ChatGPT Atlas: An AI browser that changes how users navigate the web

OpenAI launched its new browser, Atlas, this week. Despite the skepticism around privacy and security, it might just be a game-changer.

Sure, others have tried this before, including Atlassian, Perplexity, and a few niche AI browsers. But none have had the reach or trust that ChatGPT commands. And with OpenAI reportedly giving away free ChatGPT credits for users who make Atlas their default browser, adoption will definitely happen.

I had ignored the earlier AI browsers. But with Atlas, I couldn’t resist. I decided to test it on a task I actually needed done this week: booking a flight on United, my go-to airline.

Putting ChatGPT Atlas to the test: Booking a United flight with constraints

I gave Atlas a fairly open-ended prompt:

“Find me a flight from New York to San Francisco or San Jose for tomorrow. I need an aisle seat, but I don’t want the flight to be much more expensive than the cheapest option. I’d prefer flights where I might get a free upgrade with my MileagePlus status. A stopover is fine if it helps me get an aisle seat, keeps costs low, and improves my upgrade changes”

In other words, this wasn’t a simple query. It required judgment: balancing hard constraints (aisle seat) with soft ones (price, upgrade potential). Something humans are good at, but algorithms often fumble.

To my surprise, Atlas handled it quite well. It surfaced a few options that made sense, even showing me the seat map. I would have likely picked the same flight myself.

United Airlines seat picker shown by Atlas browser

For tasks like this—nuanced, multi-constraint decisions—I can already see Atlas becoming a regular part of my workflow. And I suspect many others will feel the same once they try it.

How AI browsers reshape first-party data and product analytics

Here’s where it gets more interesting, especially from the lens of my world: first-party data and user behavior analytics.

The browsing behavior I imagine Atlas followed to resolve my query looks nothing like how a human browses. I couldn’t see exactly what it did. Developer tools didn’t work, and it ignored my proxy setup for man-in-the-middle inspection (it’s still in beta, after all). But it likely crawled hundreds of pages and flight options before making its recommendations.

When the user is an agent: Intent, attribution, and measurement

From a traditional analytics perspective, that looks like bot traffic. Except it wasn’t. It was a real user (me) with a real intent. It was just expressed through a conversational interface.

So here’s the challenge:

How do brands infer user intent when the “user” browsing their site might actually be ChatGPT acting on behalf of the user?

How do you run product analytics, make recommendations, or personalize marketing campaigns when you never actually see the user’s journey, and only see the AI’s output?

Sure, you might think: maybe we’ll get the original English query. But that’s unlikely. ChatGPT (or Atlas) isn’t going to share that user input with the brand.

What brands should do now: Make your site AI-friendly

The natural instinct will be to build your own chatbot. But if ChatGPT’s browser works that well, why would users bother switching? They’ll stay within the ChatGPT ecosystem, where the experience feels consistent and effortless.

That means brands will have to integrate, not compete. They’ll need to think deeply about how to make their content and APIs understandable to AI agents like Atlas.

How do you expose inventory, pricing, or availability in a way that an AI browser can parse and present accurately?

How do you make your site “AI-friendly,” just as SEO once made it search-friendly?

And how do you do all of this while still not giving away your most valuable asset (your end users) to chatGPT?

These are open questions that we, as an industry, will need to wrestle with over the next few years. But one thing feels certain: Browsing as we know it is about to change. And fast.

Final thought: AI won’t just search the web. It will use it for us

Atlas may still be in beta, but it already hints at a future where AI doesn’t just search the web. It uses it on our behalf.

And that shift will ripple through every layer of the internet economy, from analytics to advertising to the very definition of a “user session.”

Interesting times ahead indeed.