Editorial: the week the audience became the agent
Counterpoint to the AI-browsers-are-eating-our-budgets panic. The agent is not the enemy. The agent is the audience using a better tool.
Mariano De Vitto · May 2026
The story of the last six weeks in marketing has a fixed villain. ChatGPT Atlas, OpenAI's AI-native browser, is clicking ads it should not click, comparing products on behalf of users who never see the brand's site, and generating analytics noise that costs real money to clean up. Every CMO conversation I have been in for the past month opens with some version of what do we do about the bot traffic. The framing has hardened into something that sounds true but is not.
The agent is not the enemy
I want to argue against my own previous coverage on this. The agent is not the enemy. The agent is the audience using a better tool. Every time I hear a marketing organization describe AI browsers as a threat to be detected and refunded, I hear an industry making the same mistake it made about ad blockers, about iOS 14, about cookie deprecation, and about the SERP collapsing into AI Overviews. The mistake is treating a structural shift in user behaviour as a fraud problem.
It is true that Atlas-style agents register as humans in advertiser analytics. It is also true that those agents are working on behalf of actual humans who actually want answers about products. The difference between a human visitor who clicks an ad, scans the landing page for thirty seconds, and bounces, and an Atlas agent that clicks the same ad, parses the same landing page, and reports back to its user, is not a difference in legitimacy. It is a difference in interface. The agent is the audience. Just with a better tool.
Design for the agent, do not try to detect it
The brands that will win the next phase of this transition are the ones that stop trying to detect the agent and start designing for it. A landing page that converts a human can be designed to convert a human via agent. The agent is reading structured data. It rewards pages with clear schema, direct claims in the first hundred words, and machine-readable pricing. It penalizes pages that bury the answer in three scroll-lengths of marketing copy. The work that wins on Atlas is the same work that wins on AI Overviews and on the next wave of agentic shopping interfaces. Make the page legible. Make the answer findable. Cite the primary source. Do the boring work of clean information design.
Performance compresses, owned expands
The strategic implication is uncomfortable for media buyers and welcome for content people. Performance media is going to compress. Branded search budgets that were always partly performative get harder to justify when half the clicks are agent-driven and the agent is just confirming a preference the user had already formed. Owned content and brand strength are going to expand. When the agent is the front door, the brands that show up consistently and credibly across primary sources, third-party reviews and independent media are the ones the agent will surface. That is editorial work, not media-buying work.
The line I keep returning to · every time technology gives audiences a better way to filter noise, the brands that win are the ones who were already producing signal. Cookie deprecation does not hurt brands with strong first-party data. The SERP collapsing into AI Overviews does not hurt brands with deep topical authority. Atlas does not hurt brands the agent already trusts. The question for every marketing organization in 2026 is whether the work you ship would convince the agent. If yes, the next twelve months are going to feel like a tailwind. If no, this is what every quarter is going to feel like from now on.
This week's signal: The agent is not the enemy. The agent is the audience using a better tool. Build the brand the agent already trusts.
The Signal Brief · Mariano De Vitto — Head of Marketing, Barcelona