Amazon Ads Now Sells ChatGPT Ads via Its DSP

On September 10, Amazon Ads announced a partnership with OpenAI to bring advertising to ChatGPT, something the company itself calls "a conversational ad experience on ChatGPT." Amazon's own page confirms the pilot is already live: "Select advertisers in the U.S. are currently testing ChatGPT Ads as part of a pilot." Delta Vacations is named "among the first brands testing this new approach," and Katrin Koenig, of that brand, talks about personalization for trip planning. Chris Conetta, Director of Omnichannel Supply at Amazon DSP, sums up the pitch: "Conversational ads represent the fastest growing engagement opportunity for brands to reach new and existing audiences." That's where the press release ends. The interesting part starts in what it leaves out.

A second door on a channel that used to have one

Until today, buying ad inventory inside ChatGPT meant going through one place: the Ads Manager OpenAI built, the same tool we covered when the ChatGPT Ads Manager plugin started running campaigns by conversation. What actually changes with this announcement isn't spelled out on Amazon's page. Digiday spells it out instead: "From today (September 10), advertisers using Amazon Ads, the retail giant's advertising platform, which includes its buying tool, the Amazon DSP, can buy ads inside the ChatGPT app." That detail changes the size of the story. Amazon isn't just one more advertiser inside OpenAI's system, a large third-party DSP is becoming an additional entry point into that inventory.

That widens who can buy, and more importantly, how. A team that already runs Amazon DSP could, in theory, add ChatGPT to that same account without opening a new tool. Digiday is precise about how limited this still is: "It's a pilot for now, limited to select U.S. advertisers, with Delta Vacations among the first named." None of this is generally available yet.

What Amazon says, and what everyone else says

Amazon's page speaks in the positioning language that serves the brand, while trade press adds the actual buying layer and the questions the brand doesn't answer. Per ppc.land, the company points to "trillions of proprietary browsing, shopping, and streaming signals" to promise results "without relying on ad identifiers." That's a brand claim, made by an interested party, and it's worth reading it as exactly that.

What Amazon's page doesn't say runs longer than what it does. It doesn't clarify whether the full Amazon DSP is what's being extended, or some subset of it. It doesn't say how many advertisers are in the pilot. It doesn't specify what audience signals are actually used. It publishes no pricing. It gives no expansion timeline. And it doesn't mention a single market outside the United States. Each of those gaps matters differently, but one outweighs the rest: Amazon's page doesn't say a single word about measurement or attribution.

What the DSP's low fees don't cover yet

The Amazon DSP's low fees describe its general positioning in the programmatic market, not the price of this ChatGPT inventory, which nobody has published. Digiday adds a real data point, but it needs careful handling: "Amazon Ads already undercuts rivals on fees, with its DSP often landing near zero on programmatic guaranteed deals against The Trade Desk's historical 15 to 20% or Google's similar cut." That's a known story about Amazon's DSP in the fight for programmatic budget, something we've been tracking since we wrote about consolidation in retail media, not a promise about this specific inventory. Assuming ChatGPT ads bought through Amazon will come in cheaper because Amazon's DSP tends to be cheap is exactly the kind of leap an advertiser shouldn't make on what's public today.

Measurement is still the open question

Measurement is, by the trade press's own account, the biggest open problem in OpenAI's ad business, and this announcement doesn't solve it. Digiday puts it near the top of the list: "Measurement, clunky tech and limited ad formats sit near the top of that list." ppc.land is even more direct about this specific announcement, noting it "offers no detail on how attribution will function across the ChatGPT surface," and closing with the question that actually matters: "Whether that foothold delivers measurable performance, rather than presence, is the question the pilot exists to answer." We already wrote about how to measure ChatGPT ads with the tools that exist today, and that conclusion still holds: the buying layer is moving faster than the measurement layer, and this announcement confirms that gap rather than closing it.

For an advertiser, the logical order is to audit first and scale second. What's happening here runs the other way: the channel gains a new, bigger, more familiar buying path, while the question of how to measure that spend sits exactly where it was before the announcement.

What to do while the pilot stays closed

The pilot is invite-only, so most U.S. advertisers cannot opt in today even though the U.S. is the only market enabled. The useful move is not to wait for general availability, it is to find out whether you are in scope at all. Teams already running Amazon DSP have a rep who can answer that, and the answer decides whether this is a Q4 planning question or a next-year one.

There is a second question worth settling now, because otherwise the default will settle it. If this channel ends up packaged inside the DSPs a team already uses rather than inside OpenAI's native Ads Manager, buying it stops being a budget decision and becomes a question of access and commercial terms: who has the account connected, under which DSP, and at what fee. Those are answerable today, before anyone flips a switch.

For teams running accounts outside the U.S., the read is narrower. Nothing operational changes yet, and while previous ChatGPT Ads expansions arrived in blocks, as we covered when ChatGPT Ads added Brazil and Mexico, that says nothing about when DSP-based access follows. It is a separate decision, and neither company has put a date on it.

Who this is for

This note is for growth and media teams already running Amazon DSP, or evaluating it, who need to know what actually opened on September 10 and what did not. It is also for anyone tracking ChatGPT Ads as a channel who wants to place this announcement in the right sequence: it is a distribution move, not a measurement one. It is not an implementation guide, because the pilot is closed. It is a map of what changed, and of which questions are worth answering now.

This content was developed with AI assistance and reviewed by the Zenda team. Any bad ideas are entirely our own.