AI Agents Are Buying Ads Like It's 1995 TV, and CTV Already Showed Us Where That Goes

A new report shows AI ad-buying agents skip most of the auction and just trust the price. CTV tried that first. The fraud numbers tell you how it ends.

DataBeat put out a report in June. AI agents buying ads take part in 86% fewer auctions than normal buyers. And they pay 13.4% less per thousand impressions.

At first that sounds like a win. Agents are smarter, so they get better prices. That's not what's actually happening.

Fewer auctions means less checking, not more skill

Programmatic advertising was built to check every single ad sale. One impression, one bid, one price, all verified in the open. That's the whole point of the system.

AI agents are skipping most of that. Instead of bidding on each impression, they're striking bulk deals ahead of time. Fixed price, fixed volume, agreed in advance. No auction. No per-impression check.

This is exactly how TV ads used to get sold. A buyer agrees to a price for a chunk of audience, months ahead, and just trusts the number. Nobody checks each individual ad. The whole system runs on trust between two parties.

Programmatic advertising was built to fix that problem. Now the AI agents buying programmatic ads are quietly rebuilding the old system, just with software instead of a phone call.

The industry already admitted the trust problem is real

In July, IAB Tech Lab released new rules for these AI buying agents. One of the new rules is a "pricing provenance field." It exists so an AI agent can't just make up a price when it doesn't have real market data to check against.

Think about what that means. In a normal auction, a price can't be faked, because other buyers are bidding against it in real time. That keeps everyone honest automatically.

Take away the auction, and prices can't check themselves anymore. So now the price has to be recorded and trusted instead of proven. That's not a small technical detail. That's the entire model changing from "we can verify this" to "we're trusting this."

CTV already ran this exact experiment

This isn't a guess about what happens next. It already happened, on streaming TV.

Streaming ads moved onto programmatic systems, but a lot of the old TV-style trust stuck around too. Nobody was checking each ad play as closely as they should have.

Here's what happened. A study this year found only 40% of CTV bid requests carry real, useful data about what show the ad ran next to. Another study found 43% of CTV buyers aren't sure where their ads actually played. And DoubleVerify found that 34 out of every 100 CTV ad plays it checked ran somewhere other than the streaming content it was supposed to run next to. Fraud on CTV is up 140% year over year.

That's what happens when you build a system meant to check everything, and then quietly stop checking.

Why this matters for your ad spend

You don't need an opinion on whether AI buying agents are good or bad to see the risk here.

If AI agents are striking more bulk, pre-agreed deals and skipping fewer auctions, they're also skipping the per-impression checks that used to catch fraud and waste. CTV already showed what fills that gap: fake impressions, wrong placements, and buyers who don't know where their money actually went.

The same logic applies to conversion tracking. A platform reporting a number, and that number never getting independently checked, is exactly how bad data survives. It doesn't matter if it's an ad auction price or a conversion count. If nobody's verifying it, you're trusting it blind.

The smart move isn't to avoid AI buying agents. It's to ask, for anything they touch, "who's actually checking this number, and how?" If the honest answer is "nobody, we're just trusting it," that's worth knowing before your budget finds out the hard way.

---

DataCops exists because trusting a platform's number isn't the same as verifying it. We check conversions before they reach Google, Meta, LinkedIn or TikTok, so what you're spending against is real, not just reported.