OpenAI tests sponsored sales agents in ChatGPT as its ad business already nears $1 billion a year

By Zendoric · September 17, 2026.

OpenAI has begun testing "Sponsored Agents": artificial intelligence agents paid for by a company that hold a conversation with the user who clicks on an ad inside ChatGPT, as the company itself announced on its blog on Wednesday. The test, for now, is limited to a select group of advertisers in the United States.

The mechanism, as OpenAI describes it, works like this: the user sees a relevant ad and can choose to open a conversation, clearly labeled as sponsored, with that company's agent. There they can explain what they need, ask follow-up questions and, if convinced, follow a link to the advertiser's website. OpenAI insists that this conversation stays separate from both the original ChatGPT thread and the assistant's "independent" answers, so as not to mix the two.

The new feature arrives alongside more pieces for the same business: ChatGPT Work, which allows ads to be created by writing instructions in natural language; new creative generation tools inside Ads Manager, its ad management platform; and integrations with HubSpot and Shopify to bring ChatGPT into other tools companies already use. OpenAI sums it up as "reimagining both sides of advertising": how people interact with brands in an AI-native environment and how brands create and manage their ads with AI.

The timing is revealing. OpenAI announced in January 2026 that it would bring advertising to ChatGPT in the United States and that testing would begin "in weeks." According to figures the company itself made public, by August 31 that business had already reached an annualized revenue run rate of $1 billion: an annualized projection based on its revenue pace at that moment, not a full year already booked, but one that gives a sense of how fast it is taking off. OpenAI says it already has "tens of thousands" of advertisers and, since that same date, allows ads to be bought directly via Ads Manager in India, Europe, the Middle East and North Africa.

Our reading is that OpenAI is executing, at unusual speed, the classic internet monetization playbook: identify a massive user base, turn its attention into advertising revenue and then build self-service tools so any advertiser can sign up on their own. What changes is the basic unit of the ad. It is no longer a link or a banner the user glances at and dismisses: it is a conversational agent, with the same fluency and apparent closeness as ChatGPT itself, explicitly designed to persuade. The labeling and the separation from the main thread that OpenAI promises are necessary, but they do not remove the underlying problem: the line between "the assistant is helping you decide" and "a brand is selling to you" becomes much blurrier when both speak in the same tone and inside the same window.

In the short term, this opens up concrete risks worth naming plainly: regulatory pressure over advertising disclosure rules, the possibility of designs that stretch the conversation out to maximize conversion rather than real usefulness for the user, and an underlying conflict of incentives for OpenAI, which needs to present itself as a neutral assistant while depending more and more on advertising revenue to fund the cost of its models. That the business has grown so fast (from zero to $1 billion annualized in eight months) shows, above all, the extent to which advertisers already see the AI conversation as a product-discovery channel as valuable as Google's search engine was in its day.

As industry context, OpenAI is not alone in this race: Google, Meta and Perplexity are competing to turn their assistants into the new entry point to commerce, and the integrations with Shopify and HubSpot point to the real prize, which is not the ad impression but control of the moment of transaction. Whoever wins that layer of commercial "plumbing" inside the AI conversation will have far greater business leverage than whoever simply sells the smartest model.

In the long term, a solid advertising business is also, for OpenAI, a way to make frontier AI cheaper or even free for hundreds of millions of people who would not pay for a subscription, which fits the abundance we defend as a horizon: more people with powerful tools, with no price barrier. But that promise only holds if the sponsored conversation is governed with the same rigor as a regulated television ad, and not with the laxity of any old banner; otherwise, the short-term cost in trust may end up outweighing the revenue it generates.