ChatGPT Ads Reaches 31 European Markets With No Custom Audiences: What You Can Buy

The ads reached 31 countries. The ad accounts did not.

A performance lead in Madrid goes looking for a ChatGPT Ads account on the morning of Monday 24 August, budget already signed off, because ChatGPT Ads went live in Spain that day. What OpenAI offers her is an interest form, for advertisers "located outside of these countries" who want to "register your interest and stay informed about future availability." That is the European launch in one screen.

OpenAI turned on ad inventory across a very large slice of Europe in the week of 24 August, and its own announcement is precise about the size of it: "Next week, ChatGPT Ads will expand to 31 European countries, including Germany, France, Spain, Italy, Sweden, Norway, Denmark, the Netherlands, and Austria." The post is dated August 18, 2026, which puts the go-live on Monday the 24th. Buying runs through OpenAI's sales and partner channels for now.

The constraint that actually shapes a media plan is not the account, though. It is one sentence in OpenAI's developer documentation, and it decides what any campaign pointed at these markets is allowed to do: "Custom audiences are not supported for campaigns targeting the European Economic Area (EEA) or Switzerland, where personalized ads are not yet available."

Hold that next to the map and the story tightens into something worth planning around. Every market that opened that day is inside that carve-out. The audience layer is off in all of them, which leaves geography, the conversation on screen, and whatever you put in the ad. If you want the wider picture of how advertising inside ChatGPT works, start there. This is the version of it Europe gets first, and it is a different job.

What actually turned on across Europe on Monday

Three facts define the launch, and each one narrows what the channel can be used for.

The 31 markets are exactly the EEA plus Switzerland

Count the set and the scope stops looking approximate:

• 27 EU member states
• plus Iceland, Liechtenstein and Norway, which completes the European Economic Area at 30
• plus Switzerland, which brings it to 31

The United Kingdom is not in the group, because ChatGPT Ads already ran there.

That is why the developer documentation reads like it was written for this launch. The carve-out and the expansion describe the same territory, so there is no European market where ads are live and personalization is also available. The map and the restriction are one shape.

Ads reach the Free and Go plans, and nobody else

OpenAI states the audience boundary plainly: "As in our existing markets, ads will be shown only to users on the Free and Go plans. Plus, Pro, and Enterprise subscriptions will remain ad-free." Its Help Center goes further on who is excluded, saying it does not show ads "to users on Plus, Pro, or any Business plan, or to any users who tell us or we predict are under the age of 18."

So the reachable population across those 31 countries is adults on the lower-tier consumer plans, in a mode where they are asking rather than scrolling. That is a narrower pool than the headline user numbers suggest, and it is a warmer one.

Six months from a US pilot to the largest expansion so far

The trajectory matters for how much of this is likely to change under you. OpenAI's own summary: "In February, we began testing ads in ChatGPT with a pilot in the United States. Over the past six months, we have expanded to eight additional markets. The addition of 31 European markets marks our largest expansion to date."

Nine markets in six months, then 31 in one week. A platform moving at that rate rewrites its own documentation faster than a media plan gets approved, which is an argument for checking the source pages on the day you build rather than trusting any article, this one included.

The sentence in the developer docs that sets your ceiling

One line does more work here than the entire launch announcement. It is worth reading slowly, because two of its details change what you should do.

Where it lives, and why that matters

The restriction sits in OpenAI's developer documentation for the Ads API, on the custom audiences page. It does not appear in the advertiser-facing Help Center article about the same feature, which describes uploads, matching and bid multipliers without mentioning Europe at all.

That gap is a small trap. An advertiser reading the Help Center path would find a complete-looking guide to a feature they cannot use in these markets. The precise attribution is the developer documentation, and it is the surface to check before promising a client an audience strategy.

It restricts campaigns, not companies

The wording is scoped to "campaigns targeting the European Economic Area (EEA) or Switzerland." That is a targeting condition, not a corporate address. A US or Brazilian advertiser pointing a campaign at Germany runs into exactly the same wall as a German one, and a European brand advertising into the United States does not.

Two separate restrictions are in play, and they compound rather than repeat:

• The documented one, from the developer docs: no custom audiences on a campaign that targets the EEA or Switzerland, whoever is running it.
• The structural one, from the availability page: an advertiser based in the EEA cannot open a self-serve account at all right now.

Keeping them apart is what stops a plan from being built on the wrong one. A London or New York team can buy into these markets today through a partner; what none of them can buy is an audience.

"Not yet" is doing real work

The sentence says personalized ads are "not yet available," not that they are unavailable. That is OpenAI's own choice of words on its own documentation, and it reads like a staging note rather than a settled position.

Plan for the configuration that exists. A campaign designed today should assume no audience layer for its full flight, because that is the documented state, and a campaign designed around personalization arriving in October is a campaign betting on a roadmap. If it lands, it is upside on a plan that already worked.

Nine countries can open an account, and none is in the EEA

The asymmetry between where the ads run and where the accounts live is the most quotable thing about this launch.

What the availability page listed on August 18

OpenAI publishes an Ads Manager availability table, which says the table "shows where Ads Manager is currently available and where access is coming soon." As captured on August 18, 2026, the available column held nine countries: Australia, Brazil, Canada, Japan, Korea, Mexico, New Zealand, the United Kingdom and the United States.

Nine. Not one of them is in the European Economic Area or Switzerland. So in the week ads started appearing across those 31 markets, the number of them where an advertiser could open an account through the front door was zero. People in the new countries got the ads before advertisers in the same countries got an account.

The doors that are open instead

OpenAI names the alternative route directly: "Advertisers will initially be able to access ChatGPT Ads through the OpenAI Ads Solutions team, agency partners, and technology partners. Self-service access through Ads Manager will follow later this summer."

For a large brand with a holding-company relationship, that is a phone call. For a mid-market ecommerce team, it is a form and a wait. The practical consequence is that the first cohort of European advertisers on this channel will be skewed toward managed spend, and the learning that comes out of it will be reported by agencies rather than published in a dashboard anyone can open.

What "later this summer" actually commits to

It is a short window, and worth reading as one. Summer ends around September 22, so OpenAI's own phrasing puts self-serve inside roughly a month of the European launch. The same availability page hedges with "Country availability may continue to evolve as testing expands," which is the standard escape hatch on a page that has been rewritten several times this year.

Two things follow:

• Do not build a Q4 plan that depends on Ads Manager being open in your market.
• Do not assume it will stay closed either, and keep a briefed campaign ready for the week it opens.

Build the creative now, because the creative is the part that works either way.

What a European campaign can actually buy

Strip out the audience layer and three families of control remain. They are enough to run on, and they are not the ones most media buyers reach for first.

What an EEA-targeted campaign can use today:

• Country, region and DMA geo-targeting
• Placement against the conversation on screen
• Context hints at the ad group level
• Impressions, clicks or conversions as the goal
• The OpenAI Pixel, the Conversions API and third-party measurement

What it cannot use:

• Custom audiences built from your customer or prospect lists
• Audience exclusions built from those same lists
• Custom-audience bid multipliers
• Any other form of personalized targeting, for now

Geography, at three levels of resolution

The targeting documentation is unambiguous and carries no European exception: "Use campaign targeting to control where your ads can deliver. OpenAI Ads supports country, region, and DMA targeting." Locations are looked up through a live geo API rather than a fixed published list, so the targetable set moves with the platform instead of sitting in a static table.

So a Spanish campaign can be a Spanish campaign, and a campaign that wants Bavaria and not the rest of Germany can be that too. Geography survives the carve-out intact, which makes it the only hard boundary a European campaign owns.

Context, which means the conversation and not the person

The second control is placement against what is on screen. Ads render below the response, and ad selection runs through what OpenAI calls a "relevance-weighted, second-price auction." The signals it considers include the context and intent of the current conversation, the landing page, the title, the copy, and advertiser-provided context hints.

Read that list again with the audience layer removed. Four of the five inputs are things you wrote. In a market with personalization switched off, your copy is not just the message, it is most of how you qualify for the auction at all.

Measurement, which came along for the trip

The measurement stack is not carved out either. OpenAI describes it as expanding "measurement beyond clicks through the OpenAI Pixel, Conversions API, and third-party measurement integrations." That gives a European campaign three routes home:

• The OpenAI Pixel, browser-side, the fastest to install and the easiest to lose to consent tooling.
• The Conversions API, server-side, the sturdier of the two and the one worth the engineering hour.
• Third-party integrations, for teams already measuring several channels in one place.

Conversion bidding works the same way it does elsewhere, and the thing worth knowing before you set a target is that the CPA bid you type is not the unit you are billed in.

One default deserves a calendar entry rather than a paragraph: the pixel's automatic advanced matching switch changes what your site sends back to OpenAI, and a European rollout is a good moment to confirm which consent gate sits in front of it.

When you cannot slice the audience, the creative does the slicing

Here is the part that turns a policy note into a media plan. Every targeting control a platform removes gets reassigned to the creative, and this one removed the biggest one.

That reassignment is not unique to OpenAI. In the same week, Microsoft Advertising set an October 1 date for removing the Max CPC control from new campaigns on its standalone automated bidding strategies while switching on wider search-term matching, which leaves budget, targets and creative as the inputs a buyer still sets. Two platforms, two different levers taken away, and the same residue left behind.

The unit is small, and the picture is most of it

A ChatGPT ad is a headline, a sentence and an image. The API caps ad titles at 3 to 50 characters and bodies at up to 100, which is roughly a title and one line of copy. There is no room to explain, qualify or segment inside the unit.

That puts an unusual amount of weight on the image, and it means the number of distinct images you can put into an ad group is close to the number of distinct arguments you can make. In every other channel, creative volume is an optimization. Here it is the targeting mechanism.

Context hints are a description, not a keyword list

The instrument most buyers reach for next is context hints, and it repays a careful read. OpenAI's own wording is that these hints help guide matching, "but they are not exact-match keywords and do not guarantee delivery in specific conversations."

So the hint tells the system what conversation you belong in. It does not buy that conversation. Which means you cannot fix a mismatch by editing a keyword the way you would on search. You fix it by putting a different creative in front of a differently framed conversation and reading what survives.

A worked example, in one fortnight

Take a skincare brand entering Germany and Spain on day one, with the audience layer unavailable in both.

• Two campaigns, not one. Country-level geo split, so spend and results stay readable per market.
• One landing page per market. Hold the destination constant, so the variable under test is the picture.
• Eight image treatments per market. Same product, eight arguments: texture close-up, in-hand scale, morning routine, before-and-after framing, ingredient hero, packaging on a bathroom shelf, a hand applying it, and the product against a plain surface.
• Two headlines per image. Fifty characters is not room for a message test, so vary the promise, not the phrasing.
• Read at the ad row. Per-creative rows are the clean signal the platform returns, so let them decide what survives the first two weeks.

That is 32 ad variants across two markets before anyone has agreed on a hero shot. Producing them the traditional way is a shoot, a retouch queue and a fortnight of lead time, which is exactly where a plan like this normally dies. The discipline behind it is the same one that governs structured creative testing on mature channels, and the arithmetic behind the number of variants is covered in how many ad creatives a test actually needs.

One more note for anyone porting a plan across from the US or the UK: the seven-market setup guide we published earlier this month is still accurate on mechanics, and its country list is now the smaller half of the map.

How Novoads solves the creative-supply problem contextual targeting creates

Contextual-only targeting does not need better ads. It needs more of them, consistent with each other, ready before the flight starts.

That is a production problem, and it is the one Novoads is built for. You upload a product photo and generate ad-ready stills from it through GPT Image 2 at a flat 0.3 credits per image, so eight treatments of one product is 2.4 credits and a second set in a different setting is another 2.4. The same product photo also feeds the UGC video ads you run on the channels where video is the unit, which keeps one asset library serving several placements instead of one shoot serving one campaign. What that changes is the decision: you stop choosing which three ideas are worth photographing and start choosing which of eight the auction liked.

Start for $1. That is $1 for 3 days of access. Cancel anytime.

In Europe, creative is the targeting layer

Every young ad platform is thin somewhere, and the thin part tells you where the work is. ChatGPT Ads arrived in Europe with the audience layer switched off, the self-serve door still closed and the copy capped at a sentence, and it handed advertisers geography, a conversation and a picture.

That is not a weaker channel. It is a channel where the usual order is inverted. On Meta or Google you find the right person and let a competent creative do the rest. Here the creative finds the moment, because nothing else can. The advertisers who treat variant supply as the media plan, rather than waiting for the audience controls to arrive, will have spent a full quarter learning what the conversation responds to before everyone else gets their account.