Ads in ChatGPT: OpenAI’s New Advertiser Toolkit

Ads in ChatGPT took a quiet but consequential step on or around June 17, 2026, when OpenAI published new Ad Tools Terms. The real news is not that ads exist — they have since February — but that OpenAI now has the legal scaffolding for first-party audience uploads and AI-generated ad creative, the two capabilities that separate a test from a permanent advertising business.

Most coverage in mid-June 2026 is still recapping the February launch. The new angle sits in lawyer-written language: the Terms define Audience Tools and Creative Tools, set out what data is barred and who carries liability, and include a data-reuse clause that few outlets have surfaced. Crucially, defining a tool in the Terms is not the same as shipping it — as of June 20, neither feature was confirmed live in the Ads Manager.

This guide reads the actual terms, separates announced from launched, recomputes the platform’s evolution across every milestone so far, and gives a media buyer a sober view: is this ready for budget, and on what terms. Every figure below is sourced and hedged where the primary source is secondary.

On or around June 17, 2026, OpenAI published its Ad Tools Terms at openai.com/policies/ad-tools-terms, introducing two optional features for advertisers: Audience Tools, which let an advertiser upload first-party customer data to build custom audiences, and Creative Tools, AI-powered features that generate, modify, localise, or translate ad creative from a brand’s own materials. Digiday covered the Terms update the same day; Storyboard18 framed it as moving OpenAI closer to established platforms like Meta and Google.

The careful reading matters here. The Terms define these tools and the rules around them — they do not, on their own, confirm that either feature is switched on inside the Ads Manager. As of June 20, 2026, there was no confirmed public launch of Audience Tools or Creative Tools as active, usable features. The right framing for a media plan is that the platform has published the legal groundwork for its next phase, with availability pending.

The reason this is worth a marketer’s attention now, before the features flip on, is access timing. OpenAI built the rest of the stack quickly: the pilot crossed $100 million in annualised revenue in under six weeks, the self-serve Ads Manager opened with no minimum spend, and CPC bidding and conversion tracking followed. If audience matching and AI creative arrive on the same cadence, the brands that have already cleaned their data governance and run instrumented tests will move first. Our paid media team treats this as a readiness exercise, not a launch-day scramble.

Audience Tools, as defined in Section 2 of the Terms, let an advertiser upload first-party customer data — customer identifiers, suppression lists, and audience segments — to create custom audiences for campaigns. The Terms call this Audience Data. The model is familiar from Meta Custom Audiences and Google Customer Match, which is exactly the comparison OpenAI is inviting. What stands out is the specificity of the restrictions.

Two hard limits define the boundary. First, the Terms set out an explicit Prohibited Data list. Second, data sourced from brokers, data marketplaces, or third-party suppliers is barred — Audience Data must be your own first-party data. OpenAI also states it does not guarantee audience match rates, reach, campaign performance, or delivery, and reserves the right to reject, suppress, or modify any audience segment you submit.

For a brand, the practical work is upstream of the platform. Before any audience upload, your customer data needs a clean consent trail, a documented lawful basis, and a way to honour suppression and deletion requests — the same hygiene that a well-run CRM should already enforce. If your customer records are stitched from purchased lists or unconsented enrichment, the Terms put that squarely off-limits, and the time to find out is now, not after you’ve uploaded.

The most consequential line in the Terms is also the least covered. Section 2.2 allows audience data to be used by OpenAI and its affiliates for the development, provision, and improvement of their products or services — the Terms call this the Access Purpose — in addition to enabling your custom audiences. Advertisers must agree to this to use Audience Tools. No major outlet had surfaced the clause by the time the Terms went public.

Read it precisely. This is not OpenAI selling your data — the Terms bar external sharing of Audience Data. It is OpenAI reserving the right to use uploaded audience data internally, for product improvement, beyond the immediate job of delivering your ads. For a privacy-conscious brand, that is a material consideration: the data you upload to target a campaign may also feed the platform’s own development. Whether that is acceptable is a governance decision, not a technical one — and it belongs in your data-protection review before a single record is uploaded.

Creative Tools, defined in Section 3, are AI-powered features that let an advertiser generate, modify, transform, optimise, localise, or translate ad creatives from Ad Materials — product catalogues, website content, images, text, videos, logos, and other brand assets. The pitch is speed: proven assets ready to deploy lower the barrier to experimenting with a new channel. The fine print is responsibility.

The Terms are blunt about where liability sits. Advertisers bear full responsibility for reviewing AI-generated creatives and for verifying pricing, availability, testimonials, endorsements, and performance claims before publishing. Generated Creatives may not be used outside the Advertising Services unless OpenAI expressly permits it. And the restricted-use list bars unauthorised digital replicas, impersonation of a person or organisation, deceptive synthetic media, and any use of Prohibited Data.

That single sentence reshapes the workflow. If a Creative Tool generates an ad that overstates a discount, misquotes availability, or invents an endorsement, the exposure is the advertiser’s. The efficiency of AI generation does not remove the review step — it relocates it. Brands that treat AI creative as a draft to be fact-checked against the live catalogue will be fine; brands that treat it as a publish-ready output are accepting a risk the Terms have explicitly handed back to them.

Most coverage treats each announcement in isolation. The table below synthesises the whole evolution from the February 9, 2026 pilot to the June 17 Ad Tools Terms, with the elapsed days recomputed from the launch date. The pattern is a platform maturing fast — and a final row that is policy-announced rather than confirmed live.

The cadence is the signal. The platform went from a labelled sponsored placement to a self-serve manager with conversion tracking in roughly twelve weeks, and added the legal scaffolding for audience and creative tools four weeks after that. If you read the trajectory forward, the question is not whether audience matching and AI creative arrive, but when — and whether they land before the click-through economics improve enough to justify scaled budgets. For a comparison of how this CPC model stacks up against the incumbent auction, our breakdown of Google’s June 2026 bidding changes is the natural companion read.

The headline performance number is uncomfortable. Advertisers report click-through rates as low as 0.91% per Adweek, against a Google Search benchmark of 6.4%. Two honesty notes before anyone forecasts on it: the 0.91% figure is advertiser-reported data relayed via Adweek, not an OpenAI-published benchmark — OpenAI has stated it has no published cross-advertiser performance figures — and the comparison is to a fundamentally different surface. Treat the number as directional, not as a planning input.

Here is the interpretation most coverage misses. The gap is unlikely to be a creative problem you can fix with better headlines, because it reflects how people use the two surfaces. A search results page is a launchpad — users arrive expecting to click out. A conversational interface is a destination — users arrive expecting an answer in the thread, and a sponsored link at the bottom competes with a response that has already satisfied the query. That is a structural difference in intent expression, not a tuning problem. Better creative will move the number; it is unlikely to close the gap.

Projecting forward, the channel’s case rests on two things improving together: the click economics, and the targeting precision that Audience Tools is meant to bring. If audience matching lands and sharpens relevance, a low CTR on a high-intent, well-targeted impression can still pencil out — especially at a $3–$5 CPC with no minimum spend. The brands learning that math now, on small budgets, will be the ones with a defensible position when the tools and the audience both scale.

No published comparison existed of OpenAI’s announced audience and creative tools against the established Meta and Google equivalents as of June 20, 2026. The table below maps the dimensions that matter for a brand-safety and data-governance sign-off. The OpenAI cells come from the Ad Tools Terms; remember that the OpenAI tools are announced in policy, not confirmed live.

Two cells should stop a governance reviewer. The Access Purpose row — data use beyond targeting — is the OpenAI differentiator that established platforms scope more narrowly to ad delivery and measurement. And the explicit no-guarantee on match rates is unusual to see stated so plainly. Neither is a dealbreaker; both are decisions a brand should make deliberately rather than discover after an upload. This is the same competitive shift we trace across the AI interfaces in our overview of the emerging AI search advertising market, and it mirrors how Meta is building AI-powered ad surfaces of its own.

The barrier to a first test is low: the self-serve Ads Manager is open with no minimum spend, CPC bidding around $3–$5, a $200 daily cap, and a Conversions API plus pixel tracking so you can measure post-click outcomes rather than guess. The matrix below maps a starting posture by where your brand sits today.

Whichever posture fits, the discipline is the same: keep budgets small, instrument every campaign with conversion tracking, review AI creative against your live catalogue, and put the Access Purpose clause in front of legal before uploading audience data. The brands that win the channel will be the ones that learned its economics cheaply and governed its data carefully — not the ones that waited for a press release. If you want a partner to stand up that test and its governance, that is exactly the work our paid media engagements are built for.