Amazon starts testing ads inside ChatGPT
Amazon has opened a route for advertisers to extend Amazon Ads campaigns into ChatGPT. That is the announcement, and it is worth reading carefully, because of what is not in it. There is no launch date. There is no price. There is no performance data. The only market named is the United States, where select advertisers are described as currently testing ChatGPT Ads as part of a pilot.
Treat this as something a platform is testing rather than something anyone can go and buy. That framing is not a hedge, it is the announcement’s own description of itself.
What was actually announced
The mechanism, as described, is an integration at the buying layer. Advertisers already running multi-channel campaigns through Amazon DSP can extend those campaigns into a conversational ad experience on ChatGPT, rather than building and buying that placement separately. Amazon frames the value as its own audience data reaching a surface it did not previously touch, describing the advantage in terms of proprietary browsing, shopping and streaming signals. No figure is attached to that claim anywhere in the announcement.
Chris Conetta, Director of Omnichannel Supply at Amazon DSP, calls conversational ads “the fastest growing engagement opportunity for brands to reach new and existing audiences”. Delta Vacations is named as an early tester, and its president, Katrin Koenig, says the company is “proud to be among the first brands testing this new approach and helping define what the future of AI-powered marketing can look like”.
Read those two quotes together and the shape of the thing is clear. This is a pilot with named participants, described in the language of direction rather than results. Nobody is claiming a return yet, because nobody has published one.
Tool comparison · FastMoss vs Kalodata
New ad surfaces arrive faster than the tooling that measures them, which is why sellers keep buying blind. FastMoss and Kalodata are built for exactly that gap on TikTok Shop, showing what is actually converting in a category before you commit a budget to it.
Comparison coming soon
FTC disclosure: our tool comparisons carry affiliate links. If you sign up through one we may earn a commission, at no extra cost to you. We rank on merit, never on commission, and the verdict is written by a human.
From the publisher
Everything we know about running Amazon, in one course.
The commercial strategy, taught by operators who run our own brands.
Why an absence of numbers is the story
Most product announcements come with at least one figure, usually a flattering one. This one carries none, and that is unusual enough to be informative.
An advertiser cannot model this. There is no cost per thousand impressions to compare against display, no click-through benchmark to compare against search, no reported lift from the pilot brands, no minimum spend and no reported inventory volume. Every input a media plan needs is missing. Any agency presenting a projected return on a conversational ad placement this quarter is presenting an assumption, and it is fair to ask them which published figure it rests on.
The absence also sets a useful test for the next few months. When numbers do appear, check who measured them. A platform reporting on the performance of its own new inventory, in a pilot it selected the participants for, is not a measurement of what the placement would do for a business that was not selected. Early figures from a pilot are the most flattering figures a channel ever produces, and they are produced by an interested party.
What a pilot with selected participants rules out
Three things follow from the way access is described, and each of them closes off a plan somebody is probably already drafting.
It is not self-serve. Advertisers described as selected for a pilot did not apply through a console and switch it on. There is no stated route in for a business that has not been approached, and no application process is described.
It is not a European product. The announcement names the United States and does not describe availability anywhere else. Nothing published says when or whether the Amazon route into conversational placements reaches European advertisers, and a European seller planning on it is planning on an assumption.
And it is not a change to how anything is currently bought. Campaigns running through Amazon DSP today behave exactly as they did last week. This adds a possible destination for a future campaign. It does not alter an existing one.
It is also worth being precise about who this reaches even in the market where it is being tested. Amazon DSP is a demand-side buying platform used by brands and agencies running display and video at scale. It is not the Sponsored Products console that most marketplace sellers live in. A seller whose entire advertising operation is keyword bids on a marketplace is several steps removed from this announcement, and the coverage treating it as a change to marketplace advertising is collapsing two different products into one.
The structural change worth watching
Set the pilot aside and there is a genuine shift underneath it, which is why this is worth ten minutes rather than none.
Retail media has spent a decade growing on the retailer’s own surface. A shopper searched on a marketplace, and the marketplace sold the attention of that search to the brands competing for it. The demand and the inventory sat in the same place. What is being tested here is the separation of those two: the shopping intent data stays with the retailer, and the placement moves to somebody else’s product entirely.
If that model works, the competitive question for a marketplace stops being how much traffic it holds and becomes how well it can follow its buyers off its own property. And for a seller, the longer-term consequence is uncomfortable but simple. The behavioural data you generate by selling on a platform becomes an asset that platform can monetise in places you have no listing, no presence and no control.
There is a second-order effect worth naming too. Conversational interfaces answer a question with a recommendation rather than a ranked list of results. A ranked list is something a seller can work on, with price, content, reviews and bids. A recommendation is a single outcome produced by a process nobody outside the system can inspect. If advertising arrives inside that, the question of what separates a paid recommendation from an organic one becomes the most important question in the channel, and it has not been answered anywhere in this announcement.
None of that is a conclusion. It is one pilot in one market with nothing published about how it performs. But it is the reason to keep the tab open.
Tool comparison · FastMoss vs Kalodata
The pattern here is old even where the surface is new, because conversational discovery is what social commerce has been doing for years. Our head-to-head of FastMoss and Kalodata compares how each tracks product demand on a feed rather than a search bar.
Comparison coming soon
FTC disclosure: our tool comparisons carry affiliate links. If you sign up through one we may earn a commission, at no extra cost to you. We rank on merit, never on commission, and the verdict is written by a human.
What a seller does differently on Monday morning
Honestly, for almost every European seller, nothing. There is no product to buy, no console to open, no setting to change and no deadline to meet. Saying so is more useful than manufacturing an action.
Two things are worth doing anyway, and both are cheap.
- Write down what you would need to know before you would spend on a conversational placement, while you are still calm about it. A cost basis, a measurable conversion path, an attribution window you understand and a way to turn it off. Doing that now means the eventual pitch arrives into a list of questions rather than into enthusiasm.
- If an agency or a tool vendor brings you a proposal for advertising inside a chat assistant in the next quarter, ask a single question: which published figure is your projection built on. There is currently no correct answer to that question, and the way it is handled will tell you a great deal about the vendor.
Where the honest uncertainty sits
The pilot’s scale is not described. How many advertisers, which categories and how much inventory are all unstated, so nobody outside it can tell whether this is a broad test or a handful of brands.
The commercial terms are unstated. Pricing model, minimum spend and how the placement is reported against the rest of a DSP campaign are all absent.
And the roadmap is unstated. An announcement that a pilot exists is not a commitment that it graduates, and platforms routinely test things that quietly stop. Treat the availability of this outside a US pilot as unknown rather than imminent, because unknown is what has been published.
The three-line version
- If you run the ads: nothing changes this week. Write your list of pre-conditions for buying a conversational placement now, before anyone pitches you one.
- If you have just taken on the ad budget: the gap to name is that the data your sales generate on a marketplace can be sold as targeting somewhere you do not sell, and nobody in the business has a position on that yet.
- If you sign the budget: there is nothing to approve and nothing to buy. Refuse any forecast for this placement until a cost basis and an independent performance figure exist.
Sources
- Amazon Ads: https://advertising.amazon.com/library/news/amazon-ads-chat-gpt-advertising-integration
- ChannelX: https://channelx.world/2026/09/new-amazon-ads-chatgpt-integration-for-dsp/