OpenAI opens up ChatGPT ads manager to the U.S. while promising third-party measurement, CPA bidding

Cost-per-action (CPA) bidding is similarly in motion, though Awan declined to say when. 

What he was a lot more forthcoming about was the ads manager itself, now in beta to the U.S. including SMBs, startups, global brands and agency holdcos Dentsu, Omnicom, Publicis and WPP. It marks the biggest expansion yet in the three months since OpenAI launched advertising in ChatGPT, opening the platform from a handful of large test partners to any business that wants in, category restrictions aside.

So far, the company has focused on a “limited set” of advertiser categories, such as household and consumer goods, local services, travel and entertainment, as well as digital products and education — categories which are typically deemed low risk when it comes to regulations or user harm. Its policy page states the team expects to expand “eligible categories over time as our safeguards, review systems, and compliance infrastructure mature.”

It is the same controlled approach OpenAI has applied to building its ads business since the start of the year. Three months ago, advertisers could only pay to reach every thousand users — no click-based bidding, no action-based buying and no self-serve platform to do any of it through. 

Now all three are part of the playbook and OpenAI is intent on keeping a tight grip on it. That wasn’t necessarily obvious when it began signing ad tech partners at the start of the year. Since then, the company has made the boundaries clear: the likes of Adobe, Criteo, Kargo, Pacvue and StackAdapt handle campaign budgeting, bidding and creative while OpenAI’s ads system controls all the delivery decisions. 

That need for control is table stakes for OpenAI. Show the wrong ad in the wrong conversation — something sensitive or something jarring — and the trust that makes ChatGPT’s intent signals valuable in the first place starts to erode. That is why delivery has been kept in-house. Every decision about where an ad appears and whether it is relevant enough to appear at all, runs through OpenAI’s own system. It also means conversation data stats there — advertisers receive aggregated performance metrics not the underlying exchanges that generated them.

It’s a balancing act that will only get more difficult as the platform moves deeper into performance advertising. For now, it remains a test-and-learn exercise so far — advertisers are still working out which budget to pull from and OpenAI is still working out which ones to court. But the direction is clear. The arrival of CPC bidding, the forthcoming addition of CPA, a conversion tracking pixel already live and a conversions API in development all point the same way.

“By adding things like an ads manager, CPC bidding, pixel measurement and CAPI, OpenAI is demonstrating that it understands the basic building blocks that are necessary for advertisers to feel comfortable testing on ChatGPT,” Sonata Insights’ founder and chief analyst Debra Aho Williamson said.

The conversions API in particular gives advertisers a window into what happens after someone engages with an ad such as landing page views, product catalog and page views as well as add-to-cart events. It’s not the full post-click picture that advertisers are used to but it is among the first real evidence OpenAI can offer that its ads do anything at all.

“They [OpenAI] are taking ownership of the creation of that pixel, you can’t do it yourself,” said Jellyfish’s chief solutions officer, media activation, Jai Amin. “They have to send it to you, based on what you’re trying to track.” 

Until a more robust measurement system is in place, the main lever advertisers have to improve performance is the creative itself — currently confined to a single format, a small favicon with text. That constraint reflects a deliberate choice. In every forum, from press briefings to pitch meetings, OpenAI’s ad execs make the same point: ads cannot come at the expense of the user experience. Advertiser ROI comes second.