ChatGPT Ads Reach 89 Countries: The 26 New Markets

OpenAI's October 21, 2026 announcement says it is "expanding ChatGPT Ads to 26 additional countries, extending availability to 89 countries worldwide". It follows the Europe (August) and Southeast Asia and Taiwan (September) expansions, and it reaches further into Latin America, Africa and South Asia than any earlier step. The same post gives the scale so far: more than 1.2 billion people use ChatGPT each week, and since ads launched "just over eight months ago", over 100,000 advertisers have run campaigns. Both numbers are OpenAI's own.

This post records the expansion and what it means for an advertiser in a newly opened market. The running list of every market and date lives in our maintained ChatGPT Ads market ledger, which will carry these 26 when it is refreshed from OpenAI's help-centre country list. Sources were read on October 21, 2026 (CEST). Where OpenAI names a country we name it; where it does not, we say so.

The announcement is short. Its substance is four statements: 26 countries added, 89 in total; the additions follow the Europe and the Southeast Asia and Taiwan rollouts; the two buying routes are technology partners and self-service Ads Manager; and ads appear only on the Free and Go plans. Around those sit the figures OpenAI wants quoted, 1.2 billion weekly users and more than 100,000 advertisers since launch, and a restatement of its ads principles: conversations stay private from advertisers, no customer data is sold, ads are labelled and separate from answers, and advertising does not influence what ChatGPT says.

What the announcement does not contain is as useful to know. No full country list. No launch date per country beyond "today". No pricing, no format availability by market, no language support statement, and no mention of which partners serve which regions. The "what's next" paragraph names the categories OpenAI intends to build, formats, objectives, buying options and measurement, without dates. An advertiser planning a budget for a new market is therefore working from the same facts as everyone else: the country is open, the self-serve route exists, and the audience is the Free and Go tiers.

The largest block is Latin America. After Mexico and Brazil, which opened earlier, OpenAI says it is bringing ChatGPT Ads to "14 more Latin American countries, including Argentina, Chile, Colombia, and Peru". That names four and leaves ten more countries in the region unnamed. The second sentence covers the rest of the world in one breath: more markets across Europe, the Middle East, Africa and Asia, "including Nigeria, South Africa, Serbia, Ukraine, Cambodia, Bangladesh, and Pakistan".

The example of why the gap matters: an agency in Lima can start today with confidence, because Peru is named. An agency in Quito or Montevideo cannot tell from the announcement alone, but OpenAI's help-centre Ads Manager availability page, read October 21, lists 89 countries as available for self-service, Ecuador and Uruguay among them; that page, not the announcement, is the check for an unnamed market. The failure case is a plan built on a list copied from a news story that filled the gaps; OpenAI's own pages are the only sources we treat as authoritative for a market being open.

The announcement gives two routes for the new countries. Advertisers can go through technology partners or use self-service access through Ads Manager, OpenAI's own buying interface; OpenAI's help centre adds that larger advertisers can contact its Ads team for sales support. The announcement does not say the self-serve route opens later than the partner route in the new countries; both are stated as available. It also does not name the partners active in any region.

For a local business or a regional agency the practical reading is that Ads Manager is the starting point: OpenAI's help centre requires the legal entity that advertises and is billed to be based in a country listed as available, and it is the route that makes a small test possible. The failure case is treating a new market like an existing one on day one. Bid competition, available formats and the conversation mix in Spanish, Portuguese, Bengali or Urdu will differ from the US, and the first weeks of a newly opened market are the period when OpenAI publishes least about it. Our CPA bidding guide covers when to move budget from manual to target-CPA, and the answer in a new market is: not in the first fortnight.

The rule is unchanged and worth restating because it defines the audience: "Ads will appear only on the Free and Go plans. Plus, Pro, and Enterprise subscriptions will remain ad-free." OpenAI frames the trade plainly, that advertising supports free and low-cost access. For an advertiser that means the reachable audience in any country is its Free and Go users, and OpenAI does not publish how large that share is per market.

Two consequences follow. In the new markets, where Go is the affordable paid tier and Plus is priced for a smaller segment, the ad-eligible share of users is plausibly higher than in the US, but that is an inference and OpenAI gives no number, so we print none. And the people you reach are, by definition, not the ones paying for the most capable models, which matters for B2B advertisers more than for consumer brands. Our plan comparison lays out what Free and Go users get, which is the product your ad sits inside.

On the same day OpenAI published an engineering post on how the ad stack works, and it answers the question advertisers in every market ask first: how does an ad end up under a conversation? The short version is that a custom, post-trained language model OpenAI calls the orchestrator reads the conversation and runs the whole flow through tool calls. It rewrites the conversation into one or more commercial-intent queries, retrieves a broad pool of candidate ads from semantic indexes, filters them for eligibility (campaign status, targeting, policy, budget, pacing, placement fit and per-advertiser deduplication), scores them for relevance and predicted engagement, nominates the plausible ones for an auction, and the auction tool applies pricing and final selection. The orchestrator does not set the price and does not pick the winner.

Three design choices matter to a buyer. The ads session is stateful: it remembers context across turns and across conversations, so intent that develops over several messages, the post's example is a running conversation that only later turns to shoes, can be matched without re-running everything on each turn. The ad flow runs in its own service with its own models and code, so ChatGPT's answer has no dependency on it; if the ad system is slow or down, the answer still arrives, and OpenAI states that returning no ad is "an ordinary, successful outcome". And after the auction there are two more ways an ad fails to show: the orchestrator can block one that feels wrong in context, and the ad can simply arrive too late for the response. The failure case for an advertiser is therefore not losing an auction but never entering one, which is why creative relevance to a conversation counts more here than on a search results page.

Today's expansion is the latest step in a rollout that started with a US test on February 9, 2026. OpenAI announced 31 European countries on August 18; self-serve Ads Manager opened across Europe, India, the Middle East and North Africa on August 31, reaching 52 self-serve countries by early September; Southeast Asia and Taiwan followed on September 23; and today adds 26 to reach 89. The ledger holds each step with its date. The pattern is consistent: OpenAI's posts name the headline markets, and the full list of available countries lives on its help-centre Ads Manager availability page.

The example that shows what has changed is the pace: eight months from a single-country launch to 89 markets, on OpenAI's own dates. The failure case is reading pace as maturity. The product in a market opened today has the formats, objectives and measurement of October 2026, which OpenAI itself describes as the next phase of work, and our attribution playbook explains what can and cannot be measured today.

A newly opened market is an argument for a small, structured test rather than a budget shift. The routing below is our recommendation for the first four weeks, built on what OpenAI's two posts say about how ads are selected and who sees them. The common thread is that creative relevance to a conversation is the lever, because the orchestrator can decline to show anything, and that measurement should be set up before spend, because the measurement tools in a new market are the ones that exist today.

For advertisers that want this run as a managed test with the reporting built first, our paid media team can stand up a new-market pilot with the measurement in place before the first spend.

Five things an advertiser would want are not in either post. The full list of 26 countries. Pricing or benchmark costs for any market, which OpenAI has not published and which we will not fill from third-party estimates. Format and objective availability by country. Language support for ad creative and for the orchestrator's intent matching outside English. And the partner roster by region. Each is a reason to confirm in Ads Manager rather than to plan from the announcement, and each is a line we will add to the ledger when OpenAI states it.

There is one more thing to watch. The engineering post was written by the team that built the stack, and it describes the system as self-improving, with Codex used to build and tune it at speed. That is a statement about how fast the product will change, which is the best argument for a short, measured test now and a second look in a quarter rather than a large commitment on day one.