×
OpenAI’s Atlas browser struggles with unreliable AI web assistance
Written by
Published on
Join our daily newsletter for breaking news, product launches and deals, research breakdowns, and other industry-leading AI coverage
Join Now

OpenAI’s new Atlas browser features an “Ask ChatGPT” sidebar that promises to serve as an AI-powered web guide, analyzing pages and answering questions about content. However, initial testing reveals the feature often provides underwhelming responses and contains concerning privacy inconsistencies that undermine its reliability as a browsing companion.

What you should know: Atlas is OpenAI’s Chromium-based browser that centers AI interaction above traditional web browsing, with the Ask ChatGPT sidebar available for free while agent features require paid subscriptions.

  • The browser currently runs only on macOS and looks nearly identical to Chrome, with planned features like tab groups and ad-blocking still in development.
  • Ryan O’Rouke, OpenAI’s lead designer, described the sidebar as “a major unlock” that lets users “invite ChatGPT into your corner of the internet.”
  • The sidebar compresses the main content window, making websites appear “skinnier than usual” and leaving some pages looking “incredibly janky.”

Performance issues: Testing revealed multiple instances where the AI assistant provided bland, irrelevant, or factually incorrect responses.

  • When analyzing the Xbox website for game recommendations, ChatGPT suggested “Madden NFL 26” — a generic answer that ignored the user’s gaming history despite having access to over a year of ChatGPT interactions.
  • The assistant recommended prioritizing an already-answered Gmail message when asked to help decide which email to respond to first.
  • For movie research on Rotten Tomatoes, ChatGPT’s initial summary was “too lengthy to be efficient,” forcing manual review of the actual content.

Privacy contradiction: The most concerning issue emerged during Bluesky testing, where ChatGPT made conflicting statements about its ability to access private messages.

  • Initially, the AI claimed it would “stop ‘seeing’ the page” when users opened private DMs and wouldn’t access “message list, conversation text, or sender info.”
  • However, when tested with actual DM content, ChatGPT successfully analyzed private messages and provided details about both the conversation and sender.
  • When questioned about this contradiction, the AI provided a different explanation, stating it doesn’t have access “unless you explicitly paste or show that text to me in the chat.”

The big picture: Atlas represents OpenAI’s attempt to reimagine web browsing around AI interaction rather than traditional link-following, but early execution suggests the technology isn’t ready to replace human web navigation instincts.

What they’re saying: “It’s basically you inviting ChatGPT into your corner of the internet,” O’Rouke said during the Atlas announcement livestream.

  • The reviewer concluded: “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.”
OpenAI's Atlas Wants to Be the Web’s Tour Guide. I’m Not Convinced It Needs One

Recent News

Chinese startup Noetix launches $1.4K humanoid robot for consumers

The three-foot robot costs about the same as a flagship smartphone.

7 AI stocks with highest trading volume spark investor interest

High dollar volume suggests institutional interest in companies beyond the usual tech giants.