ChatGPT Ads and Google's Bidding Shift: Who Controls Agentic Ad Buying?

Two announcements dropped within 24 hours of each other last week that, taken together, describe the same structural shift in advertising from opposite ends. One was celebrated. One was barely noticed. Both deserve a hard look.

ChatGPT’s $1B Ad Run Rate Is a Stack Question, Not Just a Media Question

OpenAI’s ChatGPT advertising business has reached a $1 billion annualised run rate, according to Digiday — calculated by multiplying current monthly revenue by 12, so it reflects the present pace rather than booked revenue. Europe is now getting self-serve access, which signals OpenAI is moving beyond managed deals toward scaled, democratised ad buying.

For marketing directors in Southeast Asia, the instinct will be to ask: should we test ChatGPT placements? That’s the wrong first question. The right one is: where does this live in our stack, and who owns the measurement?

ChatGPT operates as a generative interface, not a traditional display environment. Attribution logic that works on Meta or Google — last-click, view-through windows, pixel-based conversion — does not map cleanly onto a conversational ad unit. Before a single dollar goes in, your analytics architecture needs to account for a new surface that sits upstream of the purchase funnel in ways most current setups are not built to track. Brands that ran early LINE and Grab ad formats in Southeast Asia without adapting their attribution models learned this expensively. The same trap is open here.

Google Just Changed the Rules Mid-Game, and Most Advertisers Missed It

Almost simultaneously, AdExchanger surfaced something quieter but with more immediate P&L implications: Google Ads has changed how target-based bidding behaves on budget-constrained campaigns. The algorithm now bids more aggressively toward your stated tCPA or tROAS target, regardless of whether historical performance supports that level of aggression. Google did flag the change in advance — credit where it’s due — but the operational consequence is significant.

A campaign running a $10 target CPA that has historically performed at $14 will now see the system push harder toward $10, potentially burning through budget faster and distorting the performance data you’ve accumulated. For anyone running tROAS on Shopee-integrated campaigns or regional e-commerce during a promotional window, this is not a theoretical concern.

The deeper issue AdExchanger identifies is structural: this is agentic behaviour. The system is making autonomous decisions that override observed reality in favour of a stated objective. That’s a meaningful transfer of control — from the media buyer to the algorithm — and it happened via a product update, not a contractual renegotiation.

---

Agentic Advertising Has an Accountability Gap

Place these two stories side by side and a pattern emerges. OpenAI is building an ad platform designed from the ground up around AI-mediated interactions. Google is incrementally shifting its existing platform toward autonomous decision-making. The direction is the same: the system acts, and the human ratifies — or doesn’t notice.

This is not inherently bad. Automation at Google’s scale has delivered real efficiency gains for performance marketers. But the accountability architecture hasn’t kept pace. When a human media buyer makes a bad call, there’s a decision trail. When an algorithm makes a bad call because the objective function was misconfigured, the trail leads back to whoever set the tCPA — which is usually a junior exec who inherited the campaign settings from someone who left 18 months ago.

For Southeast Asian brands operating across multiple markets with varied budget constraints, currency considerations, and platform mixes, this gap is wider. A regional performance lead in Kuala Lumpur managing campaigns across MY, TH, and ID simultaneously cannot manually audit every algorithmic adjustment. The answer isn’t to resist automation — it’s to build governance structures that treat bidding parameters as strategic decisions, not set-and-forget configurations.

What Your Stack Actually Needs Right Now

Three practical moves for teams navigating this shift:

Treat bidding strategy as a governance document, not a campaign setting. tCPA and tROAS targets should be reviewed on a defined cadence — at minimum monthly — with sign-off from someone who understands the downstream revenue model, not just the platform interface. When Google changes how those targets behave, you need a human who knows what the original intent was.

Build a measurement layer that doesn’t depend on any single platform’s reporting. ChatGPT’s ad environment, Google’s evolving auction mechanics, and Meta’s continued signal limitations all point to the same conclusion: first-party data infrastructure and platform-agnostic analytics are no longer optional. If your attribution model is essentially the Google Ads dashboard, you are flying in a plane with one instrument.

Pilot ChatGPT placements with an explicit measurement hypothesis before scaling. Define what success looks like in terms your existing stack can actually observe — branded search lift, direct traffic, CRM-attributed conversions — before the creative brief is written. In markets like Indonesia and Thailand where ChatGPT usage is growing but purchase behaviour still skews toward Shopee and Tokopedia, the funnel connection will not be obvious.

The question worth sitting with: as advertising platforms become increasingly agentic, is the role of the media buyer shifting from decision-maker to objective-setter? And if so, who in your organisation is qualified to set objectives that an AI will execute autonomously at scale?

---

At grzzly, we spend a lot of time inside stacks that have accumulated tools faster than they’ve built governance. The shift toward agentic advertising makes that gap more expensive. If your team is trying to figure out how new AI-native ad surfaces fit alongside your existing programmatic setup — or whether your bidding configurations are actually aligned with your business targets — we’re the right conversation to have. Let’s talk