Advertisers can now build ChatGPT audiences from email addresses, phone numbers and Google Advertising IDs. On October 5, OpenAI also named partners for conversion data, attribution and early incrementality tests. The tools are maturing fast. We think the smart move is a contained test with a clear hypothesis, and a hard look at customer trust before any list is uploaded.
What changed
OpenAI's help center says custom audiences accept emails, phone numbers, hashed versions of both and Google Advertising IDs. An audience needs at least 25,000 matched users before it can be used for targeting, and OpenAI recommends 100,000 or more. Smaller lists can still be used as exclusions, such as current customers.
On measurement, OpenAI named Hightouch, Tealium and LiveRamp to help send conversion data, plus attribution partners such as AppsFlyer, Northbeam and Triple Whale. It called incrementality work early, with Haus, Measured and WorkMagic exploring geo-based experiments. Brand suitability pilots with DoubleVerify and Integral Ad Science are in development.
Why it matters
These are the tools advertisers expect from a serious channel, and they lower the cost of a fair test. OpenAI says more than 100,000 advertisers have run ads, and it announced 26 new markets on October 21. AdExchanger reports advertisers want stronger measurement before they commit more budget.
That caution is healthy. The results OpenAI has shared so far are partner-reported case studies, not independent benchmarks. And a chat assistant feels more personal than a search page, so how people react to seeing a brand there may differ. Both points argue for testing with holdouts before scaling.
Four things to consider
1. Start with a hypothesis
Write down what you expect ChatGPT ads to do that your other channels don't, such as reaching new customers at a lower cost. Set the success metric and the budget cap before launch. If the test can't prove or disprove the idea, it isn't ready.
2. Build in a holdout
Attributed conversions show who clicked, not who would have bought anyway. Use a geo or audience holdout so you can measure lift. OpenAI's own incrementality tools are early, so plan your own control group.
3. Weigh trust before uploading lists
Check that your privacy notice and consent terms cover sharing customer data for ad targeting on this platform. Ask how a customer would feel seeing your ad in a private conversation. Starting with exclusions, such as current customers, can be a lower-risk first step.
4. Keep the test separate
Fund the test from a defined test budget rather than pulling from channels that already work. Compare results against your existing search and social benchmarks. Scale only when lift holds up over more than one period.
Our view
ChatGPT ads are worth a disciplined test for many brands, especially those whose customers already use AI tools to research purchases. They aren't yet a must-buy. A clear hypothesis, a real holdout and a careful approach to customer data will tell you more than early case studies will.
Questions to ask your team
What would ChatGPT ads need to prove for us to scale them?
Do our consent terms cover uploading customer lists here?
How will we measure lift, not just attributed conversions?
Which budget funds the test, and when do we decide?
From Insight to Impact
Want help planning a fair test?
Strategy Butler's Channel Assessment and Campaign Strategy Framework help you decide whether a new channel fits, and how to test it with clear goals and controls. See what's included, and the sample pages behind them.
Sources
OpenAI: New ChatGPT ads format and measurement partners, October 5, 2026OpenAI Help Center: Set up custom audiences for your campaign, accessed October 2026AdExchanger: ChatGPT Ads Is Expanding To New Markets And Exploring New Ad Formats, October 21, 2026Customer-list targeting was also noted in AdExchanger's October 16, 2026 roundup.
