ChatGPT Ads Went Live in 31 Countries Today With Almost No Targeting

ChatGPT started serving ads in 31 more countries today. The advertisers buying them mostly cannot choose who sees them.

OpenAI announced the expansion on August 19, taking its ad business from 9 markets to roughly 40 in one move: Germany, France, Spain, Italy, Poland, the Nordics, Ireland, the Netherlands and the rest of the European Economic Area, plus Switzerland. It’s the largest geographic jump the platform has made.

The part worth reading twice is what the ads can and cannot do there. At launch, ads in Europe are not personalized. Selection draws on five things: the topic of the conversation happening right now, approximate location, device type, time of day, and language.

That’s it. No audience lists. No history. No interests.

Dave Dugan, OpenAI’s VP of ads, framed the pitch this way:

“People come to ChatGPT with goals. In the attention economy, brands have to be interesting; in the intelligence economy, they have to be useful.”

How full the surface already is

Europe joins a surface that filled up fast.

Promptwatch, which samples the live interfaces rather than asking anyone, tracked ad frequency across 90 days to August 17. ChatGPT served no ads at all until May 27, when they showed up in 1.69% of responses. The 90-day average landed at 20.1%. In the most recent week it was 32.4%, up 12.5% on the week before.

Read the denominator before you use that number, because it’s narrower than it looks. Promptwatch counts “responses where ChatGPT searched the web, returned at least one citation, and completed ad extraction.” So it covers roughly a third of the answers where ChatGPT went and looked something up, well short of a third of everything anyone asks it.

Even narrowed, the shape is the story. Zero to about a third of searching answers in twelve weeks, and the newest markets got switched on at the top of that curve rather than the bottom.

Promptwatch sells AI-search monitoring, so the incentive runs toward showing a busy surface. It also publishes the raw series and flags its own volume caveats, which is more than most of the numbers in this category come with.

The trade nobody is naming out loud

Strip out personalization and you have not made advertising worse. You have made it a different product, and one most performance teams stopped buying a decade ago.

What remains is close to early paid search: someone states an intent, and you buy against the intent rather than against the person. The targeting surface is the sentence they just typed. For brands with broad appeal and a clear category, that’s workable. For anyone whose media plan depends on retargeting, lookalikes, or a customer file, there’s nothing here to plug into yet.

Victor Batista, senior director of paid media at Jellyfish, has clients in the pilot and is not overselling it:

“This is very new, and they are building at a very fast pace.”

Access is narrow to match. Buying runs through OpenAI’s own ads team, agency partners, and technology partners. Self-service through the Ads Manager comes later. That’s a tighter door than the US market, where the Ads Manager went live and the barrier dropped earlier this month. Shamsul Chowdhury, SVP of paid media at Zeno Group, reads the geography as a revenue decision:

“Europe makes natural sense for OpenAI to tap into in order to expand their ad revenue footprint.”

The commercial pressure behind it is real. OpenAI told reporters its ad revenue has grown more than 25% since the start of August, against about a billion weekly users, roughly a fifth of whom show commercial intent.

Why the targeting is missing

European law is the reason.

GDPR requires explicit consent before a company can process personal data to target advertising, unless it can argue a legitimate interest, which is a hard case to make for ad targeting. So OpenAI built the consent step into the product. Users can opt in to personalized ads, which lets OpenAI use conversation history, purchase data and interests. Users who don’t opt in still see ads, drawn only from the current conversation and rough location.

Opting out changes which ads you see. It doesn’t change whether you see them.

Clara Westbrook, head of data privacy at Arbor Law, thinks the structure puts a particular kind of pressure on users, since ads only appear on the free and cheaper tiers:

“It almost feels like people are having to pay if they don’t want to receive ads at all.”

For a buyer, that splits your European audience in two, and the ratio between the halves is a number OpenAI has and you don’t.

You can buy it before you can measure it

Here is the timing problem, and it broke on the same day the ads did.

The IAB is building the framework for measuring AI advertising, and it isn’t due until November 12. The signals attribution runs on, tracking tags and referral data, don’t reliably survive a journey that passes through an AI platform. Caroline Gigerich, the IAB’s VP of AI, wants it to “create a shared framework for measuring and crediting AI’s role in conversions.” Asked what was hardest for the working group to agree on, she said: everything.

This is the trade body’s third AI framework this month. It already shipped guidance on measuring AI visibility, which we covered when it told you which AI visibility numbers to ignore, plus one on disclosing AI use. Attribution came last because it’s the hard one.

So the sequence for a European advertiser is: inventory today, standard in eleven weeks. Spend lands in a quarter that closes before anyone agrees how to credit it.

The steelman

The strongest argument against treating this as a problem: contextual advertising worked fine for years, and intent stated in a sentence is richer than intent inferred from a cookie. Someone typing “best CRM for a 12-person sales team” has told you more than any behavioral segment would.

That argument is good, and it’s why this inventory is worth testing rather than skipping. But it cuts against OpenAI too. If context is enough, the consent flow is doing something other than improving relevance. And the measurement gap is real either way, because a strong signal you can’t attribute still lands in your reporting as nothing.

What to do with this

Three things, and none of them is “wait.”

Test it as a brand line, not a performance line. Without personalization or attribution, last-click will produce a number that means nothing. Give it a budget you can afford to measure badly for a quarter.

Write down your own measurement rules now, before November. Whatever ships in November will be a standard, not a retrofit. With no baseline from the pre-standard period, you’ll have nothing to compare it against.

Ask your agency which of the two European audiences your ads reached. Consented and non-consented audiences behave differently, and that split is the most useful thing to learn from a first quarter of spend. If nobody can answer it, that’s the finding.

Ads shipped today. The ruler ships in November. Plan for the gap, because the spend is happening inside it.