OpenAI Turned the Ad Unit Into an Agent You Can Talk To

On September 16, OpenAI introduced three things at once for ChatGPT Ads: a test of Sponsored Agents, natural-language campaign and creative tools for advertisers, and integrations with HubSpot and Shopify.

Sponsored Agents is the one that changes the format. When a relevant ad appears in ChatGPT, the user can enter a clearly labeled conversation with a business-sponsored agent — state what they actually need, ask follow-ups, and click through to the business's site when ready.

US-based Shopify merchants can use a new ChatGPT Ads app in the Shopify App Store from September 16 to create and manage campaigns, syncing catalogs and tracking performance. International availability in ChatGPT Ads markets starts September 23. HubSpot customers can connect a ChatGPT Ads account and build ads, monitor performance, and follow up on leads without leaving HubSpot.

Why this is a different product from advertising

Every digital ad format since the banner has done the same job: interrupt, describe, and hand off to a landing page. The advertiser's control ends at the click. Everything after that — whether the page answers the question the user actually had — is a conversion-rate optimization problem that has consumed twenty years of industry effort.

A Sponsored Agent removes the handoff. The user who clicks does not land on a page written for the average visitor. They land in a conversation that can answer their specific question, and the business controls what that conversation knows.

For high-consideration purchases this is obviously better. Someone comparing two mattresses, two insurance policies, or two CRM tools has questions that a landing page cannot anticipate. A sponsored agent can answer them, in context, without the user leaving the surface they were already using.

For the advertiser, it also solves attribution. The conversation is the funnel. There is no gap between impression and intent to model.

The labeling question

OpenAI says the conversation is clearly labeled, and that label is doing an enormous amount of work.

ChatGPT's value rests on users believing the model is answering their question rather than selling them something. Introducing a surface where a business controls the model's responses — even a separate, labeled surface — puts pressure on that belief. The moment a user is unsure whether the answer they got was sponsored, the product's core proposition erodes.

The structural defense is that Sponsored Agents live in a distinct conversation entered by choice, post-click. That is a meaningfully cleaner separation than native advertising in a feed, where the sponsored item sits in the same stream as organic content. Whether users perceive it that way after six months of exposure is the open question, and it is not one OpenAI controls.

Shopify and HubSpot are the actual distribution

The Sponsored Agents test is with select US advertisers. The Shopify and HubSpot integrations shipped to everyone.

That ordering tells you where the revenue is. Sponsored Agents is a format experiment. The integrations are plumbing that lets hundreds of thousands of small merchants buy ChatGPT ads from inside software they already use, with their product catalog and customer context already connected.

Shopify merchants are the ideal first cohort. They have structured catalogs, they are accustomed to managing paid acquisition from within the admin, and their margins make performance advertising a constant operational concern. Dropping a campaign app into the Shopify App Store means the buying decision is a single install rather than a new vendor relationship.

HubSpot covers the other side — B2B, where the conversion event is a lead rather than a purchase, and where the follow-up matters more than the click. Being able to create ads, watch performance, and work the resulting leads in one system, powered by existing CRM context, is a genuinely tight loop.

What it means for the open web

ChatGPT ads are a new demand channel, and demand channels are zero-sum against the ones that already exist.

The same week OpenAI shipped this, executives from the New York Times, the Atlantic, and CBS warned publicly that AI is eroding search referrals and increasing IP pressure, and argued publishers need their own agent products. Germany moved to put Google AI Overviews and Perplexity under media law. A Munich court found Google liable over AI Overviews.

The through line is that the referral economy — search sends traffic to a page, the page monetizes the visitor — is being replaced by an answer economy where the user never leaves. Sponsored Agents is the monetization layer for the answer economy. It is also an admission that the answer economy needs one, because an assistant that never sends traffic anywhere has no other way to pay for itself at consumer scale.

The cost pressure underneath

Nobody should read this purely as product strategy. OpenAI is running inference for hundreds of millions of users on a subscription and API mix that does not cover the free tier.

Advertising is the only business model in the history of consumer software that has funded free products at that scale. The question was never whether OpenAI would sell ads — it was what form they would take in a medium with no page and no feed. Sponsored Agents is the answer: the ad unit is a conversation, because the product is a conversation.

What to watch

Pricing and auction mechanics. OpenAI has not detailed how Sponsored Agent placements are priced. Whether it is CPC, per-conversation, or per-qualified-lead will determine which advertisers can afford it.

Whether the labeled conversation converts. The format's entire premise is that a conversation beats a landing page. Merchants will know within a quarter, and Shopify app reviews will say so publicly.

International rollout on September 23. US-only tests are easy. The first non-US market with different advertising disclosure rules is where the labeling approach gets stress-tested by a regulator rather than a user.