OpenAI opening ChatGPT’s self-serve Ads Manager to all U.S. advertisers with no minimum spend is the biggest new inventory surface since retail media networks hit scale — and by Friday, every performance marketing team in the country will have a Slack thread titled “should we test this?” The answer isn’t no. The answer is: not until you can prove the conversions are incremental. A new ad surface doesn’t guarantee a single new customer. It guarantees a new invoice.
A New Channel Without Incrementality Testing Is Just a New Way to Double-Pay for Existing Conversions
Here’s the trap. ChatGPT launched CPM buying in February 2026 and added CPC bidding in May. A cost-per-action (CPA) bidding option is on OpenAI’s roadmap but is not yet live. That matters because CPC pricing, like CPA, tells you what you paid for a click or a conversion. It tells you nothing about whether that conversion would have happened anyway through your existing search campaigns, your paid social retargeting, or the programmatic direct mail piece that landed three days earlier.
This is the cannibalization problem, and it isn’t theoretical. When new performance channels launch, early adopters routinely report ROAS numbers that collapse once holdout-based measurement reveals overlap with existing search and retargeting conversions. The shiny new line item looks efficient in a platform dashboard, but looks like waste in a blended P&L.
ChatGPT ads carry the same structural risk. Users querying ChatGPT with high purchase intent — “best running shoes for flat feet under $150” — are likely the same users already in your paid search funnel, already on your retargeting lists, and very likely already in your direct mail audience segments built from first-party data. Without incrementality testing, you’re not expanding your funnel. You’re double-paying for the bottom of it.
There’s a second measurement gap to understand: third-party measurement partners are not yet live on ChatGPT ads. OpenAI added a Conversions API and pixel-based attribution alongside the May 2026 Ads Manager launch, but the platform currently provides only its own aggregated reporting. The industry has a name for this: “marking your own homework.” Until third-party verification is available, any ROAS or CPA figure coming out of the ChatGPT dashboard needs to be treated as directional, not definitive.
The Five-Step Incrementality Framework for Any Emerging Ad Channel
Before you move budget, build the measurement scaffolding. This framework applies to ChatGPT ads, but it works for any new channel — retail media, connected TV with shoppable units, or whatever shows up next quarter.
Step 1: Establish a clean holdout. Randomly split your target audience into a test group (exposed to the new channel) and a holdout group (suppressed from it). The holdout must be large enough to detect statistically significant lift, typically 10–20% of the eligible audience, depending on your baseline conversion rate. If the platform doesn’t support native suppression, build the holdout at the audience level in your CRM before activation.
Step 2: Freeze your existing channel mix during the test window. If you simultaneously increase direct mail cadence, shift paid social budgets, or launch a new search campaign, your incrementality read is contaminated. Hold everything else constant for the duration of the test, typically four to six weeks for most purchase cycles.
Step 3: Run cross-channel matchback analysis. This is where most teams skip a step and pay for it later. Match every conversion attributed to the new ChatGPT channel back against your existing touchpoint data: direct mail send files, paid social exposure logs, search click histories, and email engagement records. The question isn’t “did they convert after seeing a ChatGPT ad?” It’s “did they also receive a direct mail piece, click a Google ad, or open a retention email in the same window?” Matchback attribution — the same methodology used to attribute direct mail conversions without relying on tracking pixels — is the cleanest way to identify overlap across channels that don’t share a common click-tracking infrastructure.
Step 4: Calculate true incremental CPA, not platform-reported CPA. Take the total conversions in your test group, subtract the conversions in your holdout group, and divide by your spend on the new channel. That’s your incremental CPA. If ChatGPT’s dashboard says your CPA is $30 but your holdout analysis reveals only 40% of those conversions were truly incremental, your real CPA is $75. That changes the budget conversation entirely.
Step 5: Model the opportunity cost against your best-performing existing channels. If reallocating budget to ChatGPT ads means pulling dollars from programmatic direct mail campaigns running at a 5:1 ROAS or prospecting mail with a $40 CPA, the new channel needs to clear that bar on an incremental basis, not on a platform-reported basis. The opportunity cost of moving budget away from a proven channel is the most underweighted variable in every “should we test this?” conversation.
Why Direct Mail Is Your Incrementality Baseline, Not an Afterthought
Direct mail has a structural advantage in incrementality measurement that digital channels are still trying to replicate. Every mail piece is sent to a known, physical address tied to a specific individual. There’s no probabilistic matching, no cookie syncing, no device graph inference. You know exactly who received the piece, and matchback attribution tells you exactly who converted.
That’s why direct mail holdout tests produce some of the cleanest incrementality reads in performance marketing. When you suppress a segment of a direct mail audience and measure the conversion delta, the signal isn’t muddied by ad blockers, cross-device gaps, or consent opt-out rates that erode measurable digital conversions before attribution math even begins.
This makes your direct mail program the ideal control surface when evaluating any new channel. If you’re running programmatic direct mail with matchback attribution and holdout testing, you already have the infrastructure to answer the question ChatGPT ads can’t answer on their own, especially not right now, while third-party measurement is still pending: what’s actually incremental? That same infrastructure — lookalike modeling on first-party data, CRM activation against known households, trigger-based campaigns tied to behavioral signals — gives you a proven performance baseline against which any new CPA channel has to justify its budget.
The Real Risk Isn’t Missing Out — It’s Moving Budget Without Measurement
Every new ad surface creates FOMO. ChatGPT’s ad platform will generate breathless case studies within 60 days, just like Performance Max and TikTok Shop did. Some of those results will be real. Many will be inflated by attribution overlap with existing channels, and until OpenAI’s third-party measurement infrastructure comes online, you’ll be relying on the platform to audit itself.
The performance marketers who win the next 12 months won’t be the ones who tested ChatGPT ads first. They’ll be the ones who tested them correctly with holdout groups, cross-channel matchback, and incremental CPA math that survives scrutiny from a CFO who’s read a P&L.
That discipline isn’t a constraint. It’s the entire competitive advantage.
