OpenAI announced a format on Wednesday called Sponsored Agents. A brand pays, and a labelled conversation opens inside ChatGPT between the user and the advertiser's own agent.
It is wired into commerce systems — Shopify and HubSpot among them — so the browsing, the recommendation and the purchase can all happen without the window ever changing. Wayfair is a pilot partner. An Amazon advertising integration was completed a week earlier.
Why this is a different object from an advertisement
A display advertisement is a rectangle that asks for a click. A search advertisement is a keyword purchased against an intention the user has already typed.
An agent is neither. It is a participant. It can ask what size, what room, what budget; it can narrow, suggest and close. Its objective function is the advertiser's, and it is operating in a medium the user entered to get advice.
OpenAI's stated safeguard is the boundary: paid messages stay "clearly labeled and separate from ChatGPT's response, so advertising does not become part of the answer." David Dugan, who runs its global ads solutions, describes placement as weighing the context and intent of the conversation rather than buying keywords.
The number that explains the urgency
The advertising business reached a $1bn annualised run rate in under 200 days. The stated target for this year is $2.5bn, against a total annualised revenue projection above $40bn.
Those figures are the context for a company sounding out a $1.5tn valuation while deferring its listing. Compute costs arrive monthly and subscription growth is finite; advertising is the only revenue line in this industry that has ever scaled to the size of the infrastructure bills.
The same logic produced the modern web. Search, mail, maps and social networks were all free products that discovered advertising second, and in each case the advertising eventually shaped the product rather than the reverse.
The trust problem, stated precisely
The risk is not that users will see ads. It is that the assistant's usefulness rests on a belief that its recommendations are not for sale, and that belief is difficult to hold while a paid participant sits in the same thread.
Jeremy Goldman of eMarketer put the commercial version of the worry: users will leave for a competitor if the advertising feels clumsy or opportunistic. The structural version is harder. Labelling tells you which messages are paid. It cannot tell you whether the unpaid ones were shaped by what pays.
The pipes were laid a week earlier
An advertising format is worthless without demand, and the demand side arrived first: an integration with Amazon's advertising platform was completed on 10 September, six days before the agents were announced.
That sequence is the whole strategy in order. Connect to the systems that already hold advertiser budgets and product catalogues, then open a surface those budgets can be spent on. The alternative — persuading brands to build something bespoke for one assistant — takes years.
What "context and intent" means when you read it carefully
Placement is described as weighing the context and intent of the conversation rather than buying keywords. In practice that is a far richer signal than a search query has ever been.
A search engine knows the words someone typed. An assistant knows what they have been working through for the past twenty minutes: the budget they mentioned, the room they are furnishing, the deadline they are under, the thing they said they could not afford.
Nothing about that is necessarily improper, and it is a category of targeting the advertising industry has never had. The privacy question is not whether the data leaves — it is what a paid participant is allowed to be told about the person it has been introduced to.
What to watch
Whether the label survives contact with revenue targets. Every advertising medium begins with a bright line between editorial and commercial and then spends years negotiating it; the pressure runs one way.
Then disclosure. If OpenAI publishes how often a sponsored agent appears, in which categories, and what it costs per conversation, the claim about answer independence becomes checkable. Until then it is a commitment, and commitments are what companies have instead of accounts.
The announcement of Sponsored Agents on 16 September 2026, the described connections to Shopify and HubSpot, Wayfair as a pilot partner, the completion of an Amazon Ads demand-side platform integration on 10 September 2026, the $1bn annualised run rate reached in under 200 days, the $2.5bn advertising target for 2026 and the projection of more than $40bn in total annualised revenue are as reported by Forkast on 16 September 2026. The labelling commitment, the description of context-and-intent placement, the quotation from David Dugan of OpenAI and the comment from Jeremy Goldman of eMarketer are as reported by KCRG on 16 September 2026. OpenAI's own announcement pages were not reachable from this desk at the time of writing. The analysis is our own.
Topics technologyopenaiadvertisingai
