OpenAI Chief Revenue Officer Denise Dresser used her Cannes conversation with CNBC's Julia Boorstin to walk through the company's young advertising business inside ChatGPT, and her framing was deliberately narrow. Ads in ChatGPT, she argued, only work if they are useful. The pitch is not interruption at scale but relevance inside a conversation a user chose to start.
It is a clean story, and worth pressing on. The same qualities that make a chatbot a good advertising surface, that users arrive with intent and treat the answer as a trusted recommendation, are exactly what make advertising inside it delicate. An ad that is genuinely indistinguishable from a helpful answer is, by design, harder to recognize as an ad. The session was strong on philosophy and lighter on the mechanics that will decide whether this is good for users: how ads are labeled, how brands get ranked, and what data informs the targeting.
Why OpenAI Is Selling Ads at All
Dresser tied advertising back to the company's mission of bringing intelligence to all of humanity, with ads positioned as one way to scale the business and make that intelligence more accessible to more people. She was clear that OpenAI moved slowly and on purpose before putting ads into the product, wanting to protect the user experience first.
The early traction, as she described it, is encouraging. The product is roughly 19 weeks old, and already works with thousands of customers across seven countries, with more expansion planned. She framed advertising as a piece of OpenAI's broader revenue picture, noting the company has talked about enterprise making up a majority of revenue by year end, while declining to give financial projections for ads specifically.
The mission framing is where a skeptic would start. Casting ads as a way to make intelligence more accessible is a familiar move; Google justified search ads in similar terms, and over two decades the commercial layer steadily reshaped what the product showed and rewarded. The unanswered question is what happens when advertiser incentives and user interests diverge, which they eventually will, and which side the system favors when they do.
The Test Is Usefulness, Not Attention
The defining feature of a ChatGPT ad, in Dresser's telling, is that it has to be helpful to the conversation the user is already having. That makes it feel different from a typical display or social ad: less attention-grabbing, more intelligent about the product in question, and conversational because it lives inside the dialogue rather than around it.
She pointed to one early signal to back this up. OpenAI has seen roughly a 50% reduction in users closing an ad, which the team reads as a sign of improving relevance, since closing an ad is the clearest signal that it was not useful. The form factor, she repeated, is built around usefulness, which she argued is good for the consumer and the brand at the same time.
That metric deserves a second look. Fewer ad-closes can mean the ads got more relevant, but it can equally mean they got harder to distinguish from the assistant's own output, so users stop noticing them as ads worth dismissing. The two are difficult to tell apart from the outside, and Dresser did not address how OpenAI separates them. "Useful" is also a slippery standard in a context where the product's authority comes from feeling neutral. There is a real difference between an ad that helps a user finish their task and one that quietly nudges the answer toward whoever paid, and the session did not draw that line.
Where It Fits in the Media Plan
Boorstin pressed on the practical question agencies keep asking: whose budget does this come from. Dresser's answer was that, for now, ads in ChatGPT are largely experimental, funded out of test-and-learn money rather than reallocated from social or streaming, though she suggested it could mature into its own category over time. She resisted framing it as a direct competitor to existing channels, describing it instead as simply another way for brands to reach the right person with the right message.
Three Ideas to Take Away
- Usefulness is the stated format, and the open question. A ChatGPT ad is meant to help the conversation, but the same design makes it harder to recognize as an ad, so ad-closes alone are a thin proxy for quality.
- Intent makes the placement powerful, and that is the risk. Interactive tools like a virtual try-on can genuinely serve a user, but a paid recommendation inside a trusted answer also carries more persuasive weight than any banner.
- It is early, so judge the mechanics, not the mission. This is test-and-learn budget today; the things that will matter are labeling, ranking and data practices, none of which the session pinned down.