Amazon Ads in ChatGPT: Azoma explains what it means for organic AI visibility

Amazon announced a partnership with OpenAI on 10 September that lets advertisers extend their Amazon Ads campaigns into ChatGPT. Select US advertisers are testing it as a pilot, buying through Amazon DSP rather than dealing with OpenAI directly, with Delta Vacations among the first.

Chris Conetta, Director of Omnichannel Supply at Amazon DSP, called conversational ads the fastest growing engagement opportunity for brands to reach new and existing audiences. The commercial logic is obvious on both sides. The interesting part is what it does to the organic answer sitting directly above the ad.

What shipped

1. A buying route, not a new ad product.

OpenAI has been selling sponsored placements in ChatGPT since February, starting with retailers including Best Buy and Williams-Sonoma, and that business has reached a reported $1 billion annualised revenue run rate. What changed on Thursday is that Amazon advertisers can now reach it through a platform they already use.

Ad units appear as text or images beneath ChatGPT responses and are labelled as sponsored.

2. Amazon manages the campaign, OpenAI decides the delivery.

Amazon sets up and runs campaigns. OpenAI's ads system controls all delivery decisions, meaning when and where an ad appears. At launch it runs as a managed service, with Amazon reps helping build campaigns and providing context hints to match ads to the right topics and conversations.

That division matters for anyone modelling this. A brand buying through Amazon does not control placement inside the answer, and neither does Amazon.

3. A reversal, on Amazon's terms.

Amazon has restricted AI platforms from its store while building its own shopping agents. It has also, quietly, been buying ChatGPT ads itself, becoming the top retail advertiser on the platform between April and August according to Sensor Tower data cited by CNBC.

So Amazon gets visibility where product research now happens, without opening its catalogue or giving up the checkout.

What Amazon did not say

Several outlets have reported that Amazon's shopping data will be used to target ChatGPT users. Amazon's announcement does not state this.

The release describes Amazon Ads generally as offering trillions of proprietary browsing, shopping and streaming signals, which is standard boilerplate that appears across its materials. Neither Amazon nor OpenAI has stated that those signals are shared with OpenAI or used directly for targeting inside ChatGPT.

The honest position today is that Amazon is a demand partner and a buying route. Whether it becomes a data partnership is the question worth watching, and it has not been answered.

The question every brand will ask next

The first analyses of ChatGPT ad performance will almost certainly show that brands advertising on the platform also appear more often in organic ChatGPT answers. That correlation is likely to be real. Establishing that the advertising caused it is a separate question, and at least three explanations fit the same data.

The first is selection. Brands buying ChatGPT ads tend to be larger, better covered in the press, and already well represented in the earned media and retailer sources that organic answers draw on, which produces the correlation independently of any campaign.

The second is co-occurrence. A sponsored unit and an organic mention both respond to the same query context, so the two appear together by design rather than by influence.

The third is a genuine halo. Advertising that drives traffic and coverage could feed the organic sources over time, which would be a real causal effect, and it would be slower and harder to isolate than the first two.

Worth being precise about vocabulary here, because it is unsettled and the confusion travels into budget conversations. Agentic commerce optimisation overlaps heavily with what is variously called GEO (generative engine optimisation), AEO (answer engine optimisation), and AI search optimisation.

The distinctions are mostly emphasis: GEO names the technology producing the answer, AEO names the behaviour being optimised for, and agentic commerce optimisation names the case where the answer ends in a purchase. All of them describe the organic half of what is now a two-sided problem.

Where the budget comes from

One question worth asking early is whether this represents new budget.

Retail media budgets are already committed across Amazon, Walmart and the rest. A brand adding ChatGPT through Amazon DSP is most likely moving a dollar rather than finding one, and moving it into a channel with no established benchmark for what good looks like.

The buying route got easier. The budget question did not change.

The placement itself is worth noting. A sponsored unit appears beneath a ChatGPT response that has already made a recommendation, so the ad is read in the context of whatever the answer above it said.

What this means for GEO programmes

GEO, generative engine optimisation, is the work of getting a brand and its products discovered, correctly described and recommended inside AI-generated answers. A paid layer arriving alongside it changes three things and leaves the work itself untouched.

Ownership gets contested. GEO has typically sat with search, content or digital shelf teams. Retail media sits with the media team. Both now buy against the same surface, and the surface count keeps rising, with Meta Muse and Instacart Clementine both launching in the past fortnight.

Attribution gets harder. Once both are running, an organic movement has a plausible paid explanation available and vice versa. Separating the two afterwards requires a baseline recorded before the campaign started.

The organic half gains value. A labelled sponsored unit and a recommendation inside the answer are read differently by shoppers. The thing paid cannot buy is being the answer.

That last point is worth dwelling on for anyone asking how to get recommended by ChatGPT in the first place. Recommendations are assembled from cited sources, and Azoma's analysis of millions of shopping agent citations found that in Q2 2026 ChatGPT's shopping source mix ran at 41% earned media, 37% retailer, 19% UGC and 3% brand.com.

Buying an ad beneath the answer does nothing to that mix. It is the same 78% of earned media and retailer coverage deciding the recommendation as it was last week.

What brands should do about it

Three things, and none of them are new work.

Establish where you stand organically before the campaign starts. Once spend is live, the pre-period is gone and any question about what the advertising did becomes unanswerable. This is a two-week job now and an impossible one in January.

Hold something back. Advertise against one set of categories or prompt themes and deliberately not against a matched set. Without that, category-level movement and campaign effect look identical.

Watch the sources, not the score. If paid spend genuinely lifts organic recommendation, it does so by feeding the sources an answer is assembled from: coverage, reviews, retailer content. A brand-level visibility number tells you something moved. Knowing which sources ChatGPT cites when recommending products in your category tells you what moved it, and that is the only version that survives an argument about causation.

➡️ Azoma is an AI visibility platform that tracks which sources ChatGPT cites when recommending products, at product level and across Alexa for Shopping, Walmart Sparky and Gemini.

What this does not change

It does not change how ChatGPT decides what to recommend. A sponsored unit sits beneath the answer. The answer above it is still assembled from product data and the sources the model retrieves, so ChatGPT visibility is still earned through complete attributes, accurate product information and presence in the sources it cites.

It does not mean paid and organic are interchangeable. A shopper who reads a recommendation and a shopper who clicks a labelled ad are in different states. Substituting spend for the work of being recommended buys attention rather than endorsement.

It does not tell us about Alexa for Shopping. Amazon buying ChatGPT ads and Amazon selling ChatGPT ads to its advertisers say nothing about whether spend affects retrieval inside Amazon's own agent, which is a separate system with separate incentives and no published answer.

Important caveat: none of the above should be read as scepticism about the channel. ChatGPT reported 900 million weekly active users in February and a $1 billion annualised ad run rate, so this is a real market forming quickly. The caution is about attribution, not about whether to participate.

Summing Up

Advertising is going to be the theme of 2027 in agentic commerce. This announcement points squarely at that, and the budget conversations will start well before the measurement does.

The practical change for brands is narrower than the announcement suggests. There is now a paid lever in a channel where the organic answer still does most of the persuading, and in most organisations the two are bought by different teams on different timelines.

For anyone running a GEO or answer engine optimisation programme, paid arriving in the answer does not reduce the organic work. It raises the cost of not knowing whether the organic work is landing, because now there is a plausible alternative explanation for every movement.

Paid placement in an AI answer is straightforward to buy and difficult to attribute. Brands that record where they stand organically before they start spending will be the ones able to tell the difference later.

Azoma is built for exactly this. We measure how AI shopping agents retrieve, compare and recommend products across Alexa for Shopping, Walmart Sparky, ChatGPT and Gemini, and fix the sources those answers are built from. Get in touch today to set an organic baseline before your first ChatGPT campaign runs.