← Back to blog

What do AI browsers like ChatGPT Atlas and Perplexity Comet mean for publishers?

AI browsers such as ChatGPT Atlas and Perplexity Comet put an agent between the reader and the page, summarising content and completing tasks so people click through far less often. For publishers that means fewer page views, attribution that collapses into 'direct', and a fast-growing slice of audience that conventional analytics cannot see. The browsers that actually pay, like Comet Plus, are still the exception.


An AI browser is a web browser with a large language model built into its core, so that the assistant - not the person - does much of the reading, navigating and clicking. Instead of opening ten tabs to compare hotels, a user asks the browser and it visits the pages, extracts what matters, and returns a single answer or even completes the booking. OpenAI's ChatGPT Atlas, launched on 21 October 2025, and Perplexity's Comet, which went free on iOS in March 2026 and climbed to number three on the App Store, are the two highest-profile examples. For publishers the consequence is structural: the page still gets read, but the human who used to land on it, see the ads and move through the funnel increasingly never arrives.

What is an AI browser?

An AI browser wraps a normal rendering engine around a conversational agent that can act on the user's behalf. It does three things a traditional browser does not. It answers questions in a side panel using the content of the pages you are on or pages it fetches for you. It runs an agent or "agent mode" that can carry out multi-step tasks - searching, filling forms, comparing options, buying - by driving the browser itself. And it uses your logged-in sessions and browsing context to personalise those answers. ChatGPT Atlas and Perplexity Comet are the leading consumer products; Google and Microsoft have folded similar agentic features into Chrome and Edge. The shared design goal is to keep the user inside the assistant and reduce the number of pages they have to visit themselves.

How AI browsers differ from a Live Search Agent

This is the distinction that trips up most publishers. A Live Search Agent is a server-side bot an assistant dispatches from its own infrastructure to fetch a page, read the HTML and feed one answer, without running JavaScript and without ever touching the user's machine. An AI browser runs on the user's device inside a real browser engine, so JavaScript usually does execute and the request can carry the user's cookies and login. The retrieval bot and the agentic browser are two different surfaces, but they create the same commercial problem from opposite directions: the Live Search Agent reads without a human attached, while the AI browser has a human attached who never actually looks at the page. In both cases the content is consumed and the monetisation that depends on human attention does not fire.

What AI browsers mean for publisher traffic

The headline effect is fewer clicks. When the browser summarises an article in its sidebar or completes a task using several sources at once, the user gets what they came for without opening the publisher's page, so referral visits, ad impressions and affiliate clicks all fall. The Associated Press described Atlas as an attempt to make ChatGPT the gateway to the web, warning it could cut referrals by keeping users inside AI-generated summaries. The pattern is already visible in citation data: a SparkToro analysis in January 2026 found only 12 to 18 per cent of Perplexity citations produced an actual click-through to the cited site. The most valuable mention - your brand named inside the answer the user reads and acts on - now frequently generates no visit at all.

Why AI browser traffic breaks your analytics

Even the visits that do happen are getting harder to attribute. AI browsers and mobile AI apps routinely strip the referrer header, so a click that genuinely came from an assistant lands in Google Analytics as "Direct" or "(not set)" with no trace of its origin. One analysis of just under 450,000 AI-adjacent visits found 70.6 per cent arrived with no referrer and were logged as direct. The deliberate use of a noreferrer attribute on AI links makes some of this traffic untraceable in any client-side tool. The result is a measurement gap that runs in two directions at once: the agent reads you and leaves no analytics record, and the human who does arrive is misfiled as direct, so the channel looks smaller than it is and gets under-invested in.

Do any AI browsers pay publishers?

A few have started to, but payment is the exception rather than the norm. Perplexity's Comet Plus, announced on 25 August 2025, is the clearest example: a five-dollar-a-month subscription that commits 80 per cent of revenue to publishers, seeded with a 42.5 million dollar pool, and splitting that money across three kinds of engagement - direct human visits, citations in answers, and agent actions where the assistant uses your content to complete a task. Early partners include Fortune, Gannett, Der Spiegel, The Independent and Time. It is a meaningful precedent because it puts a price on agent-driven reads that produce no click. But it is one product from one company, the pool is finite, and most AI browsing today still reads publisher content without compensating anyone for it.

What publishers should do now

Three moves matter. First, measure server-side. Because AI browsers and agents bypass or strip client-side tags, the only reliable record of who is reading you sits in your logs and at the CDN edge, not in JavaScript analytics. Second, treat the read itself as the monetisable event rather than waiting for a click that increasingly never comes - the question is no longer only "how do I get the visit" but "how do I get paid when the answer is built from my content". Third, join the compensation schemes that exist while pushing for more, and look at edge-level approaches that monetise the AI response directly. This is the layer blankspace works in: detecting AI and agent traffic at the CDN edge and placing contextual brand mentions into the answer, so value is captured at the moment of retrieval instead of being lost when the user never lands on the page.

Frequently asked questions

What is the difference between an AI browser and a normal browser?

A normal browser renders pages for a person to read and click through themselves. An AI browser adds a built-in assistant that can read those pages for the user, answer questions about them in a side panel, and run an agent that completes multi-step tasks like searching or buying. The practical difference for publishers is that the human increasingly consumes a summary or an action rather than visiting the page itself.

Does ChatGPT Atlas reduce publisher traffic?

It is designed to keep users inside ChatGPT, which tends to reduce click-throughs to source sites. The Associated Press characterised Atlas as a gateway that could cut referrals by holding users inside AI-generated summaries. There is no single industry-wide figure yet, but the direction of travel matches the wider pattern of AI surfaces lowering referral traffic to publishers.

Why does AI browser traffic show up as "direct" in Google Analytics?

Because AI browsers and mobile AI apps frequently strip the referrer header, so the analytics tool sees a visit but has no information about where it came from and defaults to labelling it direct or unassigned. One analysis of around 450,000 AI-adjacent visits found more than 70 per cent arrived with no referrer. The fix is to read server logs and edge data rather than relying on client-side tags alone.

Do AI browsers pay publishers for content?

Mostly not, though that is starting to change. Perplexity's Comet Plus shares 80 per cent of its subscription revenue with publishers from a 42.5 million dollar pool, paying out across human visits, citations and agent actions. It is the leading example rather than the standard, and most AI browsing still reads publisher content without compensation.

How can publishers monetise AI browser and agent traffic?

The first step is to measure it server-side, since client-side analytics miss most of it. From there, options include joining revenue-share programmes where they exist and monetising the AI response itself at the CDN edge, where the content is being read. blankspace operates at this layer, detecting AI and agent traffic and injecting contextual brand mentions into the answer so the read is monetised even when no human visits the page.