OpenAI阐述广告业务逻辑：将用户信任与用户价值置于广告主之前

Tuesday, November 10th, 2026 – 1:00 am

SHARE:

Ads in ChatGPT, it turns out, are a humanitarian effort.

Right at the top of OpenAI’s charter, the company states that its main mission is to bring the benefits of artificial general intelligence (AGI) – meaning AI that can match or outdo people at intellectual tasks – to all of humanity.

“Asterisk, without killing us,” quipped LUMA Partners CEO Terry Kawaja during an on-stage chat with Benji Shomair, OpenAI’s VP of monetization, at LUMA’s Digital Media Summit in New York City on Monday.

But to achieve this lofty goal depends on getting far more people to use ChatGPT than will ever pay for it.

In other words, Shomair said, for OpenAI to fulfill its mission “actually requires [us] to build an ad business.”

“We believe this technology is so beneficial, so incremental, so powerful,” he said, “that we need to create economic means to make it available to everyone.”

Trust first, advertisers third

But putting ads inside a product that people confide in is a very delicate exercise.

“In my experience, ads or commercial messages haven’t ever been introduced into as intimate an environment as ChatGPT,” Shomair said. “Just think about the things that people are sharing with ChatGPT.”

And so OpenAI is building its ads business around three priorities, in this order: user trust, user value and, last, advertiser value. If you get the first two right, the third will follow, he said.

Which is all well and good as a sentiment. But what do you do when someone asks you to push the boundaries? Which does happen.

Since launching its advertising business less than a year ago, OpenAI has onboarded more than 50 ad tech, measurement and agency partners. “We’re taking a partner-first approach,” Shomair said. “We can’t do it all ourselves; it’s pragmatic.”

Sometimes, however, partners have a different idea of what’s fair game, and OpenAI just won’t go there, Shomair said.

“There are places where I think our partners would like us to do certain things that we just can’t,” he said, “and that’s because of how we think about making decisions.”

Rules of engagement

The decisions that OpenAI makes when it comes to its advertising business are guided by five principles: mission alignment (the whole bring-AGI-to-the-masses thing), answer independence, conversation privacy, choice and control and prioritizing user trust and user experience over revenue.

OpenAI devised its principles before it even started testing ads internally, and they still shape every product decision, Shomair said.

“We reference these principles on a pretty consistent basis,” he said, “and we use this as a way to communicate to our audiences how we’re thinking about trust.”

Even so, ads in AI chats are an easy punchline. Anthropic’s Super Bowl spots for Claude earlier this year gleefully poked fun at the very idea.

In one, a scrawny guy asks a buff trainer how to get a six-pack fast, in response to which the trainer starts hawking height-boosting insoles. In another, a man asks his therapist for advice on how to communicate better with his mom, and the therapist starts plugging a dating site that connects “sensitive cubs” with “roaring cougars.”

Those Anthropic ads had an impact. Claude’s daily download count went from roughly 63,000 in January to a peak of about 225,000 in mid-February, according to Appfigures, and its daily active users rose 11% after the game, per BNP Paribas.

Shomair acknowledged that the ads were funny. But they don’t reflect OpenAI’s approach to monetization in ChatGPT, he said, where the answers “are separate from the ads and not influenced by the ads.”

‘You’ve got to get your hands dirty’

Which raises the question of what an ad even is or should be in a product like ChatGPT.

Shomair, who helped build Meta’s ads business before joining OpenAI last December, said part of the fun of his job is that new modes of consumer engagement force him to think about ads differently.

What, for instance, should advertising look like when consumers talk to a brand’s AI agent? Or when their own AI agent does the shopping for them? Or when two agents are just talking to each other?

Shomair said he doesn’t presume to know the answer. Even people working on the bleeding edge, the ones developing these things, are just experimenting, he said.

“It’s hard to forecast beyond six to 12 months,” he said. “The products I was working with 12 months ago versus the products I’m working on now – it’s like night and day.”

But there’s one thing he’s sure of: Nobody should sit on the sidelines. His advice to advertisers: “You’ve got to get your hands dirty,” he said. “You’ve got to jump in and play with it.”