What OpenAI has actually announced (and what it has not)
The signal came through trade press, not a product launch event. In two Digiday articles published on 30 June 2026, OpenAI's head of advertising David Dugan outlined the company's approach: an open-platform build, third-party measurement as a design principle, and a format roadmap that goes well beyond the limited sponsored-content test currently running with selected partners. A third piece published 1 July 2026 confirmed that OpenAI is hiring San Francisco-based software engineers specifically to build text, image, video, native, conversational, and interactive ad formats.
What is live as of July 2026: a narrow sponsored-content pilot with a small number of partners, not a self-serve buy. What is planned: a broader platform with multiple format types, third-party measurement integration, and an open-ecosystem approach where agencies and measurement vendors co-build with OpenAI rather than receive a finished product. The distinction matters because media-plan decisions for 2027 budgets happen now, and the risk of over-indexing on a platform that is not yet available to most buyers is real.
Dugan's framing on the build approach, as reported by Digiday, is that there are "two ways to build these platforms." OpenAI is choosing the open path: inviting the ad ecosystem in early, sharing the architecture with measurement partners, and deferring some hard questions rather than shipping a closed system. That is a meaningful signal about the company's intent to work with existing ad-industry infrastructure rather than replace it.
| Dimension | Current state (July 2026) | Forthcoming plans |
|---|---|---|
| Access | Limited pilot, selected partners only | General availability (timeline unconfirmed) |
| Formats | Sponsored content (text-adjacent) | Image, video, conversational, native, interactive |
| Measurement | Platform-reported only | Third-party measurement: described as "a natural step" |
| User data shared | No conversation data shared | No conversation data shared (stated principle) |
| Ecosystem approach | Closed pilot | Open: agencies and measurement vendors co-build |
Three formats: what image, video, and conversational actually mean for buyers
Two of OpenAI's three planned formats are familiar. One is not.
Image and video ads are conventional digital formats adapted to a new surface. An image ad is a static visual unit served within or alongside a ChatGPT response. A video ad is motion creative in the same context. Both formats have established production workflows, measurement methodologies, and buying infrastructure. The challenge is not the format, it is the context: a user inside a ChatGPT conversation is in a different intent state than a user scrolling a social feed, and creative that ignores that context will underperform.
The conversational ad is the category that does not have a direct analogue elsewhere. Rather than appearing in a fixed display slot beside the content, it participates in the dialogue itself. The ad content surfaces as part of what ChatGPT says. For a user asking "what travel insurance should I look at for a trip to Japan," a conversational ad format would have a brand's response surface as part of the answer stream, not as a banner running alongside it. This is why Dugan's framing is significant: ChatGPT ads are not just another display surface. The format architecture is different.
- What it is: a static visual unit served within or adjacent to a ChatGPT response.
- Production workflow: standard display creative. Existing assets can be adapted.
- Measurement question: viewability standards for a chat interface have not been published. The MRC (Media Rating Council) has not released ChatGPT-specific standards as of July 2026.
- Buyer implication: lowest execution lift from existing programs. Highest risk of misfit creative: a banner built for a scrolling feed reads differently in a conversational context.
The measurement trap: why third-party verification is harder here than on Google or Meta
David Dugan described third-party measurement as "a natural step" in his 30 June 2026 Digiday interview. The context in which he said it matters: OpenAI will not share conversation data with advertisers, because doing so conflicts with the company's user-privacy principles. External verification is not just a buyer request. It is the only path to independently confirming what ran and what converted, given that the richest intent signal sits behind a privacy wall.
This is structurally different from the Google or Meta measurement problem. On those platforms, the issue is that the platform grades its own homework and has a revenue incentive to show positive numbers. The platform data is available, it is just self-interested. On ChatGPT, the data that would let you understand what a user was asking when your ad appeared is not available at all. Third-party measurement vendors will receive a constrained signal set: delivery confirmation, some contextual category data, and outcome signals from your own first-party systems. They will not receive conversation content.
| Platform | User-level data to advertisers | Contextual signal | Third-party verifier access |
|---|---|---|---|
| Google Ads | Limited (post-ATT, privacy sandbox changes) | Search query (partial) | IAS, DV, MOAT supported |
| Meta Ads | Aggregated (Privacy Enhancing Tech) | Interest/behaviour categories | IAS, DV, Nielsen supported |
| ChatGPT (planned) | Not shared (stated principle) | Topic/category only (inferred) | "A natural step" per Dugan; specifics unconfirmed |
The practical implication: design your measurement infrastructure now, before inventory is available. The analytics and insights work is the prerequisite, not the follow-on. Build independent outcome measurement, first-party conversion tracking, and incrementality testing frameworks before you need to evaluate whether a ChatGPT placement is working. By the time the buy is live, it is too late to retrofit the measurement layer.
Regulated verticals: who is excluded and what to do instead
OpenAI has not published a final category exclusion list as of July 2026. What is predictable, based on how every major platform has handled regulated categories at launch, is that financial services and health will face the most significant restrictions. Google's mandatory financial-advertiser verification program expanded across Europe in 2026, having required verification in the US and other markets earlier. Meta and TikTok both impose category-level approval processes for financial products. OpenAI will almost certainly follow a similar sequence: open to most categories initially, then impose verification and restrictions as regulators take interest and brand-safety incidents accumulate.
For leapbuzz clients in banking, insurance, and fintech across Singapore, Australia, the US, and Canada, the near-term implication is direct. The ChatGPT ad surface is probably not available to regulated financial products at launch, and may not be for 12 to 18 months after general availability. The MAS (Monetary Authority of Singapore) Notice FAA-N03 framework, ASIC's product-advice rules in Australia, and equivalent frameworks in the US (FINRA) and Canada (IIROC) create a class of required disclosure and human-oversight obligations that conversational ad formats are not yet designed to satisfy.
The right 2026 play for regulated-sector advertisers is not to wait for a ChatGPT ad slot. It is to build organic AI-citation authority through Generative Engine Optimisation (GEO) now, so the brand appears in ChatGPT answers without needing to pay for placement. A bank or insurer that is cited reliably in ChatGPT's organic answers on relevant intent queries is already present in the platform before the first regulated-sector ad slot exists. That organic position costs less per impression and is harder for competitors to displace than a paid slot.
The companion post on GEO for leapbuzz clients covers the content and schema optimisation steps. For fintech and insurance buyers specifically, the most tractable GEO signals are authoritative explainer content, structured FAQ markup, and named citation from trust-signal sources. These take 60 to 90 days to index and propagate through AI engines. Starting now means having organic presence when the paid surface opens.
Planning for your 2027 media mix
The structural question for 2027 budget planning is not "should we be on ChatGPT." It is "at what allocation, with what measurement infrastructure, and alongside what organic-citation program." Media planning decisions for H1 2027 happen in Q3-Q4 2026. That timing means you need a position now, before the platform is fully visible.
- Allocate a test budget, not a committed line. A 3 to 5 percent experimental allocation in the H1 2027 budget creates the learning opportunity without exposing the broader program to an unproven platform. The objective of the test is measurement data, not volume.
- Define the measurement success criteria before the buy. What outcome would make this test a success? What would make it a failure? Define both before the campaign runs, not after. This is the same discipline required for any new channel activation.
- Build the independent measurement layer in parallel. First-party conversion tracking, tagged landing pages, and an incrementality holdout group should be in place before the first ChatGPT impression runs. These are not ChatGPT-specific. They improve measurement on every channel.
- Run organic GEO alongside the paid test. An AI performance marketing program that combines paid ChatGPT placement with organic AI-citation optimisation gets two signals from the same intent pool. The organic signal validates or challenges what the paid signal shows.
- US-first, then APAC and CA/AU. Self-serve ChatGPT ad access, when it arrives, will follow the US-first sequencing that Google, Meta, and TikTok all used. Singapore, Australia, Canada, and Malaysia buyers will have 3 to 9 months of US market data to learn from before local inventory is available. Use that window.
The firms that will extract the most value from ChatGPT advertising in 2027 are not the ones who buy the most inventory first. They are the ones who show up with measurement discipline, first-party data infrastructure, and organic citation authority already in place. Those are the inputs to build this year.
