When OpenAI unveiled an artificial intelligence-based browser in October, Alphabet Inc. investors expressed concern about what it would mean for Google Chrome, the ubiquitous gateway to the internet used by billions globally.
And yet, current versions of AI browsers are far from making legacy products like Chrome obsolete: New offerings from companies such as OpenAI and Perplexity AI Inc. present occasional bugs and trip over some seemingly straightforward requests, Bloomberg News found after testing them for a month.
The new browsers, including OpenAI’s Atlas and Comet from Perplexity, put artificial intelligence assistants front and center, replacing search engines as the preset option when users enter requests. Many also offer a feature called agentic browsing, enabling them to carry out multi-step tasks on behalf of users, like completing shopping orders and extracting a list of to-dos from unread emails.
The goal, as AI developers see it, is for consumers to use their chatbots not just from their own apps or websites, but within browsers and mobile operating systems, potentially opening more avenues for ad targeting and revenue streams. The most advanced features are currently only available on a paid tier, given that AI agent features can be more expensive to run.
For now, the two categories of browsers are encouraging different kinds of user behavior, forcing app developers, web services and publishers to rethink whether they’re designing their tools for humans or robots crawling the web — and whether they can be served by the same products.
“We are seeing folks change the way that they’re interacting with the web and how they’re finding information in different categories,” said Laura Chambers, chief executive officer of Mozilla Corp., which develops Firefox.
Legacy developers like Google, Microsoft Corp. and Mozilla have taken a different approach, keeping search as a default in their browsers, while adding AI assistant features over the past year.
“People are still using Google search for things like shopping and travel, areas where you really want to research, look at a bunch of different things and make purchasing decisions,” Chambers added.
A Mozilla user research survey conducted last year found that 60% of participants were only comfortable using generative AI for low-stakes matters or things they know enough about that they can readily verify the quality of output and spot mistakes. The results, which haven’t been previously reported, were based on a survey of 1,000 people in the US and 400 Firefox users in May 2024.
Chambers said generative AI is more helpful with informational searches because it can synthesize information and provide assorted links that users can refer back to. The same survey found that saving time can encourage users to replace search with generative AI, but a need for accuracy drives them to combine the two methods.
In user research interviews, browser developers have hailed productivity wins by their AI-assisted products and observed a desire from consumers for web browsers that can do more.
A popular use for these AI browsers has been summarizing long YouTube videos and asking the assistant about a topic mentioned in the video, according to representatives from OpenAI and Chrome. Comet users are also asking “six to 18 times more questions” than they did with the regular Perplexity chatbot, said Jesse Dwyer, the company’s head of communications.
And consumers are wanting more. Adam Fry, a product lead working on OpenAI’s Atlas browser, said OpenAI has been getting a lot of requests from power users for the ability to schedule tasks so that the browser can repeat them on a regular basis.
For example, a finance professional might employ such a feature to generate a report or dashboard from an online tool every month. Ted Choc, a director of engineering for Chrome, said users want the browser to fill out government forms on their behalf and do their taxes.
