OpenAI opened its self-serve Ads Manager to all US advertisers on May 5, 2026. That move eliminated the $50,000 minimum spend that had kept most buyers out during the managed-access period, added CPC bidding alongside the existing CPM model, and launched a Conversions API alongside pixel-based measurement. Four months after the company started running any ads at scale, ChatGPT has over 2,000 brands active via Criteo alone. The question for programmatic buyers is not whether to pay attention — it is how to read what the platform actually is before deciding how much to bet on it.

What OpenAI Built and What It Is Borrowing

OpenAI’s current ad infrastructure is a hybrid of its own tooling and borrowed plumbing. The self-serve Ads Manager at ads.openai.com runs CPC bidding natively. The programmatic access layer runs through partner integrations: Criteo, StackAdapt, Pacvue, Kargo, and Adobe can all route spend into ChatGPT inventory through their existing platforms. That means a buyer can technically add ChatGPT as a placement inside a StackAdapt campaign without opening a separate OpenAI account.

This is a deliberate design. OpenAI is following the same playbook Netflix and Walmart used before building their own ad stacks — partner first to prove the demand signal and build the inventory case, then internalize once the revenue is large enough to justify the infrastructure. Digiday reported that OpenAI’s ad tech partnerships are a means to an end, and the better those partners perform, the stronger OpenAI’s business case for eventually owning the stack itself. Partners understand this. They are participating anyway because early access to a new inventory surface — before CPMs get bid up by competition — is typically worth the eventual displacement risk.

As of June 2026, OpenAI published Ad Tools Terms defining two additional advertiser features: Audience Tools for first-party data upload, and Creative Tools for AI-generated ad production. Neither was confirmed as an active, publicly available feature by the time of publication, though the terms represent OpenAI signaling the direction.

The Inventory and Performance Reality

Early performance data is genuinely interesting and appropriately uncertain at the same time. Criteo reported that AI-referred conversion rates in its client base approach twice those of traditional search in several retail categories. StackAdapt dropped its spend minimum to zero and reported similar engagement signals. Over 600 advertisers registered for the platform in early testing, crossing $100 million in annualized revenue within weeks of the managed beta opening.

The caveats matter. Criteo and StackAdapt are not neutral measurement sources — their business depends on ChatGPT ad success, which makes their performance data structurally optimistic. The 70% traffic-share figure cited in Criteo’s pitch deck came from the pitch deck, not third-party attribution. Inventory is still thin relative to what a scaled programmatic buyer expects — early advertisers during the high-minimum period reported limited fill at the $60 CPM starting point, though fill rates improved as OpenAI built out relevance scoring.

CPM and CPC benchmarks will compress as more advertisers enter and competition for available impressions increases. Buyers moving early get the better pricing environment. Buyers moving before measurement infrastructure matures get the attribution uncertainty that comes with any new channel.

What Makes ChatGPT Inventory Structurally Different

The case for ChatGPT inventory is not that it reaches more people than Google or Meta — it does not. The case is that the intent signal available inside a ChatGPT conversation is qualitatively different from a keyword. A user asking ChatGPT a multi-paragraph question comparing enterprise software vendors is disclosing intent depth, competitive awareness, and decision stage in a way that no keyword query can match. That signal is valuable for high-consideration purchases where the sales cycle is long and audience quality matters more than volume.

OpenAI used Cannes Lions 2026 to tell media planners that ChatGPT should be treated as a gatekeeper-level platform alongside Google and Meta. The projection underpinning that claim — $2.5 billion in ad revenue this year, $100 billion by 2030 — is aggressive. Google built a $224 billion annual search ad business over two decades. Meta built its $200 billion ad business in roughly 17 years. OpenAI is projecting to match Google’s scale in four. That is possible in the same sense that any hockey-stick revenue model is possible in its early innings. It is not inevitable.

EMarketer’s projection for AI-driven search ad spending in the US is more measured: roughly $1.1 billion in 2025 to $26 billion by 2029 across the category, not just OpenAI. The total addressable market is real. OpenAI’s share of that market is unknown.

Where This Leaves Programmatic Infrastructure

The emergence of ChatGPT as a formal ad channel has implications that extend beyond the OpenAI budget line. If ChatGPT captures meaningful search ad share — even a fraction of the Google search ad market — it shifts where intent-based budgets flow. That is a DSP optimization problem: platforms that integrate ChatGPT inventory cleanly will attract budget that previously went into search. Platforms that do not will see planned search allocations reappraised.

The AdExchanger question — where does a new walled-garden AI surface leave existing DSPs and SSPs? — does not have a clean answer yet. OpenAI’s current partners get preferred access now. What OpenAI builds over the next 18 months determines whether those partnerships become lasting channels or transition agreements.

For SSPs, the near-term reality is that ChatGPT inventory is not in the open programmatic ecosystem. OpenAI controls delivery decisions even when buyers access through partner platforms. There is no supply-side bid stream, no OpenRTB integration, no exchange dynamic. That is a deliberate call — the same walled-garden architecture that keeps CPMs defensible on the surface also keeps SSP infrastructure out of the loop.

The Practical Buyer Checklist

Before allocating budget to ChatGPT inventory, programmatic buyers should work through a short set of questions that the current coverage largely skips.

Attribution setup: OpenAI’s Conversions API and pixel-based measurement are live, but they are new. Map your conversion events into the platform before spending, and run a parallel view-through test using your existing attribution stack. Do not rely only on platform-reported conversions during the first 90 days.

Category fit check: ChatGPT is strongest for high-consideration, longer-cycle purchases. Consumer packaged goods, low-margin retail, and impulse categories are not natural fits for the current ad formats or the intent density of the user base. B2B software, financial services, and healthcare are categories where the intent signal justification is most defensible.

Performance baseline: Run a 60-day test with a defined holdout before scaling. The Criteo and StackAdapt performance numbers are directional, not guaranteed. Your actual CTR and conversion data will look different depending on category, creative quality, and audience configuration.

Spend minimum reality: The platform-level minimum is gone, but the practical entry point for getting meaningful data is not zero. A 60-day test at a budget that can generate statistically meaningful conversion volume in your category is the right floor — not what the lowest recommended bid implies.

OpenAI’s ad platform is real, and the intent signal case is legitimate. The measurement infrastructure is early, the inventory is thinner than the marketing volume suggests, and the performance data in circulation is partner-reported rather than independently verified. None of that means avoid it. It means run a structured test, use your own measurement stack, and hold the projections lightly until you have 90 days of proprietary data to work with.