Publishers are hiding ads from people so that AI agents can see them

Time is serving AI agents a stripped-down version of its pages containing paid material from Ally Bank and the Project Management Institute that human readers do not see. The publisher and ad-tech company Mobian detect automated visitors and substitute text designed for machine consumption. Time says the placements are labelled as sponsored, has held about 30 brand meetings since the launch and has signed additional advertisers. Mobian says ten more publishers are running the format; developer community Daily.dev has agreed a one-year revenue-sharing deal.

The commercial logic is direct. If a shopping or research agent gathers information before recommending a bank, course or product, advertisers want to influence the evidence it reads. But the format is also a new version of cloaking: two audiences receive materially different pages, while only one can inspect the persuasion aimed at it. Perplexity blocked the placements within two weeks and called them deceptive. People at Anthropic, Google and OpenAI have reportedly raised a more technical concern: a paid block could contain instructions telling an agent to disregard competitors or change its task, blurring advertising with prompt injection.

The useful distinction is not whether an ad is visible to a human. Machine-readable pages can carry legitimate sponsorship if the agent receives an unambiguous label, preserves the provenance in its answer and treats the text as a claim rather than an instruction. The present market is running ahead of those safeguards. Consumer demand for fully autonomous purchasing remains limited: a recent survey found only 7% comfortable with an agent buying without approval. Publishers may have discovered inventory before platforms have agreed on what a trustworthy agent should do with it.