Editorial: when the audience is a machine
Perplexity is blocking ads served to its AI agent, exposing a measurement gap the ad industry has not yet solved.
Mariano De Vitto · August 2026
Perplexity has drawn a line in the sand that the advertising industry will be arguing over for years: it is now blocking ads that The New York Times served inside its AI agent's browsing sessions, arguing that showing ads to a machine acting on a user's behalf is deceptive. The decision, first reported by Digiday, is not just a spat between a publisher and an AI startup. It is the first visible skirmish in a much larger war over what counts as an audience when the reader is no longer a person.
A New Kind of Reach
Perplexity's AI agent works by browsing the web on behalf of users to answer their queries, visiting pages, reading content and synthesizing answers without a human ever looking at the page itself. When publishers like The New York Times serve the same display ads to that agent as they would to a human visitor, the entire logic of impressions, viewability and engagement starts to break down. Those metrics were built on the assumption that a person's eyes, attention and emotions were on the other end of every pageview.
Perplexity's argument, as reported by Digiday, is straightforward: counting an ad shown to a bot as reach misrepresents what actually happened. A machine cannot be persuaded by a banner, cannot be annoyed by an interstitial and cannot be delighted by clever creative the way a human can. Whatever value advertisers thought they were buying when they paid for that impression simply was not delivered, which is why Perplexity frames the practice as deceptive rather than merely inefficient.
Who Pays When Bots Browse
The bigger issue, though, is that nobody, including Perplexity, has fully defined what fair looks like in this new arrangement. Blocking ads is a clean, headline friendly gesture, but it sidesteps a much harder question: how do publishers get paid when their content is consumed by an agent instead of a person? The New York Times, like most major publishers, built its digital business on a mix of subscriptions and advertising, and advertising revenue depends on the assumption that a human being is on the other end of the page. If that assumption no longer holds for a growing share of traffic, publishers need a new model, and right now no one, including Perplexity, has proposed one.
There is also an unresolved question of authority. Who decides what counts as deceptive versus simply new? Perplexity is unilaterally setting the rule for its own agent, but publishers, ad tech platforms and marketers were not part of that decision, and there is no shared standard for what agent traffic should look like, how it should be labeled, or whether it should be monetized at all.
What Marketers Should Do
Why it matters: brands should treat this as an early warning, not a settled standard. Measurement systems built for human eyeballs, click through rates, viewability thresholds, engagement scores, will not automatically translate to AI agents that scan a page in milliseconds and never see a single pixel rendered. Marketing teams need to apply human judgment now, before agent traffic becomes a meaningful line item in any media plan.
Concretely, that means questioning vendor claims about agent reach before folding them into planning assumptions, demanding clarity from platforms like Perplexity and publishers like The New York Times about how agent visits are counted and disclosed, and resisting the instinct to treat more traffic, even machine traffic, as automatically more valuable. The industry spent two decades learning to separate real engagement from bot fraud in a human dominated web; that same discipline now has to be rebuilt for a web where legitimate agents outnumber accidental bots, and where the audience on the other end may never be a person at all.
The Signal Brief · Mariano De Vitto — Head of Marketing, Barcelona