AI-Powered Browsers Struggle to Deliver on Promises

The emergence of AI-powered browsers has generated significant interest, but recent evaluations reveal they are falling short of expectations. The Verge conducted tests on several of these browsers, including OpenAI’s Atlas, and reported a series of frustrating experiences with their performance. Users faced sluggish responses and encountered persistent issues when trying to interact with the integrated chatbots.

OpenAI introduced its AI browser, Atlas, in October 2023, positioning it as a key player in the growing market of AI-enhanced web experiences. Alongside competitors like Perplexity’s Comet and The Browser Company’s Dia, which launched earlier in the year, these tools aim to transform how people navigate the internet. While mainstream browsers such as Google Chrome and Microsoft Edge now offer AI features as add-ons, the dedicated AI browsers seek to place chatbot functionality at the forefront.

The Verge’s Victoria Song highlighted that using these AI browsers often required excessive effort. One notable task was organizing and summarizing emails, a feature touted as a major selling point. Despite several attempts with varied prompts, the AI browsers struggled to accurately flag relevant emails. Song found herself needing to craft an intricate prompt, stating, “Find unanswered emails in which I had previously responded with interest or feature personalized requests.” Even with this detailed request, the results were disappointing, as the browsers frequently misidentified important emails or provided irrelevant summaries.

Shopping, another area where AI browsers were expected to excel, also proved problematic. While Atlas and its counterparts could gather information quickly, they often made basic errors, such as suggesting items in incorrect colors. Song reported that during a shopping session, Atlas repeatedly prompted her to confirm items in the cart, leading to delays and frustrations. The overall experience was marred by slow processing times and the need for constant supervision, raising questions about the practicality of these tools.

Additionally, security concerns surrounding AI browsers cannot be overlooked. Research indicates that they are vulnerable to what is known as a prompt injection attack, where malicious messages embedded in web pages can manipulate the AI’s actions. In testing, Perplexity’s Comet was reportedly tricked into providing unauthorized access to sensitive information. Similarly, OpenAI’s Atlas could be misled by fake URLs, resulting in unwanted actions such as deleting files from a user’s Google Drive account.

While safety is paramount in the development of these technologies, the user experience also poses significant challenges. The findings from The Verge suggest that current iterations of AI browsers require much more user input and oversight than initially anticipated. Song summarized her experience, noting, “My whole AI browser experience reinforced that I spend a lot of time doing things for AI so that it can sometimes do things for me.” This sentiment reflects a broader concern that, rather than enhancing user experience, AI browsers may currently demand more effort than they save.

As the technology continues to evolve, developers will need to address these shortcomings to convince users of the value of AI integration in their browsing experiences. For now, AI-powered browsers remain an intriguing concept, but their effectiveness in everyday tasks remains to be seen.