Two days ago, Amazon announced it would let advertisers buy ChatGPT inventory through Amazon DSP. Delta Vacations is among the first brands testing the integration, which is limited to U.S. advertisers at launch. The mechanics are straightforward: you buy through Amazon's demand-side platform on a CPC or CPM basis, Amazon handles campaign setup and optimization, and OpenAI controls where and how ads appear inside ChatGPT.
The partnership matters less for what it does today than for what it signals about where conversational ad inventory is headed. Amazon operates the third-largest digital advertising platform, with $68.63 billion in ad revenue last year and Q1 2026 sales of $17.24 billion. OpenAI's ad business hit $1 billion in annualized revenue run rate in under 200 days. Connecting those two ecosystems creates a new buying path for brands already spending through Amazon DSP, and it gives OpenAI access to a massive advertiser base as it scales ChatGPT Ads globally.
For marketing leaders evaluating this pilot, the question isn't whether conversational ads are interesting. The question is whether you can model the economics, measure the lift, and justify the spend to your CFO.
The Split-Control Problem
The architecture of this partnership creates an unusual dynamic. Amazon provides the buying and campaign-management layer, while OpenAI controls ad delivery and placement through its own systems. You're buying through one platform and trusting another to serve the ads.
This split matters for forecasting. Amazon's shopping and streaming data can inform audience targeting, but OpenAI decides which ads appear based on conversational context. The two systems optimize for different signals. Amazon knows what you've browsed and bought. OpenAI knows what you're asking about right now. Whether those signals align for your product category is an empirical question, not a given.
Advertisers testing the integration will receive aggregated performance data including impressions, clicks, cost per result, CPM, and CPC. That's table stakes for any paid channel. What's missing is the closed-loop attribution that makes Amazon's retail media so attractive in the first place. On Amazon's owned properties, you can tie ad exposure to purchase. In ChatGPT, you're back to tracking clicks and hoping your downstream conversion tracking holds.
The Attribution Gap Is Real
Conversational AI introduces a measurement problem that traditional attribution models weren't built to solve. A prospect might ask ChatGPT about project management software on Monday, revisit the conversation on Wednesday to compare pricing, and convert on Friday through a completely different channel. The attribution system struggles to connect those dots.
This isn't a ChatGPT-specific problem. Marketing attribution is breaking down because customer behavior is complex and rarely follows a single, trackable path. But conversational AI accelerates the breakdown. When a user engages with your brand mention in message three of a fifteen-message conversation, with the actual conversion decision forming gradually across subsequent exchanges, your attribution system sees "ChatGPT referral traffic" without understanding the influence that occurred upstream.
AI-referred buyers convert four to five times faster than the average organic visitor in B2B, according to some early data. But proving that to your CFO requires measurement infrastructure most teams don't have. The click that measurement depended on is happening later, happening less, or not happening at all.
Early Benchmarks and What They Don't Tell You
The limited performance data available on ChatGPT Ads suggests the channel can return money, but the variance is high. One account running scaled ChatGPT Ads campaigns over 15 days reported a 1.49x blended ROAS at roughly $1.72 per click. Day-to-day ROAS swung from 0.2x to 2.9x.
Compare that to B2B advertising benchmarks across established channels: LinkedIn runs $9.39 CPC at a 0.67% click-through rate, producing leads at $202 each. Facebook delivers $1.95 CPC with a 0.79% CTR and $145 cost per lead. Google Ads runs $9.76 CPC at a 6.33% CTR, but the cost per lead hits $524 because the click-to-lead conversion rate is lower.
ChatGPT Ads pricing sits somewhere in the middle of that range, but the comparison is misleading. A search impression and a feed impression and a conversational mention are not the same unit. The intent signal differs. The user's mental state differs. The path to conversion differs. Comparing CPMs across these channels tells you what you're paying, not what you're getting.
The Pilot Math
If you're considering this pilot, here's the framework I'd use to model it.
Start with your current CAC payback by channel. If you're running Amazon DSP for awareness or consideration, you already have a baseline for what that inventory costs and what it produces. The question is whether extending into ChatGPT improves or degrades that baseline.
Next, estimate the attribution leakage. If 30% of your conversions currently come through channels where you can't trace the influence back to a specific ad exposure, adding ChatGPT inventory will likely increase that percentage. Build that into your model. If you can't tolerate more attribution uncertainty, this isn't the right pilot.
Then, define your learning objective. The value of early-mover access isn't necessarily the ROAS you'll achieve in month one. It's the signal you'll gather about how your audience engages with conversational ads, which creative approaches work, and whether the channel can scale. Price that learning into your pilot budget.
Finally, set a kill threshold. If the pilot runs for 60 days and you can't demonstrate lift above your baseline Amazon DSP performance, what do you do? Having that answer before you start prevents the pilot from becoming a permanent line item that nobody can justify but nobody wants to cut.
Who Should Test This
The pilot makes the most sense for brands that already use Amazon DSP at scale, have products with broad consumer appeal or high search volume, can tolerate attribution ambiguity in exchange for early-mover learning, and have the measurement infrastructure to track downstream conversions even when the click path is indirect.
Nonendemic advertisers, companies that don't sell products on Amazon but already use Amazon DSP to reach audiences elsewhere, may find this particularly interesting. The integration gives them a path into ChatGPT without building a direct relationship with OpenAI's ad sales team.
For B2B marketers with long sales cycles and narrow ICPs, the calculus is harder. Conversational AI is better suited to products where the path from question to purchase is short. If your buyer asks ChatGPT for marathon training tips and sees an ad for running shoes, the connection is obvious. If your buyer asks ChatGPT about enterprise data governance and sees an ad for your compliance platform, the connection exists but the conversion path is longer and harder to track.
The Forecast Question
Amazon is becoming another gateway into ChatGPT Ads, but OpenAI isn't handing over control of its inventory. That's the structural tension in this partnership. Amazon manages the buying relationship. OpenAI decides how and where ads appear. For advertisers, that means you're optimizing against a system you don't fully control, using data from a platform that doesn't see the full picture.
The right response isn't to avoid the channel. It's to model the uncertainty, run a bounded pilot, and measure what you can while acknowledging what you can't. If the pilot produces signal, scale it. If it doesn't, kill it and reallocate the budget to channels where you can prove lift.
Model or it didn't happen.
