ChatGPT ads went self-serve in May 2026. Here is what actually changed for buyers.

What changed on May 5 2026

OpenAI's US pilot launched February 9 2026 with a select group of ad-tech partners. Minimum spend at that stage was reported at roughly the range required for a managed deal. International expansion to Canada, Australia, and New Zealand was announced March 26, with those markets going live in mid-April via managed and agency arrangements. The UK followed. By April the minimum had dropped substantially, still keeping the surface out of reach for most advertisers.

On May 5, OpenAI removed the minimum-spend requirement entirely and opened a self-serve Ads Manager at ads.openai.com to US businesses of all sizes. The same release shipped a conversion pixel, Conversions API, attribution windows, and CPC bidding alongside the existing CPM model. The advertiser interest page is at openai.com/advertisers. That is the actual news cluster: access democratised, and the measurement stack matured, simultaneously.

OpenAI has stated an ad revenue ambition of approximately $2.5 billion for 2026, according to reporting by Search Engine Journal and Axios. Those are targets, not confirmed results. Treat them as signals of strategic commitment, not validated performance.

Three of leapbuzz's five markets can transact today. Two cannot. That asymmetry shapes the APAC operator move covered in section 6.

How ChatGPT ads actually work

Ads appear below the end of a ChatGPT response, clearly labelled as sponsored and visually separated from the model's answer. Per OpenAI's Help Center documentation (updated June 17 2026), one ad unit typically appears per response. That unit may feature one or more items or advertisers.

Ad selection matches to what is being discussed in the current chat thread. With personalised ads enabled, it also draws on past chats, ad-interaction history, and memory. The ranking is relevance to the conversation, not a keyword bid auction in the traditional sense.

The sharpest mechanic is the three-dot menu on an ad: a user can tap it and choose "Ask ChatGPT" to share the ad in-chat and ask questions about it directly. ChatGPT then answers independently. Advertisers do not influence that response. By default, ChatGPT cannot even see which ads were shown to a user when generating its answer.

Why the keyword playbook does not transfer

Google and Microsoft search advertising is built around the keyword as the primary unit of intent. You buy presence on a query. Exact match, phrase match, broad match, single keyword ad groups, responsive search ads, bid strategies tied to conversion signals - the entire discipline is engineered to capture declared intent at the moment of search.

ChatGPT ads do not work this way. The ad selection engine matches to "what's being discussed in your current chat thread," per OpenAI's own documentation. There is no keyword. There is no query. There is a deliberation in progress, and your ad enters that deliberation when the topic matches your campaign parameters.

This changes what good looks like. On Google, a well-structured campaign with tight match types and granular ad copy beats a loose campaign. On ChatGPT, the brand that wins the "Ask ChatGPT about this ad" moment is the one whose underlying content, website, and offer hold up to direct questioning by an independent AI. The ad is a door. The AI-interrogated brand is what waits behind it.

That is Generative Engine Optimization (GEO) applied to a paid surface. Your owned content and schema architecture become the ad's substrate. Brands that have invested in structured, citable, primary-sourced content will perform better on this surface than brands that have not, regardless of budget. The same work that earns organic AI citations in GEO makes the paid placement more credible when interrogated.

The muscle memory of the search-advertising discipline, exact match terms, single keyword ad groups, impression share, Quality Score optimisation, does not transfer cleanly. Build the question-intent map first: what are buyers discussing when they would encounter your brand? That conversational territory, not a keyword list, is the campaign architecture for this surface.

Which sectors cannot buy this yet, and what to do instead

This is the section most of the coverage is missing.

OpenAI's Help Center states two hard exclusions as of June 2026. Per OpenAI's published principles on advertising: "Ads are not eligible to appear near sensitive or regulated topics, including personal health, mental health, or politics." The Help Center adds: "Advertisers in sensitive or regulated verticals such as dating, health, financial services, or politics are excluded from advertising on ChatGPT at this time."

Financial services is in that list. Health is in that list. Both are explicit and unambiguous in OpenAI's own documentation.

This covers the majority of leapbuzz's regulated-sector client base. The question for those sectors is not "how do we buy this?" The question is "how do we appear in AI conversations about financial products, insurance, or fintech when we cannot pay to be there?"

The answer is organic AI citation. GEO and Answer Engine Optimization (AEO) are the tools that earn a brand's place in ChatGPT, Perplexity, Claude, and Google AI Overviews responses without buying an ad. The same AI engine that runs the ad surface also runs the organic citation surface. Regulated-sector brands that own the organic citation surface now are building the only form of AI presence currently available to them.

The GEO end-to-end playbook covers the mechanics. The regulated-sector angle is specific: financial services and health content that is factually dense, primary-cited, and schema-rich gets cited at higher rates because the AI engines need credible sources on high-stakes topics, and thin content from unreliable sources gets filtered out. The compliance discipline that feels like a constraint in paid advertising becomes a content advantage in organic AI citation.

Measurement: what you get, and what you do not

The measurement story changed meaningfully on May 5 2026. OpenAI shipped a conversion pixel, Conversions API (CAPI), attribution windows, and CPC bidding alongside the existing CPM model. This was a significant maturation: advertisers can now close the loop to a conversion event on their own site and feed CPC signal back into the platform.

What has not changed is the user-data boundary. OpenAI's Help Center documentation on ads in ChatGPT is explicit: advertisers do not receive individual chat histories, user memories, names, emails, precise location, or IP addresses. Conversations are never shared with advertisers. Ad data is retained up to 30 days after deletion. All reporting on the user side stays aggregated and non-identifying.

The practical frame: ChatGPT ads are becoming a real performance channel in the sense that on-site conversion tracking and CPC optimisation now exist. They are not yet a mature performance channel in the sense that reach is small versus Google or Meta, the audience skews heavily toward Free and Go plan users, and user-level signal is structurally off the table by design. Fund accordingly.

The practical implication: fund this from an experimental or test budget rather than cannibalising a proven performance marketing line. Conversion tracking now exists, but reach is still small, the audience is Free/Go-weighted, and the channel is too new to carry the ROAS expectations of a mature Google Search or Meta campaign. Allocating budget from those lines to fund ChatGPT ads in Q3 2026 creates an unfair comparison that will push you to defund channels with deeper signal. Keep the performance lines intact.

What you can measure usefully: branded search lift and direct traffic lift in the markets where you run ChatGPT ads, triangulated against markets where you do not. That gives you an independent signal on whether the placement is affecting downstream research behaviour. Bayesian incrementality testing applied to this holdout design is covered in the cookieless MMM and Bayesian incrementality post. The principle applies here. If your measurement infrastructure cannot run a clean holdout, build that capability before you allocate serious budget to a channel that will not give you first-party conversion data.

The five-market operator move

Three of leapbuzz's five markets have access to ChatGPT ads today. US, Canada, and Australia are live. Singapore and Malaysia are not. OpenAI has not published a timeline for APAC expansion beyond the five live markets.

The wrong response to the Singapore and Malaysia gap is to wait. The right response is to use the asymmetry as a planning advantage.

Brands that enter a new ad surface when it launches in their home market typically spend the first two to three months learning the surface mechanics, not generating returns. Question-intent mapping. Landing page adaptation. Creative iteration. Bid strategy calibration. Audience exclusion logic. All of that takes time on any new platform.

US, Canadian, and Australian subsidiaries or divisions of APAC-first brands have the opportunity to run that learning curve now, at low cost, before Singapore and Malaysia go live. When the home market switches on, the team that has already run two quarters on the surface will activate with judgment, not with hypotheses. That is a genuine competitive advantage over brands that wait for their home market.

• Build the question-intent map now. What are buyers in your category discussing in ChatGPT threads in the months before they convert? Commission qualitative research and use the findings to structure campaign topic targeting before the surface exists in Singapore.
• Adapt your landing experience. A user who tapped "Ask ChatGPT" on your ad before clicking has already interrogated your brand independently. The landing page should acknowledge that research state, not restart the awareness funnel. This is a website development brief, not just a media brief.
• Run the surface in live markets. Even a modest test budget in the US or Australia generates real platform knowledge: how the ad selection engine interprets topic targeting, which creative formats hold attention, how the "Ask ChatGPT" mechanic changes downstream engagement. That knowledge transfers when Singapore goes live.
• Invest in GEO for Singapore and Malaysia now. The organic AI-citation surface is live in both markets today, regardless of whether the paid surface is. Brands that own organic AI citation in Singapore and Malaysia when paid ChatGPT ads arrive will have a structural head start.

The brands that treat the geography gap as a pause will learn the surface and the market simultaneously when Singapore opens. That doubles the cost of the learning period.

Who this channel is actually for

One structural fact limits the B2B case for ChatGPT ads in a way that most coverage is not addressing. Per OpenAI's own documentation: ads appear only for Free and Go plan users. Plus, Pro, Business, Enterprise, and Edu subscribers see no ads.

That is significant for enterprise B2B. The buyer persona that signs software contracts and services agreements typically sits on a Business or Enterprise plan, either as an individual subscriber or via a company-provisioned account. Those buyers are on ad-free tiers. They will not see ChatGPT ads.

The channel fits better for:

• High-consideration B2B or prosumer purchases where the buyer researches on Free or Go before escalating to an enterprise conversation. Early-funnel consideration plays where the buyer is still personally subscribed.
• Considered consumer purchases with a research phase. Travel, automotive, property, education, consumer technology. Categories where the buyer spends time in ChatGPT exploring options before making a decision.
• SMB-directed B2B where the buyer is a small business owner or founder on a personal Free or Go subscription making their own purchasing decisions.

This does not mean enterprise B2B brands should ignore the surface entirely. Brand presence in AI conversations has value independent of whether the exact buyer in that session is on a Free plan. But the direct-response expectation (ad impression, click, pipeline entry) is weakest for brands whose ICP lives on Enterprise plans. Set the objective accordingly.

The strongest immediate fit is mid-market B2B with a research-heavy buying cycle, and high-consideration consumer categories in the live markets. For pure-enterprise B2B in regulated sectors, the work is GEO and organic AI citation, not paid ChatGPT placements, for at least the next two to four quarters.

For AI marketing strategy that accounts for this channel alongside your existing mix, the starting point is a channel-fit audit: which buyer personas are on Free or Go plans, and what are they asking during their research phase? That audit shapes the ChatGPT ads brief, the GEO brief, and the relationship between them.