the existing Performance, Accessibility, Best Practices, and SEO categories.
No. It's a separate pass/fail category and doesn't change your existing 0–100 Performance score.
Not currently. Google has described the category as experimental and has not tied it to Search rankings.
It's optional. Google's Search team has stated it doesn't affect ranking or AI Overviews, but it's inexpensive to create if you want to hedge for the future.
An experimental protocol that lets a site register its forms and actions in a format AI agents can call directly, instead of inferring functionality from page layout.
Run your URL through PageSpeed Insights, or run a Lighthouse audit in Chrome DevTools with the Agentic Browsing category selected.
Yes. Core Web Vitals remain part of page experience signals, and a fast, stable site is also the foundation for passing the new agent-focused checks.
Our team will answer all your questions, We ensure a quick response.
