AI browsers are still too small to hurt publishers — but the ad-blocking default is the real warning sign

Perplexity's Comet browser had one of the fastest pricing collapses in recent software history. It launched in July 2025 locked behind Perplexity's $200-a-month Max subscription tier — a price point that signaled a premium, cautious rollout. By early October 2025, it was free globally. By November, it was free on Android. By March 2026, free on iOS. Nine months from luxury product to zero-cost download, complete with built-in ad and tracker blocking, an AI-generated answer engine, and a limited allotment of autonomous “agent credits” that let it complete multi-step tasks on a user's behalf without direct clicking.

That trajectory matters more than any single feature, because it tells you Perplexity isn't trying to monetize Comet as a product — it's trying to win browser market share as fast as possible, subsidized by the company's other revenue lines. OpenAI's ChatGPT Atlas, Google's Gemini-in-Chrome integrations, and various experiments from The Browser Company (Dia, the successor to Arc) are all racing along a similar track: turn browsing from a page-by-page clicking activity into something closer to giving an assistant an errand and letting it act inside the browser on your behalf.

For publishers, this isn't an abstract shift. It attacks the two pillars their business model rests on simultaneously. The first is referral traffic — when an agentic browser reads, summarizes, and acts on information without the user ever navigating to the source page, the publisher loses the pageview that used to come from being the answer to a search query. The second is ad revenue, which depends on tracking and impressions. Comet's ad and tracker blocking is on by default, not opt-in, meaning every session run through it is invisible to the ad tech stack that funds most of the open web.

According to reporting from Digiday, publisher sentiment on this right now is a mix of genuine alarm and studied calm. One unnamed publisher described agentic browsers as a “3 percent problem” — real, but not yet large enough to reorder editorial or business priorities. That's a defensible read of the current numbers: Chrome alone still commands roughly 70% of global browser market share, and none of the AI-native browsers have meaningfully dented that dominance yet. Comet, Atlas, and their peers remain nascent by comparison, still building habits and awareness rather than displacing incumbent browsing behavior at scale.

But “3 percent problem” is a description of today, not a forecast. The IAB Tech Lab, the standards body that has spent two decades building the infrastructure of programmatic advertising, is running early pilots of something called a Trusted Server framework — an attempt to give publishers and advertisers a way to distinguish legitimate agentic-browser traffic from bot traffic or scraping, so that impressions and referrals can still be counted and monetized even when a human isn't the one clicking. At least half a dozen U.S. and European publishers are reportedly involved in early pilots. That's a tell: publishers aren't dismissing the trend, they're quietly building the plumbing to survive it, even while they downplay its current urgency in public.

The deeper problem for the ad-supported web is structural, not just about traffic volume. Current ad tech has no reliable way to tell the difference between a human scrolling a page and an AI agent parsing it on a human's behalf — which means advertisers, worried about paying for impressions nobody actually saw, are already pulling spend at the margins rather than waiting to find out how big the problem becomes. That's a rational response to uncertainty, but it also means agentic browsers can shrink ad budgets before they've captured meaningful market share, simply by making advertisers nervous about measurement integrity.

What publishers should actually do about it: waiting for agentic browsers to reach meaningful scale before building a response is a mistake, because the infrastructure question — how do you get paid when an AI agent, not a human, is the one consuming your content — takes years to solve, not months. The realistic paths forward are a hybrid of the IAB's Trusted Server approach (verified agentic traffic that still counts commercially), direct licensing deals with AI browser and assistant makers for content access (the model several major publishers have already struck with OpenAI and other AI labs), and diversifying revenue away from pure pageview-driven advertising toward subscriptions and direct reader relationships that don't depend on someone clicking through in the first place. None of these are complete solutions today. But the publishers already piloting them will be in a materially better position when the “3 percent problem” becomes a 15 or 30 percent one — and given Comet's nine-month path from $200 subscription to free download, that shift could arrive faster than the current calm suggests.