OpenAI’s recently launched Atlas browser is a fascinating inversion of what users may expect from a browser, centering AI answers above traditional web links. Every click in a regular browser is a chance to see a new part of the web. Every click in Atlas is a chance to use ChatGPT.
Just typed a question into the address bar? That’s now a ChatGPT query. Want help contextualizing a web page? The Ask ChatGPT sidebar can see and analyze what’s on your screen. Want someone to buy your Halloween costume? The “agent mode” can click around on Amazon and throw some vampire fangs into your cart.
Ryan O’Rouke, OpenAI’s lead designer for the browser, demonstrated the Ask ChatGPT feature during the livestream announcement of Atlas. He asked it to summarize GitHub code appearing on the web page in his browser. He called it “a major unlock,” since ChatGPT can now see what’s happening on the page. “It’s basically you inviting ChatGPT into your corner of the internet,” O’Rouke said.
The Ask ChatGPT feature is notably free to use in Atlas, unlike the browser agent tool that’s reserved for subscribers to ChatGPT Plus and ChatGPT Pro. Here, ChatGPT shows up as another column-constrained bot on the right side of your screen, in the same vein as the AI sidebars in Perplexity’s Comet browser and Microsoft’s Edge.
After a few days of testing OpenAI’s Ask ChatGPT feature in the Atlas Browser, I’m convinced that I want to surf the web alone. Sans sidebar. In peace.
The browser is built using Chromium, an open-source browser project managed by Google that provides the codebase for Chrome, Opera, and others. Because of this, Atlas looks a lot like Chrome. (As I swiped between the two browsers on my laptop during testing, I forgot which was which a couple of times.) The nascent browser’s best days are still ahead of it, with a roadmap of upcoming features such as tab groups and an ad-blocker that could build it up to be a more fully realized competitor to Chrome. It’s available only on macOS for the time being.
With that in mind, the Ask ChatGPT sidebar in Atlas felt clunky during my first couple of days. Not only was it difficult to get used to, I found myself having trouble thinking of questions to ask about the news articles, online recipes, and other web pages as I scrolled around. Also, the sidebar squishes the main content window, so websites you visit may appear skinnier than usual. It compressed the WIRED homepage, leaving it looking incredibly janky.
While browsing the Xbox website for what game to potentially play next, Madden NFL 26 was one of ChatGPT’s suggestions after it analyzed the page. A pretty vanilla answer, and not very in line with what the bot knows about me, despite having more than a year’s worth of my ChatGPT interactions to pull from for insights.
In another example of the sidebar being underwhelming in this first iteration, I was looking through my personal Gmail inbox with Ask ChatGPT open. I requested help deciding what message to answer first, and its suggestion was to prioritize one I had already answered.
OK, then. Next, I was on the Rotten Tomatoes website trying to decide if Bugonia would be a good date night movie to see in theaters. In the Atlas sidebar, ChatGPT’s initial summary of the reviews and other info on the web page was too lengthy to be efficient. I ended up just manually combing through the reviews to get a better sense of the movie’s tone.
The oddest, and most memorable, interaction I had with ChatGPT Atlas occurred as I scrolled around on Bluesky and asked it to highlight any trends it saw. (ChatGPT listed “political anxiety” and “everyday absurdism” as two of the trends on my timeline.) I was curious what would happen if I opened my Bluesky DMs, which amounted to just a few old messages from friends. So I asked the bot what it would see if I opened that more private page.
“Opening your DMs won’t expose anything to me—I’ll simply stop ‘seeing’ the page until you go back to a public view (like your feed, profile, or a post),” read part of ChatGPT’s answer.
Knowing that my Bluesky DMs don’t have anything sensitive, I gave it a whirl. Even though the bot said it wouldn’t “see the message list, conversation text, or sender info,” that wasn’t the case.
I opened a DM inviting me to an event, then asked the bot a follow-up question asking about the message and what the invitation was about. I received a response from ChatGPT that included facts culled from the conversation, along with details about the sender. What gives? I asked the bot to explain itself, just to see whether it would backtrack the privacy claims it made earlier in the chat.
“I do not have access to your private messages or account data unless you explicitly paste or show that text to me in the chat,” read ChatGPT’s response. “What you just did—asking ‘what was this an invite to?’—temporarily surfaced part of the message context so I could answer your question. But I don’t have any background or visibility into your inbox beyond what you actively bring into our conversation.”
The first answer the Ask ChatGPT sidebar spit out is a potential example of an AI “hallucination,” aka error, that’s common during interactions with generative AI tools. The second is more aligned with how the tool actually works.
While some users may appreciate having a chatbot always pulled up on the side of their screen, ready to surface related facts or summarize details, it felt like an unreliable tour guide to me. One who was overly confident in its bland responses and taking up too much space.
I’ll keep testing Atlas as my main browser for the next few weeks, but for now, I’m leaving that sidebar closed. I prefer the fullscreen version of the internet.
Disclaimer : This story is auto aggregated by a computer programme and has not been created or edited by DOWNTHENEWS. Publisher: wired.com
