OpenAI launches product carousel ads in ChatGPT

OpenAI launched product carousel ads inside ChatGPT: multiple products from one retailer, side by side, at the bottom of a conversation. OpenAI decides when a carousel fires instead of a single result. If you already run Google Shopping, the same catalogue feed likely pipes straight in. Setup cost is near zero. Spend will flow.

The question is what that spend actually measures.

The attribution problem is real, and it's immediate. OpenAI owns the impression-to-carousel logic. That black box will misattribute in whatever direction flatters its own numbers, and you won't be able to audit it. The trap is the sequence: low CPMs and small dollars through Q3 while nobody has figured out conversion tracking, then Q4 arrives and budgets balloon before attribution is anywhere close to settled. Early Facebook Shopping misfired in exactly that order. Run a small, instrumented test now. Do not wait until October to stress-test your measurement.

There is also a structural ceiling on the format. The carousel sits at the bottom of the conversation, which is banner-blindness real estate. Users trained themselves to ignore that placement fifteen years ago. For this to compete with Google, bottom-of-conversation clickthrough has to justify CPMs at Google-competitive levels. That has not been demonstrated. Until it is, this is an R&D line, not a performance line. Budget it accordingly.

Who this pressures. The direct threat lands on Alphabet (Google's parent company). Google Shopping already trades at a discount on AI-eats-search fear. A real, annualized ad run-rate out of ChatGPT by Q4 turns that fear into a number analysts have to model, not just a thesis they can defer. The Trade Desk (the largest independent ad-buying platform) is a quiet beneficiary if OpenAI eventually needs programmatic pipes for scale. For agency trading desks, this is a third auction environment your team has never run, and the staffing and measurement costs are the real budget item, not the media.

The trust risk is the one OpenAI cannot buy back. People ask ChatGPT for advice. Every carousel spends a little of the sense that the assistant is on their side. OpenAI clearly knows this, which is why the unit sits at the bottom and why the system, not the advertiser, controls when it shows. That placement protects the user experience and caps advertiser upside at the same time.

Our call: OpenAI's 2026 ad revenue will come in below $2.5 billion, most likely under $1.5 billion. Attribution cannot mature fast enough through a single Q4 to unlock real budget rather than experimental trickle. New ad surfaces almost always undershoot their first big public target because measurement lags adoption. For the $2.5 billion figure to hit, everything would have to break right at once in one quarter. That is the less likely path. We revisit by 2027-03-31.

OpenAI just bolted a Google Shopping clone onto ChatGPT. Multiple products from one retailer, side by side, at the bottom of your conversation, with OpenAI deciding when a carousel shows instead of a single product. For an ad-tech operator, the question isn't whether this is clever. It's whether a third commerce auction is now real enough to plan budget against, who it takes from, and how fast.

Reversibility: Type 2 for advertisers. Piping a Google Shopping feed into a new surface is a low-cost experiment you can turn off. Type 1 for the incumbents whose intent moat this erodes over years. What's actually being decided: not "does OpenAI have an ad business" but "is conversational intent a better mousetrap than keyword intent, and does it arrive before 2027 budgets get locked." Forcing function: Q4, when advertisers set 2027 plans. That's the window in the source, and it's the one that matters.

The Market Analyst. The plain-English version: OpenAI is trying to sell the same shopper twice, once when they type a query into Google, once when they ask ChatGPT. The pressure lands on Alphabet, whose search advertising already trades at a discount for "AI eats search" fear. What isn't in the price is timeline. Google shares absorbed the theoretical threat years ago. A real annualized ad run-rate out of ChatGPT by Q4 turns theory into a number analysts have to model. The $2.5 billion 2026 target and $100 billion by 2030 are aspirations, not bookings, so treat them as direction, not forecast. The Trade Desk is a quiet beneficiary if OpenAI ever needs programmatic pipes for scale.

The Skeptic. Steelman the doubt. This ad sits at the bottom of the conversation. That's banner-blindness real estate, the spot users trained themselves to ignore fifteen years ago. For this to work, ChatGPT holds hundreds of millions of active users, those users tolerate commerce inside a tool they trust to be neutral, click-through justifies CPMs that beat Google, and OpenAI builds attribution from scratch. None of that is proven. In plain terms: people asked ChatGPT for advice, not a sales pitch, and shoving products under the answer risks the one thing OpenAI can't buy back, which is the sense that the assistant is on your side. OpenAI also decides when carousels fire. Performance buyers hate ceding that control, because it's a black box they can't optimize against.

The Operator. Tuesday morning, this is nearly free to try. If you already run Google Shopping, the same catalogue file likely pipes straight in, thanks to the feed automation OpenAI shipped three months back. So it will get tried. The trap is the sequence. Spend trickles in through the rest of Q3 at low CPMs while measurement is immature, everyone shrugs because the dollars are small, then Q4 arrives and budgets balloon before anyone has figured out attribution. That's the exact order early Facebook Shopping misfired in. Stress-test your conversion tracking now, in August, not in October. OpenAI owns the impression-to-carousel logic, and a black box you can't audit will misattribute in whatever direction flatters its own numbers.

The Customer / End User. Two customers here, and they want opposite things. The advertiser wants a new high-intent channel and near-zero setup, and this delivers both, which is why early spend will look encouraging regardless of quality. The person using ChatGPT wants an answer, not a storefront. The whole reason conversational search feels valuable is that it doesn't feel like Google's ten blue links with four ads stapled on top. Every carousel spends a little of that goodwill. OpenAI clearly knows it, which is why the unit sits at the bottom and why the system, not the advertiser, decides when to show it. That placement protects the user experience and caps the advertiser's upside at the same time.

The CFO. The measurement gap and the opportunity cost of staffing a third auction environment your team has never run are the real budget items, not the media itself. Early carousel dollars are cheap, but cheap CPMs against unproven attribution can mean you're buying distribution and calling it performance. Ask the uncomfortable question: how many of the first advertisers here are buying results, and how many are buying the press release that says they were early on ChatGPT ads? Until post-click attribution closes, this is an R&D line, not a performance line. Fund it like one. A small, instrumented test budget you can defend, not a Q4 reallocation you'll struggle to explain when the numbers come back muddy.

Where the council splits.

The Strategist-versus-Skeptic fight is the whole story. One read says intent lives inside the conversation itself, no keyword proxy needed, and that's a structurally better signal that pulls budget out of Google's Performance Max faster than anyone expects. The other says a unit at the bottom of a chat window is the oldest dead zone in digital advertising, and trust is the asset OpenAI is quietly spending. Both can't be very right.

Second split: the Operator and the CFO agree spend will flow, and disagree on what it means. Easy setup guarantees dollars move. Easy setup tells you nothing about whether those dollars perform. Volume of adoption will get read as proof of value, and it isn't the same thing.

What it hinges on. Two facts settle this, and neither is public yet. First, does OpenAI close the attribution loop with clean post-click measurement. Without it, budget stays in experiment mode no matter how many advertisers pile in. Second, does bottom-of-conversation clickthrough justify CPMs that compete with Google, because that's the number that separates a real auction from a science project. The council leans cautious-but-serious: real enough to instrument now, not real enough to reallocate Q4 dollars into.

Prediction: By OpenAI's next reported ad-revenue update covering full-year 2026 (early 2027), its 2026 ad revenue will come in below the $2.5 billion target, most likely under $1.5 billion.

Confidence: Medium Attribution and CPM proof can't mature fast enough by Q4 to justify that ramp.

Why: The $2.5 billion figure assumes a standing-start ad business scales through a single Q4 while advertisers still lack clean post-click attribution, which is the thing that unlocks real budget rather than experimental trickle. The source itself frames the rest of Q3 as "about scale" and Q4 as the pressure point, meaning most of the year is still ramp, not run-rate. New ad surfaces almost always undershoot their first big public target because measurement lags adoption, and a unit placed at the bottom of the conversation has a lower clickthrough ceiling than the top-of-results real estate Google monetizes. For the target to hit, everything would have to break right at once in a single quarter, which is the less likely path.

Revisit by 2027-03-31: We're right if OpenAI's reported or credibly leaked 2026 ad revenue lands under $1.5 billion. We're wrong if it hits or clears the $2.5 billion target.