Now open in Europe · Incrementality test for ChatGPT Ads
We have run ChatGPT Ads in the US since the channel’s first weeks. Long enough to know what works and what does not. The rest of the work is measurement. In a holdout test, part of the audience sees no ads, and the difference is what the channel really added. The accounts behind that spend $50k to $1M a month.
Measuring it on your own account is what this page is for, and your category is still open.
The feasibility check is free and comes before any quote. It is built for brands that can put €15,000 or more into this channel over six weeks.
Short on time? The whole page in four lines
No slides. Bring your numbers and leave knowing whether a test on your account can be read.
About us
$50k – $1M
per account, per month
US accounts under management, judged on contribution margin, not on what the platform reports about itself.
Since launch
the channel's first weeks
Long enough to know which account structures work.
Not measured yet
the honest state of this channel
Being the first in your category with a holdout result is the whole offer.
Most of the brands we have tested found the same thing. For all the novelty, this channel did not beat the cost per acquisition of what they already ran. Sometimes it came close. Sometimes it was worse.
That is not an argument against the channel. It is an argument against buying it because it is new.
Your category, your margins and your consideration cycle decide the answer. The only way to get yours is to measure it on your own account, before the budget decides for you.
One detail to know before you budget: the ads only show to people on the free and cheaper ChatGPT plans. Subscribers on the top tiers never see them.
Whether that helps you or hurts you depends on who buys from you. It is the kind of thing that decides a channel long before the ads themselves do.
Side by side with the channels you already buy. Everything here comes from live US accounts, not from the documentation.
| Category | Google & Meta ads | ChatGPT Ads |
|---|---|---|
| Where the ad sits | Google & MetaBeside a list of links, or in a social feed. | ChatGPT AdsUnder the model's answer, while someone is still narrowing a choice. |
| What you target | Google & MetaKeywords, demographics, audiences built from cookies. | ChatGPT AdsThe intent inside the question and the answer. No cookies involved. |
| What people type | Google & MetaTwo or three words. Often ambiguous. | ChatGPT AdsWhole questions, with the constraint and the budget already in them. |
| The auction | Google & MetaMature. Everyone in your category has been bidding for years. | ChatGPT AdsEarly, with few bidders. That is the one real reason to look now. |
| Cost per conversion, as the platform reports it | Google & MetaYour current cost per conversion. | ChatGPT AdsIn the US accounts we run, often around half of what Google Ads reports. |
| Cost per new customer, measured against a holdout | Google & MetaYour current cost per new customer. | ChatGPT AdsAbout the same as Google Ads. Sometimes worse. |
The last two rows are the same accounts over the same weeks. One is what the channel says about itself. The other is what it actually added. Everything else on this page exists because those two numbers disagree.
Three things that change how you run it
01
The system reads the question and the answer, so you write context hints as whole phrases: “Comfortable sandals for a holiday under €50”, not “sandals”. Get this wrong and the click still arrives cheaply, which is exactly what makes the mistake expensive to spot.
02
CPM for reach, CPC for traffic, conversion mode for what happens after the click, billed on a qualifying click rather than on the conversion itself. The model you pick decides what the system learns.
03
A campaign is a business goal, an ad group is an intent cluster, an ad is a creative variant. Putting everything into one group saves you an hour now and costs you a month of learning later.
The platform will report conversions from your first week. Two things go wrong when a channel reports its own contribution, and both have played out on channels where the money is already big.
01
Attributed conversions come from the same system that sold you the impression, and whichever tool has the longest attribution window claims the order. Adding a channel without a holdout usually moves credit around inside your reporting rather than adding revenue to the business.
What this costs: On an established channel, the same budget added real revenue during a promotion and close to none outside it. The platform reported both periods as broadly similar.
02
Right now the channel is small and separable, so a holdout is cheap to run. The platform is also subsidising traffic to build the marketplace, and that will not last. Once the channel is blended into your mix, isolating its effect takes far more spend and a far longer window.
What this costs: Wait a quarter and the same answer costs several times more.
A free feasibility check, then six weeks live, roughly eight weeks end to end, to find out whether this channel does anything for you. Keep going and you move onto a monthly fee, set by what you spend on ChatGPT Ads. It starts only if you decide to keep the channel after the report.
The prices below are founding-client prices, at half the US price. You get them in exchange for two things: a written case study when the test is done, and permission to use your logo. Decline either and the US price applies.
Six-week holdout test
€3,800 US price
€1,900
Six weeks, paid once
Six weeks to find out whether this channel moves anything for you. A holdout sized against your real conversion volume, run properly, and measured against your baseline.
What that price assumes
Monthly fee, set by your monthly ChatGPT Ads spend
Up to €20k
€5,700 US price
€2,850
per month*
*starts after test, if you keep the channel
Typical for
€200k–400k a month across all channels
What grows with the tier
€20k to €45k
€9,300 US price
€4,650
per month*
*starts after test, if you keep the channel
Typical for
€400k–900k a month across all channels
What grows with the tier
€45k to €90k
€14,800 US price
€7,400
per month*
*starts after test, if you keep the channel
Typical for
€900k–1.8m a month across all channels
What grows with the tier
Above €90k
US price on request
Quoted
scoped to your account
Typical for
€1.8m+ a month across all channels
What grows with the tier
The first is whether a test can produce a readable answer at all. The second is whether the channel can pay for itself once it does. You can clear one and fail the other.
Condition one
A holdout can only separate a real lift from random noise with several hundred conversions across the test. In practice that means at least €15,000 of ChatGPT Ads media over six weeks, about 330 conversions at a €45 cost per acquisition. That figure is an estimate. The feasibility check replaces it with your real numbers.
Total budget does not help. A brand spending €60k a month that puts €3k behind this channel is in the same position as one spending €10k.
Condition two
The feasibility check brings the other half: what a conversion is likely to cost on this channel, from your own numbers rather than a benchmark. If your figure sits well above that cost, the channel can pay for itself. If it sits below, you hear that instead of a quote.
If we size the test wrong and it fails to collect enough data, we extend the window at no extra fee. That is our error, not yours. If the test has enough data and shows no lift, that is the answer, in writing, with the same detail as a positive result. A test that can only return good news is not a test.
Tell us your spend and your conversion volume. Within one working day you hear whether a test on your account can be read, and if it cannot, why.
Would rather talk it through first? Book 30 minutes. No slides, no quote on the call.
Not ready to spend? Run it yourself first
Leave your email and the ChatGPT Ads dashboard replica opens, with a ten step tour of what each part does. Tick the box and six short tips follow, one every two days, from people who have already spent on the channel.
