Some do, and the list is short because the channel is new. OpenAI started testing ads in ChatGPT with a small group of brands in February 2026 and opened a beta self-serve Ads Manager to US advertisers on May 5, with cost-per-click bidding, a measurement pixel, and a Conversions API. Attribution tools are adding it the way they add any ad platform, by pulling spend from the platform and matching it against orders. ThoughtMetric, which sponsors this blog, added a ChatGPT Ads integration in late August that puts OpenAI spend next to its own order attribution. The harder question is how much the numbers mean yet.
What does a ChatGPT ads integration give you?
The same two things any ad integration does. First, spend and delivery data from OpenAI, so cost sits next to revenue in your reporting. Second, order matching, where the tool ties purchases to specific ChatGPT campaigns and ads using tracking parameters on the landing page URL. OpenAI supports static tracking parameters like UTMs on ad destinations, which is what makes independent measurement possible at all.
ThoughtMetric reports ChatGPT Ads at the campaign, ad group, and ad level, with spend, clicks, CPC, ROAS, CAC, and new versus returning customers. Connecting it takes an API key from your ads account and no changes to your store, and adding ThoughtMetric’s tracking parameter to the ads improves the match. The reason to run an independent tool here is the same as on Meta. OpenAI’s reporting is aggregated and self-graded, and your store’s order log is not.
Why is ChatGPT ads attribution harder than it looks?
Volume is the first problem. Most brands are running small test budgets, and a test that produces eleven orders cannot tell you much about ROAS no matter how good the tracking is. Give it enough spend and enough time to produce a number that would survive a few orders landing differently.
Behavior is the second. Ads appear below ChatGPT’s responses, in the middle of someone researching. Plenty of people who see a relevant product in that context will not click right away. They finish the conversation, search your brand name later, and buy through Google or a direct visit. Click attribution gives that order to search. It is the same pattern I have been watching with organic AI referrals, which I wrote about in the AI traffic I’m seeing in my attribution reports, and paid placements will not escape it.
The third is that the platform is still changing month to month. Custom audience targeting arrived in May, and conversion-optimized campaigns started rolling out in June. A result from your July test may not hold under the October product.
Is this the same as connecting ChatGPT to my analytics?
No, and the names make them easy to confuse. A ChatGPT Ads integration treats ChatGPT as an ad channel you pay for and measure. An AI connector lets you ask ChatGPT or Claude questions about your existing data in plain language. ThoughtMetric has both, and they do unrelated jobs.
How should you run a first test?
Treat it like any new channel with a fixed budget. Put tracking parameters on every ad before launch, decide in advance how many orders you need before you judge it, and evaluate it on new-customer orders in your own attribution rather than the ROAS OpenAI reports. Add ChatGPT as an answer in your post-purchase survey so you can see how many buyers credit it without a matching click. If the survey number runs well ahead of attributed orders, the channel is influencing more than its clicks show.
What should the numbers look like?
Expect attributed ROAS in your own tool to come in below what OpenAI reports, the same way it does for Meta. Expect your survey mentions to run ahead of your attributed orders, because of the research-then-search pattern above. And expect a high share of new customers, since people asking ChatGPT for product recommendations are often shopping a category rather than a brand. If ChatGPT is mostly reaching your existing customers, something about the targeting is off.
A reasonable first test runs at least four weeks, so you see a full cycle of delayed purchases, and holds budget steady throughout. Changing bids halfway through makes the result impossible to read.
Common questions
Do I still need OpenAI’s pixel if I use an attribution tool? Yes, if you want conversion-optimized campaigns to work. The pixel and Conversions API feed OpenAI’s optimization, and your attribution tool measures the result independently. Those are separate jobs.
Are ChatGPT ads available outside the US? The self-serve beta launched for US advertisers. Check OpenAI’s Ads Manager for current eligibility before you plan a test elsewhere.
Should a small store bother yet? Only with money it can afford to learn with. The channel is promising and early, and early channels reward brands that measure carefully more than brands that spend first.
Leave a comment