OpenAI starts U.S. test of Sponsored Agent chats in ChatGPT

OpenAI has begun a limited U.S. test that lets some advertisers open separate, business-sponsored conversations inside ChatGPT after a user clicks an ad, according to the company’s product announcement and an independent launch report.

The test started on September 16, 2026, with select advertisers in the United States. Under the setup described by OpenAI, ChatGPT first gives its own answer to a user’s question. A paid placement may then appear, and only if the user clicks it can that person enter a clearly labeled conversation with a sponsored agent tied to the business.

The design is meant to keep standard ChatGPT responses apart from advertiser-backed exchanges. The source material says the test does not give advertisers control over ordinary answers. PPC Land’s independent launch report also confirmed the September 16 date, the limited U.S. rollout and the separate post-click conversation.

A three-step handoff

The product flow described in the source has three stages: ChatGPT generates an independent response, a relevant ad may appear, and a user who selects that ad can move into a separate sponsored conversation.

Within that sponsored exchange, users can ask follow-up questions about the promoted product or service and can follow a link to the business’s website. OpenAI’s example involves a dining-table ad, where a shopper might ask whether the table fits a room, how many people it seats or how to care for its finish.

The source says the label and the separate chat matter because the sponsored agent speaks for a business and is not presented as a neutral continuation of ChatGPT’s advice. But it also says that this is an intended product boundary, not an independently audited guarantee that the separation cannot be crossed.

Campaign tools sit on the advertiser side

The article says OpenAI’s Ads Manager tools affect campaign creation, ad creative and the sponsored interaction, rather than the independent answer ChatGPT shows before a click.

An Ads Manager plugin in ChatGPT Work lets advertisers use natural-language prompts to create, update and analyze campaigns, turn a website or brief into campaign material, review performance and ask for recommendations. It can also suggest copy and imagery from a landing page and campaign objective, with advertisers able to review and edit those suggestions before deciding whether to add them to a campaign.

A separate opt-in feature can adapt existing headlines and descriptions to the context of a conversation and translate the copy into the user’s preferred language.

The source says OpenAI has not explained the technical enforcement behind the separation between ordinary answers and sponsored conversations.

Shopify and HubSpot link into the ad system

Shopify and HubSpot are described in the source as supporting the advertiser layer rather than ChatGPT’s standard response generation.

HubSpot, identified as the first CRM partner, allows businesses to connect a ChatGPT Ads account, create ads, track performance and follow up on leads using HubSpot context.

Shopify is the first ecommerce partner. U.S.-based merchants can use the ChatGPT Ads app to create and manage campaigns, pull products from Shopify Catalog and track performance. International availability is scheduled to begin September 23 in markets where ChatGPT Ads is offered, but the source says that timing applies to the Shopify app, not to any expansion of the U.S.-only Sponsored Agents test.

Open questions remain

The source says several points have not been disclosed, including what conversation data advertisers receive, which model runs the sponsored agent, how the format is priced, how the separation from the original chat is enforced and what measurements will be available for the interaction.

It also points to a paper submitted on September 16 that audited AI-generated product recommendations. That study built a dataset of 2,528 real commercial-advice queries and evaluated 1,536 product-query responses from ChatGPT, Gemini and Google Search AI Overviews. The researchers found that recommendations could change across repeated requests and that consumer interfaces differed substantially from corresponding APIs. The study did not examine Sponsored Agents and does not show that advertising altered ChatGPT’s independent answers.

The source says its narrower relevance is methodological: claims about independence need repeated observation in the consumer-facing product, rather than inference from a single answer or an API result.

For now, what is established is the product’s intended structure: ChatGPT’s independent answer comes first, an ad may offer a labeled entry point, and a user can then choose to enter a separate conversation sponsored by the business. The source says whether that separation holds reliably at scale will require more technical disclosure and independent testing.

Read the source