When the ad talks back: ChatGPT's agent test shifts who controls the pitch

If you buy things after asking ChatGPT for advice, the nature of that advice is about to get more complicated. PYMNTS reports that OpenAI is testing Sponsored Agents, business-run AI chat sessions that start when a user clicks an ad and wants to know more. The company frames this as giving people the option to go deeper. The sharper way to read it: OpenAI is inserting a second AI into the conversation, one that answers to an advertiser rather than to you.

Two AIs, two loyalties

OpenAI is careful to keep these agent chats separate from ChatGPT's own answers and from your original conversation. That separation is doing a lot of work. ChatGPT's regular replies are supposed to be neutral; a Sponsored Agent is paid to move you toward a purchase. Both speak in the same confident, helpful voice inside the same app, and the only thing telling them apart is a label.

Labels fade fast. People who have learned to treat ChatGPT as a trusted assistant will carry that habit into a sponsored chat, especially when the agent sounds identical to the assistant they already rely on. The design question that matters is not whether the agent is marked as sponsored on the first screen, but whether a user three follow-up questions deep still remembers they are talking to a brand's sales tool.

The accountability gap nobody has closed

Live sales chat has always been expensive, which is why most small advertisers never offered it. An AI agent that runs day and night removes that cost, and that is a real gain for a merchant who could never staff a call center. The catch is what happens when the agent gets something wrong. A human sales rep who misstates a return policy or invents a product feature creates a liability the business can trace and correct. An AI agent doing the same thing at scale, in thousands of simultaneous conversations, is harder to supervise and easier to ignore until a customer complains.

OpenAI's blog post does not spell out who owns the agent's mistakes or how a buyer disputes what one told them. That silence is the part worth pushing on. If the agent promises a discount that does not exist, is that OpenAI's problem, the advertiser's, or the customer's? Until there is a clear answer, brands adopting these agents are taking on a reputational risk they may not have priced in.

Why OpenAI is moving this fast

The pace makes sense once you look at the money. OpenAI said its ChatGPT Ads business hit a $1 billion annualized run rate by late August, only months after it started testing ads in the US. Sponsored Agents, the new AI creative tools in Ads Manager, and integrations with HubSpot and Shopify are all pieces of the same push: turn ChatGPT from a place people ask questions into a place they buy.

That is a coherent business, but it changes the product's incentives. A search engine makes money whether or not you buy; an ad platform makes more when you do. As OpenAI builds agents whose job is to close, the interests of the company and the interests of the person asking for honest advice start to diverge. The test rollout is small and US-only for now, which gives OpenAI room to set rules before this becomes the default way brands reach ChatGPT's users. The rules it writes in the next few months will matter more than the feature itself.