ChatGPT Ads Expansion: 26 More Markets, 89 Total
OpenAI says ChatGPT Ads now spans 89 markets and describes the separate systems that retrieve, rank, and serve sponsored placements.
OpenAI announced a ChatGPT Ads expansion on October 21, adding 26 markets and bringing its advertised availability to 89. A second post the same day explains the ad-selection stack. Together they describe broader advertiser access and how sponsored placements are selected separately from ChatGPT answers. Neither post gives a per-market consumer rollout schedule or guarantees that an eligible campaign will appear.
ChatGPT ads expansion: what the 89-market figure means
OpenAI says more than 100,000 advertisers have run campaigns since launch. The current Ads Manager availability table is the place to check the full market list; the announcement does not print all 26 additions. Argentina, Nigeria, Pakistan and Uzbekistan appear as available in that table. Taiwan was already available. Our September Southeast Asia and Taiwan report covers that earlier change.
The support page separates market availability from advertiser eligibility. Self-service access depends on the advertiser’s country, and the legal entity being advertised and billed must be based in an available country. For newly added European markets, OpenAI says businesses can currently start through Ads Solutions, agencies, or technology partners. An available market row is not proof that every advertiser can open a self-service account. The announcement does not disclose campaign pricing, auction thresholds, guaranteed impressions, or a country-by-country consumer start date.
Which ChatGPT plans may see ads
The expansion announcement says ads appear on Free and Go, while Plus, Pro, and Enterprise remain ad-free. The current Ads in ChatGPT FAQ adds Business and Edu to the ad-free plans. Eligible Free and Go users in selected regions may see placements as rollout continues, so country availability does not mean every account already sees ads. Accounts identified as under 18 are excluded; Temporary Chats do not show ads. The FAQ also excludes sensitive-topic placements and says political ads are not allowed.
Free users have a separate Ads-Free option that comes with lower message limits and reduced feature access. It is not a Go-plan option. Keep these categories distinct: a market can be advertiser-available; an account can be on a plan that may show ads; and an individual placement can still fail a topic, policy, or delivery check. None alone promises an impression. OpenAI says ads are labeled and separate from answers, do not influence responses, and do not expose conversations to advertisers; it says customer data is not sold. Our October visual-ads and measurement report covers a different U.S. test.
How the ChatGPT Ads stack selects a placement
The engineering post describes a custom, post-trained frontier LLM orchestrating retrieval, eligibility filtering, relevance scoring and an auction. It publishes no exact model ID. A final delivery check can still reject the result; an ad that arrives too late is not displayed.
The sponsored placement has advertiser creative and a neutral model-written preamble. OpenAI says answer generation runs in separate services, models and code. Returning no ad counts as success; an ads outage does not hold up the answer.
A practical way to read access, eligibility, and delivery
The following is our editorial interpretation, not an OpenAI workflow or guarantee. Treat an ad opportunity as passing through four checks: advertiser market access, user plan and regional eligibility, suitability of the conversation and placement, and successful delivery before the response is ready. A failure at any stage can mean no ad. This avoids reading 89 markets as universal user exposure or guaranteed campaign delivery.
For an internal launch review, record each check separately rather than collapsing it into “available.” A compact log might include the date checked, advertiser legal-entity country, access route, audience market, documented plan eligibility, observed policy exclusion, and delivery status if available. This is an analyst-created checklist, not an Ads Manager schema. Do not infer private user attributes or inspect another person’s conversation.
For a campaign in a newly listed market, verify the live table and billing-entity route first. Then apply current help rules to the intended audience and placement. Treat setup, auction selection, and a user actually seeing an ad as different stages; if a dashboard does not disclose one, mark it unknown. This record helps teams avoid inventing a rollout date or promising volume. For example, a team can approve an advertiser’s billing entity while leaving consumer exposure unconfirmed. In that record, access is verified and reach remains unknown. That supports a launch decision without guessing how many people saw a placement.
What the two announcements leave open
The October 21 posts do not enumerate all 26 additions in the blog or provide per-country consumer rollout dates, auction formulas, bid thresholds, model identifiers, latency targets, or guaranteed placements. The engineering post explains system stages but not numerical weights or an advertiser-level prediction.
OpenAI says additional formats, objectives, buying choices, measurement capabilities, and more native business interactions are future development or exploration. Talking directly to a business through an ad should not be described as launched by these posts. The stack description is not a new named model release. Our Sponsored Agents report covers a separate advertiser announcement; it does not turn the future-facing language here into a generally available feature.