OpenAI is testing ads in ChatGPT that answer follow-up questions

An ad you scroll past is one thing. An ad that asks what your kitchen looks like is a different product.

What is being tested

OpenAI announced Sponsored Agents on September 16, 2026, describing them as being tested with select advertisers in the United States. After a relevant ad appears, the user can choose to start a conversation with the advertiser’s own AI agent inside ChatGPT rather than loading a page. Digiday describes the entry point as a “Chat with us” button. The agent takes follow-up questions, and OpenAI says users can then follow a link to the business’s website when they are ready to take the next step.

The labeling commitments are specific. OpenAI says the exchange is clearly labeled, kept visually distinct from ChatGPT’s independent answers, and separated from the conversation the user originally started.

Angi confirmed in its own announcement that it is among the first brands in the pilot. Angi co-founder Angie Hicks said homeowners “can go directly from discussing a home project in ChatGPT to connecting with a skilled local pro who can help.” Wayfair’s paid media director told Digiday that Wayfair is participating as well, though OpenAI has not named it. Digiday reported the tests running on September 14, two days before the announcement.

The part labeling does not cover

Home services, the one category OpenAI’s named pilot brand operates in, is an area where a buyer describes a problem in detail before committing, which is exactly the input an agent needs to be useful and exactly the input that makes the conversation valuable to sell.

ChatGPT advertising has expanded steadily through 2026, across the United States, Europe and India. The change here is the format rather than the reach. A banner is static and a link is a destination, but an agent is a participant, and the label establishes that it is paid without establishing what it optimizes for.

The measurable question is whether the separation OpenAI describes holds in the product: whether a sponsored agent’s answers stay confined to its own labeled surface, or begin shaping what the assistant says in the conversation the user actually started.