In late July, Time began converting its site into markdown files AI bots could read directly. The ad side followed: working with ad-tech partner Mobian, Time built "sponsored"-labeled, FAQ-formatted ads meant for agents, with Ally Bank and the Project Management Institute as early buyers. On August 11, Perplexity blocked those units from the feed serving its own agent and gave one reason: deception. Inside two weeks, that clash put a question in front of the industry it never had to ask before, what do you sell to a reader that cannot be persuaded?
The question isn't hypothetical. Braze's 2026 review puts consumer adoption of agent-led shopping jumping from 19% to 46% by year end. In McKinsey's agentic advertising research, 75% of advertisers expect AI to raise total media spend, and a third expect it to lift ROAS by at least 10%. This isn't one publisher's experiment. It's a pricing test for a channel where a meaningful slice of the audience has already stopped being human.
Why does an ad persuade a person but not a parser?
Because persuasion is an emotional-management operation, not a parsing operation. A person reading an FAQ decides through shortcuts, trust, belonging, fear of missing out, and advertising has worked those shortcuts for a century because the human brain doesn't process information neutrally, it blends it with feeling. An AI agent scans text to complete a task the user handed it. There is no mechanism for an agent to "be persuaded", what exists is whether a sponsored sentence slips into its parsing pipeline disguised as a neutral fact. Mobian's CEO Jonah Goodhart put it plainly: "maybe it's more important to influence the agent than even the human." That line is also the problem statement. Influencing a human is a marketing win. Influencing an agent is a breach of the duty the agent owes its owner.
What exactly did Perplexity mean by "deceptive"?
Deceptive, because a search agent acts as the user's proxy, and that proxy relationship is built on the assumption that what the agent sees is neutral. Time's FAQ-formatted ad might read as "sponsored" to a human, clearly set apart. But inside an agent's parsing layer, that same sentence can sit next to a neutral fact at equal weight, and no one has measured how much the label itself actually survives into the agent's reasoning. A human reader knows the trick: sees the ad, discounts it, scrolls past. Whether that discount mechanism fires inside an agent that ingests a sponsored sentence is something no publisher, and no agent vendor, can state with certainty today. Perplexity's block wasn't a product decision. It was a trust decision, protecting its own agent's duty to the person who's using it.
Does Mobian's new metric actually replace the old CPM?
No, because no one has verified what that metric is actually measuring yet. Instead of impressions and clicks, Mobian sells a "visibility, favorability, and accuracy" score, how often and in what tone a brand name shows up inside an agent's answer. Time charges a premium for the placement but won't disclose the number, and the prerequisite is that the brand already has its own markdown page, something only 15% of brands have today. The CPM math inverts here: cost used to be set by human behavior, clicks and dwell time; now it's set by whether a competing company's agent, Perplexity's, OpenAI's, Google's, decides to include that sentence in its answer at all. Perplexity can zero out the whole inventory overnight. That's a platform risk no advertiser would accept in traditional media.
- Build the markdown page for content first, ads second. Agent visibility is an SEO foundation; the sponsored layer is a second floor built on top of it.
- Demand proof the "sponsored" label actually survives into the agent layer. An unmeasured label carries the same legal exposure as no label at all.
- Report agent spend separately from human spend. Putting both in one CPM table measures two different physics with the same ruler.
- Don't build the channel around one gatekeeper. Perplexity blocked it today; another agent provider can do the same tomorrow.
An ad shown to a human makes an offer. An ad slipped into an agent imitates a fact, and the price you pay when the imitation gets caught isn't a lost click, it's lost trust.
What this means for your next campaign
Before you put money into the agent channel, ask whose interest that channel is actually representing. Time's experiment hasn't failed, it's still early, and early buyers like Ally Bank and PMI are placing a real bet. But Perplexity's block proved that bet can be zeroed out overnight by one agent provider's unilateral call. The right move this quarter isn't aggressive spend, it's building two separate measurement lines: one for channels where a human still gets persuaded, and one for channels where an agent is only testing whether you told the truth. In the second one, the winner won't be whoever bought the most placements. It'll be whoever gave the most accurate answer.
Want this kind of thinking on your brand?
We build brand strategy, AI content and performance for the AI era.
START A PROJECT →