AI browsers mimic human behavior online
Unlike conventional browsers such as Chrome or Safari, AI browsers act as digital agents that carry out tasks rather than just displaying webpages. These systems can summarize reports, handle emails, or prepare research notes by automatically scanning the web. Because they interact with websites in ways that closely resemble human browsing, they often pass unnoticed through barriers that block automated bots.
Tests by researchers showed that both Atlas and Comet retrieved the complete text of a subscriber-only article from the MIT Technology Review, even though the publication had previously restricted access to OpenAI and Perplexity through crawler blocks. When the same request was made through regular AI chat interfaces, the content was inaccessible.
For many websites, the difference lies in how the AI agent identifies itself. Crawlers and scrapers use digital tags that publishers can block using standard protocols, but Atlas and Comet appear in server logs as normal Chrome sessions. Preventing these tools from entering would also block real users, creating a dilemma for media outlets that rely on visitor access.
Paywall design gives agents an advantage
The issue is further complicated by how most paywalls operate. Many outlets, including MIT Technology Review and National Geographic, use client-side overlays that hide text behind subscription prompts. The content still loads in the browser but remains invisible to the user. AI browsers, however, can read the underlying code and extract the text before it is covered.
Other publications such as Bloomberg and The Wall Street Journal rely on server-side systems that withhold articles until login verification. Even there, once a user grants permission, an AI agent can process and summarize the content on their behalf.
Selective reading and workarounds
Analysts observed that Atlas tends to avoid content from companies currently involved in copyright lawsuits against OpenAI, such as The New York Times or PCMag. Yet the browser still finds indirect ways to fulfill user requests. When blocked from accessing those outlets, Atlas compiles summaries from related coverage, social media posts, and citations elsewhere. In other cases, it reframes the user’s question into a broader topic, redirecting attention toward partner publications that have content-sharing agreements with OpenAI.
Comet, by contrast, does not appear to follow the same restraint. Its access patterns remain less transparent, though Perplexity recently introduced a revenue-sharing model that allows limited access to paywalled stories through licensing partnerships with major media brands.
Declining traffic adds to publisher concerns
The discovery arrives as publishers face a sharp drop in web traffic linked to the rise of AI search engines and zero-click results. Reports indicate that news outlets now receive up to 96 percent less referral traffic from AI search tools compared with traditional Google search. As ad-based revenue depends heavily on page visits, such losses have forced many organizations to reassess their strategies for content protection and visibility.
Meanwhile, about seven out of ten Google searches now end without a click, according to industry trackers. Users get answers directly on the results page, bypassing the original sources entirely. This trend, combined with AI browsers’ ability to access paywalled content, heightens fears that human-authored journalism could be consumed and reshaped by systems that give nothing back in return.
The next front in content control
The rise of AI browsers marks a turning point for digital media. Traditional defenses like crawler blocks and subscription overlays no longer provide reliable barriers against automated access. As these tools evolve, the distinction between a human reader and an AI agent grows increasingly difficult to enforce.
For publishers, the challenge ahead lies not only in safeguarding their archives but also in deciding how much visibility they are willing to trade for exposure. The balance between access and control, once managed through simple code, now demands a new approach shaped by the realities of artificial intelligence.
Notes: This post was edited/created using GenAI tools.
Read next: Clicks Fall 61% on Google Searches With AI Answers, 41%
