Google Chrome can now browse the web for you. Not just search. Browse. Click. Scroll. Fill out forms. Complete purchases.
On January 28, 2026, Google rolled out "Chrome auto browse," an AI-powered feature that turns the browser into an autonomous agent. Tell Gemini what you want, and Chrome handles the rest. Book a flight. Compare prices. Fill out that government form you've been avoiding. The browser does the clicking while you do something else.
Chrome has over 3 billion users. This is not some niche product announcement. This is the starting gun for mainstream agent traffic.
How it actually works
When you give Gemini a task in Chrome, the browser opens a new tab with a small sparkle icon. That's your signal that AI is taking over the session. The Gemini side panel shows each step as it happens. Scrolling. Clicking. Typing. You can watch or ignore it.
The agent runs locally on your device with cloud model support from Gemini 3. It can see the page the same way you do, identifying buttons, forms, and navigation elements through multimodal understanding. When it needs to log in somewhere, it pulls credentials from Google Password Manager (with your authorization). When it's about to do something consequential, like confirm a purchase or post to social media, it stops and asks.
TechCrunch reports the feature is designed to compete with the wave of AI-native browsers that have emerged over the past year. But there's a difference between a startup browser with a few million users and the browser that runs on two-thirds of all computers.
The use cases Google is pitching
Google's examples reveal what they think people will actually use this for:
- "Go to Etsy and find craft supplies under $75"
- Schedule appointments across multiple calendar systems
- Compare flight prices across airline websites
- Fill out expense reports and reimbursement forms
- Manage subscription renewals
- Handle license renewals and government paperwork
Notice a pattern? These are all tasks that require visiting multiple pages, filling out forms, comparing information, and making decisions based on what's found. The tedious middle of the internet. The parts nobody enjoys.
Right now, auto browse is limited to AI Pro and AI Ultra subscribers in the US (in case you were wondering why there are no screenshots in this article), with a daily cap on agentic actions. But limitations at launch rarely reflect what happens six months later. Gmail was invite-only once.
What this means for your website
Here's where it gets interesting for anyone who runs a website.
Chrome auto browse doesn't care about your carefully designed hero section. It doesn't notice your brand colors. It's not impressed by your animation on scroll. It's looking for the button it needs to click, the form it needs to fill, the information it needs to extract.
When an AI agent visits your site, it experiences a fundamentally different version than a human does. It sees the DOM, not the design. It parses the structure, not the story. And if your structure is a mess? If your buttons don't have clear labels? If your forms use JavaScript that breaks without human interaction? The agent fails. It moves on. It finds a competitor whose site actually works.
This isn't hypothetical. Websites wites with clean semantic HTML consistently outperform sites that were built purely for human eyes.
The accessibility connection
There's a silver lining here! Optimizing for AI agents and optimizing for accessibility are nearly the same work.
Screen readers need clear labels. So do AI agents. Keyboard navigation requires logical tab order. AI agents follow similar patterns. Semantic HTML helps assistive technology understand page structure. AI agents rely on the same signals.
The sites that will perform best with Chrome auto browse? They're the sites that already work well with screen readers. The overlap isn't coincidental. Both audiences need to understand what's on a page without relying on visual design alone.
If your site passes WCAG accessibility guidelines, you're already ahead of most competitors for the agent era. If your site was built with "move fast and break things" energy and zero accessibility consideration, you have a problem that just got more urgent.
What breaks agent browsing
Based on what we know about how these systems work, certain patterns will cause problems:
- Anti-bot CAPTCHAs on every page. If your site treats every visitor as a potential threat, you'll block the agents trying to complete legitimate tasks. Chrome auto browse has some ability to handle authentication (via Password Manager integration), but sites that throw up walls at every turn will see agents fail and give up.
- Forms that only work with mouse interaction. If your dropdown menus require hover states, or your date pickers only respond to click events that JavaScript simulates, agents will struggle. Forms need to work with keyboard input and programmatic interaction.
- Infinite scroll without pagination. AI agents need to know when they've reached the end of a content list. Infinite scroll that loads content based on scroll position creates ambiguity about whether more content exists.
- Content hidden behind "show more" buttons. If critical information only appears after clicking an expansion button, and that button isn't clearly labeled, agents may not find what they're looking for.
- Heavy client-side rendering with no server fallback. If your page is a blank div until JavaScript runs, and that JavaScript assumes human interaction patterns, agents may not see your content at all.
I talked about these in an episode I released earlier today, btw. Check out No Hacks episode 215: The Agent-Broken Web here.
The competitive angle
Google is doing this because they're worried about losing browser market share to AI-native competitors. The subtext is clear: if people start using AI browsers to get things done, Chrome loses relevance. By building agentic features directly into Chrome, Google keeps users (and their data) in the ecosystem.
But there's another competitive angle that matters more to website owners.
When AI agents can comparison shop across sites autonomously, the friction advantage disappears. The site that was slightly harder to use but had better products? Now it competes on equal footing, or worse, loses because the agent couldn't navigate it properly. The site with the complicated checkout flow that humans tolerated because they'd already invested time? The agent has no sunk cost. It just leaves.
Agent-friendly sites will get more completed transactions. Agent-hostile sites will get abandoned sessions. The browser won't tell you an AI gave up on your checkout. You'll just see the conversion rate drop and wonder why.
What to do about it
Some practical steps:
- Audit your site with a screen reader. If VoiceOver or NVDA can navigate your site successfully, AI agents probably can too. If they can't, fix that first.
- Check your semantic HTML. Are your buttons actually
<button>elements? Are your form inputs properly labeled? Does your heading hierarchy make sense? AI agents rely on these structural cues more than humans do. - Test without JavaScript. Disable JavaScript and see if your core content and navigation still work. If the page is blank, you have a rendering problem that affects both search engines and AI agents.
- Review your bot policies. Check your robots.txt and bot detection systems. Are you blocking Googlebot? Probably not. But are you blocking the new generation of AI crawlers? Many sites are, without realizing the cost.
- Add an llms.txt file. This is a new standard (I use it on this site) that tells AI agents how to interact with your content. It's like robots.txt but for LLMs. Still not widely adopted, but it's a simple text file and won't harm anyone.
Glimpse by Web Performance Tools can help you identify most of these issues. And it's free.
The timeline accelerates
A year ago, AI agents browsing websites autonomously was a cool demo. Six months ago, it was a feature in niche tools. Today, it's being built into the browser that runs on 3 billion devices.
The timeline for this shift keeps compressing. Every prediction about "when agents will matter" has been wrong by being too conservative. Chrome auto browse isn't the end of this trend. It's barely the beginning.
The question isn't whether AI agents will visit your website. They already are. The question is whether your website works when they do.
Do you know the answer?