A new ad channel just launched with no way to measure itself.

At its first Cannes Lions, OpenAI made its pitch for ChatGPT advertising, and the pitch rests almost entirely on one word. The users are “super intentional.” They show up mid-decision, already in the middle of working something out, which supposedly makes the inventory inherently higher-intent than a search result or a social feed. It is a good pitch. It is also a pitch about the audience, and people keep hearing it as a claim about measurement, which it is not.

Two things landed in the same week, and they belong in the same sentence. The first is that ChatGPT ads launched without a way to measure themselves. There is no channel-native, neutral attribution layer, so the channel wired in a third party’s conversions infrastructure to handle cookieless measurement. The second is that this third party is mid-acquisition into an advertising holding company, and at least one rival holding company has already said it will stop using them once the deal closes. So the brand-new channel’s measurement spine is borrowed, and the neutrality of the thing it borrowed is now openly contested.

I am not writing this as a skeptic from the sidelines. We are running the ChatGPT Ads beta. I think being early here is correct, and I will say plainly that the performance layer is real and getting better quickly. That is exactly why the measurement question deserves a clear head rather than either hype or reflexive dismissal.

Intent is not incrementality

Here is the conflation at the center of the whole thing. “These users are highly intentional” is being received as “ads on this surface work especially well.” Those are different claims, and the gap between them is the oldest trap in performance marketing.

If a user arrives already mid-decision, already leaning toward a purchase, and they see an ad, and they convert, the ad gets the credit in any proximity-based measurement. But the high intent is the reason for both the ad’s apparent success and the conversion. The very thing that makes the inventory attractive, that these people are close to buying, is the thing that makes naive attribution overstate the ad’s impact. The more intentional the audience, the more the ad gets to take credit for decisions that were already happening.

This is last-paid-touch over-credit. We have spent years explaining it on search, where branded queries from people who already decided look like conversions the ad produced. ChatGPT ads are a new surface with an unusually intent-rich audience, which means they are an unusually strong setup for exactly this error. The pitch and the trap are the same fact viewed from two angles.

A borrowed spine, and whose interest it serves

Set the intent problem aside and there is still the structural one. A channel that cannot measure itself has to import measurement, and what gets imported carries the assumptions and the incentives of whoever built it. When that provider is being folded into a holding company that buys media, “neutral measurement layer” stops being a safe description. It does not mean the numbers are wrong. It means the numbers now have an owner with a position in the market they are measuring, and an operator should treat that the way they treat any network grading its own performance.

The pattern is familiar if you have watched measurement infrastructure get acquired before. The substrate everyone agreed to trust because it sat outside the buyers and sellers gets pulled inside one of them, and the question quietly shifts from “is this accurate” to “accurate in whose favor.” You do not have to assume bad faith to insist on verification. You just have to remember that a measurement number is a claim, and claims have authors, and the author here just changed.

What being early should and should not mean

So how do you run on a channel you do not yet trust to measure itself? The same way you run on any young surface. You take the platform’s reporting as one input and not as the verdict. You hold out a portion of spend and look for the lift the closed report will never show you. You watch whether your own downstream signal, the conversions you own and define, moves when you move budget into the channel, rather than whether the channel’s own dashboard says it is working. And you stay early, because the cost of waiting until the measurement is mature is handing the learning curve to your competitors.

Being early is an edge. Trusting the channel’s self-report because being early made you enthusiastic is how the edge turns into a bill. Those two can coexist. We are leaning into the channel precisely because it is new and the performance is real, and we are refusing to grade it on its own numbers precisely because no one else’s incentives are aligned with telling us the unflattering version.

The takeaway

A genuinely new ad channel arrived, with a real performance layer and an audience that really is further down the funnel than most. It also arrived without a neutral way to know whether its ads cause the outcomes it reports, leaning on a measurement spine that is being absorbed by a media holding company, and it is being sold on “intent” in a way that invites the oldest over-credit error in the business.

Run on it. Early is right. Just do not let the newness or the intent pitch talk you out of the one discipline that has survived every prior channel: the number a seller reports about its own performance is a starting point for verification, never the end of it. Intent is not incrementality, and the only way to tell the difference is to measure it yourself.