OpenAI Ads and Dynamic Take Rates: Who Really Controls the Stack

Two things happened in the adtech infrastructure last week that, read separately, look like routine industry news. Read together, they describe the same underlying dynamic: the platforms that sit between brands and audiences are quietly rewriting the terms of the relationship — and calling it progress.

OpenAI Enters Advertising on Its Own Consent Terms

ChatGPT ads are now live in the U.K., and OpenAI has published its EU framework: personalized ads will only run to users who explicitly opt in. Digiday reports this positions OpenAI closer to Apple’s ATT model than to Google’s default-on infrastructure — a meaningful architectural choice that will shape how advertisers can realistically target within the platform.

For brands, this is not a small distinction. An opt-in-only personalization environment means the addressable audience inside ChatGPT will be structurally smaller than what you’re used to in open web or social. But the users who do opt in are, almost by definition, more intentional. They’re in a high-attention, query-driven context — closer to search intent than scroll behavior.

For Southeast Asian markets, the consent architecture question matters differently. In markets like Thailand, Vietnam, and Indonesia, where data privacy regulation is still maturing and user awareness of consent mechanisms is uneven, OpenAI’s opt-in framework may actually set a regional ceiling for addressability rather than a floor. Brands betting on ChatGPT as a mid-funnel channel here need to stress-test their targeting assumptions before they build campaign architecture around them.

The broader signal: OpenAI is not entering advertising to play by the existing rules. It’s setting its own, and doing so before regulators force the issue.

---

Dynamic Take Rates: The Fee Structure That Wears a Product Hat

Meanwhile, on the supply side, AdExchanger’s guest columnist made explicit what many programmatic buyers have suspected for a while: dynamic take rates are becoming standard practice across exchanges, dressed up as optimization features rather than disclosed as margin extraction.

Google has operated under the “Average Revenue Share” model for years — a mechanism that varies the fee it takes per auction rather than charging a fixed percentage. Index Exchange has now published its own version, framing it as a transparency-forward feature. Others are following, some publicly, most not.

The mechanism works like this: exchanges charge a higher take rate on impressions where the buyer is likely to pay more, and compress it where competition is lower. For publishers, this can smooth yield floor volatility. For buyers, it quietly increases effective CPMs on the inventory they actually want — without the change appearing as a rate card revision.

This is not innovation. It’s variable pricing that benefits the exchange at both ends of the transaction, and it compounds in programmatic environments where buyers are already operating with incomplete auction transparency. For regional trading desks running campaigns across Lazada display, open web, and app inventory in Southeast Asia, the practical implication is straightforward: your effective take rate is no longer a fixed input in your media math. It’s a variable you probably aren’t modeling.

What These Two Stories Share

At the infrastructure level, both developments represent the same strategic posture: platforms expanding their margin of control over the conditions under which advertising works, while giving buyers just enough framing to make it feel like a feature rather than a constraint.

OpenAI controls the consent layer and the context. Exchanges control the fee layer and the auction dynamics. Neither change is inherently malicious — both have defensible rationales. But for brands and agencies building media strategies that depend on predictable reach, cost, and signal, the cumulative effect is a stack that’s becoming harder to audit and easier to get quietly squeezed by.

The response isn’t paranoia — it’s infrastructure hygiene. For programmatic teams, that means pushing exchanges for explicit take rate disclosures and running periodic supply path audits, not annually but quarterly. For teams evaluating AI-native ad placements like ChatGPT, it means building consent-rate assumptions into your reach forecasts from day one, not retrofitting them after campaign launch.

The Identity Layer Underneath Both

There’s a third thread running through both stories that doesn’t get named directly: identity. OpenAI’s opt-in personalization framework only works if users have accounts and OpenAI can resolve them to ad segments — a walled garden identity model. Dynamic take rates are partly a response to signal degradation in cookieless environments, where exchanges are trying to recapture margin lost when targeting precision drops.

Both are symptoms of the same post-cookie fragmentation: as third-party identity signals weaken, every platform with a logged-in user base is building the infrastructure to monetize that advantage. The exchanges are compensating for margin pressure with variable pricing. The AI platforms are entering advertising with consent architecture that doubles as a data moat.

For brands in Southeast Asia operating across fragmented identity environments — where LINE, Grab, Shopee, and TikTok each hold meaningful first-party graphs with limited interoperability — the message is consistent: the platforms that own consent own the targeting, and they will price accordingly. Clean room partnerships and direct publisher relationships aren’t just privacy compliance exercises. They’re the hedge against a stack where the toll keeps going up.

Key Takeaways

• Audit your effective take rates quarterly, not annually — dynamic pricing means your programmatic cost structure is a moving target, not a fixed input.
• Model ChatGPT ad reach conservatively: opt-in-only personalization in a new platform means addressable audience size will be materially smaller than open-web equivalents, especially in markets with low consent literacy.
• Treat first-party identity as supply chain infrastructure: every platform with a logged-in user base is building a monetization layer around it — brands without their own first-party graph are permanently at the toll booth.

---

The adtech stack has always had friction built into it. What’s shifting now is that the friction is becoming dynamic — variable by auction, gated by consent, priced by platform leverage. The question worth sitting with: if both the AI platforms and the exchanges are optimizing for their own margin expansion simultaneously, what does a genuinely buyer-side media infrastructure actually look like in 2027?

---

At grzzly, we work with growth and media teams across Southeast Asia on exactly this kind of infrastructure question — from supply path optimization and take rate auditing to first-party identity strategy and AI-native channel evaluation. If the stack is shifting faster than your playbook, that’s usually the right moment to pressure-test your assumptions with someone who spends their days at this frontier. Let’s talk