--- title: "Amazon Just Opened a Side Door Into ChatGPT Ads. Your Measurement Stack Wasn't Built for a Third Buying Path." description: "Amazon's new DSP pilot makes ChatGPT ad inventory easy to buy, but the aggregated reporting you get back stops well short of what your attribution model needs. Here is the readiness checklist before you fund it." author: LETSGROW Dev Team date: 2026-09-12 category: AI Tools tags: ["ChatGPT Ads", "Amazon DSP", "AI Advertising", "Marketing Attribution", "Retail Media"] url: "https://letsgrow.dev/blog/amazon-chatgpt-ads-dsp-pilot-readiness" --- Amazon just gave its advertisers a third way to buy ChatGPT ads, and almost nobody is talking about what that does to attribution. On September 10, Amazon DSP opened a pilot letting select US advertisers, with Delta Vacations named as the first, extend campaigns directly into ChatGPT through Amazon's own buying tool instead of going through OpenAI. That sounds like a distribution win. It is actually a measurement problem wearing a distribution win's clothes, and most marketing teams have no plan for it. ## What Actually Launched Strip away the press language and the mechanics are simple. Advertisers buy ChatGPT ad inventory through Amazon DSP on a cost-per-click or CPM basis, the same way they buy Netflix, Roku, or Disney inventory today. Amazon also offers product feed ads that auto-generate creative from an advertiser's existing catalog. OpenAI still decides which ads actually appear beneath a given chat response and where they get placed. Amazon just becomes the buying and campaign-management layer sitting on top. That split matters more than the headline. You are not buying ChatGPT ads. You are buying access to ChatGPT ads through a second company's black box, layered on top of a first company's black box. Each layer adds a translation step, and every translation step is a place where your reporting can quietly stop matching reality. Advertisers in the pilot get aggregated performance data: impressions, clicks, cost per result, CPM, CPC. Notice what is missing. There is no mention of conversion-level attribution tying a ChatGPT ad exposure to a purchase in your own systems. Amazon is pitching its shopping and browsing data as the differentiator here, the idea being that Amazon's first-party signal makes ChatGPT placements smarter than buying blind. That pitch only pays off if you can actually connect what Amazon knows about a customer to what happened after the ad ran, and right now that connection lives entirely inside Amazon's reporting, not yours. ## Why the Buying Path You Choose Now Determines What You Can Measure Later Marketing teams evaluating ChatGPT Ads are choosing between three paths that look similar on a slide and behave completely differently in a data warehouse. ::compare-table{title="Three Ways to Buy ChatGPT Ads Right Now"} | Path | Who controls delivery | What you get back | Where this breaks | |---|---|---|---| | Direct with OpenAI | OpenAI | Platform-native reporting, direct relationship | Immature tooling, limited formats, no retail media overlay | | Through Amazon DSP | OpenAI serves, Amazon manages buying | Aggregated CPC/CPM/impressions via Amazon | No native tie to your CRM or conversion events, second black box | | Through an agency holdco desk | Varies by holdco integration | Blended reporting inside existing agency stack | Attribution logic buried in agency methodology you don't own | :: None of these three paths hands you row-level exposure data you can join to your own conversion events out of the box. That is not a hypothetical risk. It is the exact failure mode marketing teams already lived through with the walled gardens: rich targeting, thin measurement, and a growing gap between what the platform says happened and what your revenue system says happened. Amazon adding a fourth walled garden's inventory to its DSP does not close that gap. It adds another wall. ## The Part Everyone Is Skipping: OpenAI's Ad Business Is Still Unfinished Infrastructure It is worth saying plainly: OpenAI started testing ads in February and is reportedly targeting $100 billion in ad revenue by 2030, a number that requires sustaining growth above 200% a year from here. That is an aggressive target sitting on top of an ad stack that industry analysts openly describe as still missing basics: team, technology, vendor partnerships, formats, and pricing discipline. Amazon's deal gives OpenAI a fast-track to Amazon's existing advertiser base and Amazon's retail media credibility. It does not give OpenAI a finished measurement product overnight. If you buy into this pilot expecting Amazon-grade attribution because Amazon's name is on the buying interface, you are borrowing trust the underlying inventory has not earned yet. The two companies split responsibility cleanly: Amazon owns the buy, OpenAI owns delivery. Nobody in that split owns making sure your finance team can defend the spend in a QBR. ## The Readiness Checklist Before You Touch This Pilot ::checklist{title="ChatGPT Ads Via Amazon DSP: Pre-Pilot Checklist"} - Confirm whether Amazon's aggregated reporting includes anything beyond impressions, clicks, CPM, and CPC. If not, build a plan for closing the loop to your own conversion events before spend goes live. - Treat this as retail-media-adjacent spend, not search or social spend, when it hits your attribution model. It behaves like a walled garden, not like Google Ads. - Ask whether your MMM or incrementality testing framework has a slot for a new channel with no historical baseline. If it doesn't, this spend will show up as unexplained variance for at least a quarter. - Set a hard cap on test spend until you can answer, in writing, how a ChatGPT-via-Amazon impression gets tied to pipeline in your CRM. - Watch whether OpenAI expands this beyond the US pilot before you build permanent reporting infrastructure around it. Early access is not the same as stable access. :: ## The Actual Takeaway Amazon is not making ChatGPT Ads better. It is making ChatGPT Ads easier to buy, which is a different thing and a more dangerous one. Easy buying paths get budget before measurement paths get built, every single time, because procurement moves faster than analytics engineering. The teams that come out ahead here are not the ones who buy first. They are the ones who write down, before a single dollar moves, exactly what data they will and will not get back, and who decide now whether that gap is acceptable or a blocker. If you cannot currently answer what an Amazon-bought ChatGPT impression will look like in your own attribution model, that is your answer for this quarter: watch the pilot, do not fund it. Travel and other high-consideration verticals will be the test case. Let Delta Vacations spend the learning budget first.