OpenAI’s new AI browser, ChatGPT Atlas, launches with ambitious goals but delivers a mixed experience that feels more like a step forward than a leap. Built on Chromium and currently available only on macOS, the browser aims to place ChatGPT at the center of web navigation. However, it still falls short when compared to existing AI-powered alternatives like Perplexity’s Comet and Google Gemini in Chrome. The interface is clean and minimal, with a collapsible chat history panel on the left, a URL bar in the center that doubles as a search input, and an “Ask ChatGPT” button on the right. This button lets users query the AI about the current page, a feature that works similarly to AI assistants in other browsers. For Plus, Business, and Pro users, an agentic mode allows the browser to perform tasks like adding items to a shopping cart or booking reservations. The homescreen offers some initial suggestions, such as “Find the best restaurants near me,” and gradually populates with recommendations based on past searches. However, the personalization feels inconsistent—after multiple queries about Halloween costumes, the browser suggested related content, but failed to surface restaurant recommendations despite repeated searches. It also lacks the rich, customizable discovery pages found in browsers like Perplexity Comet or the news and weather widgets in Microsoft Edge. One of the biggest issues is the search experience. When you enter a query, ChatGPT first provides an AI-generated summary, then offers a list of 10 web results. These results are often inaccurate—searching for “news near me” returned links to news sites in cities far from my location. While the browser does show maps and basic details for local businesses, it doesn’t allow users to view full reviews, see their sources, or access additional information, putting it at a disadvantage compared to Comet, which integrates with TripAdvisor. The browser saves all search results and chat history in the same place as ChatGPT conversations, which can quickly become cluttered. To help with navigation, it includes a “browser memories” feature that tailors responses based on recent activity and can automatically organize and close tabs. However, the visual separation between different search topics is subtle—just a faint gray line and a label—making it hard to track what’s been discussed. Despite these shortcomings, the core AI interaction works reasonably well. Highlighting text on a webpage and asking ChatGPT for context or related information functions smoothly, similar to what’s available in Gemini and Comet. The agentic mode can complete tasks like sending emails or creating calendar events—though it’s slow. In one test, it took 10 minutes to add three items to an Amazon cart, with the AI narrating its own struggles, like waiting for pages to load. In contrast, Perplexity Comet completed the same task in about two minutes. The browser also shows promise in handling complex requests, like finding and booking a restaurant, though it made a mistake by scheduling a reservation for this Friday instead of next Friday. Ultimately, while ChatGPT Atlas introduces useful features and demonstrates OpenAI’s vision for an AI-driven web experience, it still feels like a prototype. With existing tools like the ChatGPT app’s integrations with Zillow, Spotify, and Walmart, users can already perform many of the same tasks without needing a dedicated browser. For now, ChatGPT Atlas offers a novel but underdeveloped approach to AI-powered browsing—more of a proof of concept than a compelling reason to switch from Chrome or other established browsers.