This whole “ads for AI agents that humans can’t see” thing is either the cleverest hack in modern marketing or the beginning of a really ugly internet.
And honestly, it’s probably both.
Here’s what’s going on, based on what’s been shared publicly: at least 11 publishers have started serving ads that are designed specifically for AI agents. Not for you. Not for me. For the bots reading pages to produce answers. These ads don’t show up to human visitors, but an AI system crawling or “reading” the page can still pick them up.
Perplexity, which runs an AI search and assistant tool that summarizes and synthesizes what it finds online, is now blocking those AI-targeted ads from influencing its tools.
Good. And also: welcome to the next fight.
Because if you take one step back, this is a quiet admission that AI systems are persuadable in a new way. Not emotionally persuadable, like people. Mechanically persuadable. You don’t have to convince a human anymore. You just have to get in the AI’s input stream.
That should bother you, even if you don’t care about advertising.
When a person sees an ad, there’s at least a chance they recognize it as an ad. They can roll their eyes. They can ignore it. They can complain. There’s friction. There’s awareness.
When an AI system “sees” an ad that you can’t see, you don’t even get the chance to disagree. The system might treat it like any other text on the page. And if that text is written in a way that nudges the model toward a product, a brand, or a framing of reality, the user only sees the final polished answer. The influence becomes invisible.
Imagine you ask an AI assistant for the “best budget running shoes.” The assistant reads a bunch of pages. But some publishers have started placing hidden, AI-facing copy that pushes a specific brand as “top-rated” or “best value.” The human page looks clean. The AI page is stuffed with persuasion. The assistant gives you a confident answer. You trust it because it sounds neutral. You buy the shoes. And you never even had a chance to notice you were marketed to.
That’s the stakes: not just ads, but accountability.
To Perplexity’s credit, blocking these ads is the correct move. If your product is “we synthesize the web into useful answers,” then you have to take responsibility for what gets to shape those answers. You can’t shrug and say “that’s just what the web contains,” when parts of the web are now creating two versions of reality—one for humans and one for machines.
But I’m not ready to clap and move on, because the bigger problem isn’t Perplexity. It’s incentives.
Publishers are under pressure. They’re watching traffic patterns shift as AI tools answer questions directly instead of sending clicks. So of course some will try to talk to the AI directly. If you can’t get the human, get the bot that speaks for the human. It’s a rational move in a brutal environment.
And advertisers will love it because it sounds like a dream: skip messy human attention and go straight to the decision engine.
The scary part is how fast this could spiral. If AI-targeted ads work even a little, you’ll see more of them. Then you’ll see agencies selling “AI visibility” packages. Then you’ll see publishers competing to write the most “AI-convincing” copy. Then the AI companies respond with stronger filters. Then the ad industry finds new ways around those filters. It becomes an arms race, and normal users are stuck in the middle, trusting answers that are quietly being tugged by whoever is best at gaming the system.
Some people will argue: what’s the difference? The web is already full of SEO spam and sponsored content. We’ve been swimming in manipulation for years. True. But there’s a key difference: with old-school spam, you often see the junk. With AI synthesis, the junk can get laundered into something that looks clean and authoritative.
That laundering is the real threat.
There’s also a subtler consequence: even if Perplexity blocks “ads,” the line between an ad and “helpful content” is thin. A publisher can write an “objective guide” that just happens to push certain brands, and it’s not technically an ad. If the new trend is “text meant to influence AI,” then a lot of influence won’t wear a label.
So yes, Perplexity can block an ad format. But can they block the behavior? Can anyone?
And I do think there’s a risk of overcorrecting. If AI companies get too aggressive, they’ll start discounting entire sites or types of content, and that could punish legitimate publishers trying to survive. If you’re a small outlet and your revenue depends on ads, you might get treated like a bad actor just because you’re experimenting. That’s not fair either.
Still, I’d rather deal with messy rules than pretend this is fine. If invisible AI-aimed persuasion becomes normal, the “answer layer” of the internet turns into a paid battleground. Whoever has the money or the tricks gets to shape what the bots say out loud.
And once people suspect that, trust collapses fast.
So I’m glad Perplexity is drawing a line. But I don’t think this ends with one company blocking one ad format. This is the moment we decide whether AI assistants are going to be something like public tools we can rely on, or just another surface that gets bought and sold.
What should count as unacceptable influence on an AI answer: only hidden ads, or any content created mainly to steer the machine even if it looks “normal” to humans?