Insights

ChatGPT Ads launch in Germany: the complete guide for companies

Since 24 August, ads appear in ChatGPT, here too. This is not search advertising with a new logo. Instead of individual keywords, the conversation context is at the center, and in the EEA a narrower frame applies on launch day than the platform's feature descriptions suggest. What companies can really do now, and what not yet.

By ·9 min read·
Ein Gespraechsverlauf, darunter deutlich abgetrennt ein gekennzeichneter Anzeigenblock, rechts der Kontext statt einzelner Suchbegriffe

Anyone who knows search advertising has a practiced routine in mind: research keywords, set match types, aim ads at them, steer bids by search term. That routine does not carry over to ChatGPT Ads. Not because the tools are missing, but because there is no search term to bid on. There is a conversation.

This guide sums up what is documented on launch day, what of it applies in the EEA, and what a preparation looks like that holds even when the platform expands its options.

1. What launched on 24 August

ChatGPT Ads have been available since 24 August 2026 in 31 European markets, including Germany, Austria, France, Spain, Italy, the Netherlands and the Nordic markets.

Important for understanding the reach:

  • Ads appear only for users on the Free and Go plans. Plus, Pro, Business, Enterprise and Edu stay ad-free.

  • Accounts that OpenAI identifies as belonging to people under 18 get no ads.

  • Ads sit below the answer, are marked as sponsored, and are visually separated from the answer.

  • According to OpenAI, they do not influence the organic answer.

The announcement said access would run at first through OpenAI's Ads Solutions team, through agency partners and through technology partners, with self-service access via the Ads Manager to follow later.

Update, 1 September 2026. The correction directly below is itself out of date: self-service is open in Germany. OpenAI has listed Germany as "Available" in its availability overview since the morning of 1 September and is emailing advertisers that the Ads Manager is available to advertisers based in Europe. Checked against a real account: a campaign for Germany can be created.

We are leaving all three passages standing, the original sentence, the correction of 28 August and this update. The situation flipped twice in four days, and that is precisely the state this market is in. Working with silent edits here would erase the one reliable piece of information: that none of it holds for long.

Correction, 28 August 2026. This passage first said that self-service was already usable in Germany. That was too broad. What is accurate: creating an advertiser account works from Germany, and the German interface walks you through building a campaign. Self-service itself is still listed as "Coming Soon" for Germany in OpenAI's own availability overview, which separates the two explicitly: "Account creation and self-service availability are separate: access depends on your country." Advertisers in the new European markets who want to start today go through the OpenAI Ads Solutions team, agency partners or technology partners.

We still describe the setup in full in the next section, because several decisions are made there that cannot be corrected later, and because they come up the same way in partner onboarding.

2. Four decisions on setup that are final

Before the first ad runs, you create an advertiser account. In doing so, decisions are made that the interface explicitly marks as irreversible. This is the part you do not click away in passing.

Account country, account currency and time zone are final. On setup you choose country, currency and time zone. Below the continue button is the sentence that these details cannot be changed after proceeding. For a German company that means: deliberately set country Germany, currency EUR and the Berlin time zone. Without changing it, the platform bills in USD. The note there is unambiguous: if EUR is chosen, EUR applies to campaign budget and bid instead of USD, and the amounts are adjusted to the minimum amounts of the account currency.

The time zone is not cosmetic. It determines when a daily budget resets and to which day a report books its numbers.

The budget type is locked after the campaign starts. You choose between a daily budget and a total campaign budget. Once the campaign has started, the type can no longer be switched. The amount stays changeable at any time, the type of budgeting does not.

The target domain is locked, only the path is free. For the target URL, the generated address is fixed; you can add a path or query at the end to point to a more specific page. So https://your-shop.de becomes https://your-shop.de/collections/your-category, but no other host.

That sounds technical and is in truth a substantive fork in the road: if you need a separate landing page per buying context, those pages must exist on your own domain before the campaign starts. Whoever has only the homepage cannot serve the context they just won. Product, category and content pages are explicitly intended as landing pages; only the generic homepage is discouraged. A category page is therefore a full-fledged target, not a fallback.

The bid strategy is preset to automatic. The strategy on offer is "Maximum results", described as automatic bid adjustment to achieve as many results as possible with the given budget. So you steer via budget, goal and context, not via individual bids per search term.

3. The point most overviews skip: what really works in the EEA on launch day

The platform's feature descriptions and the reality in the European Economic Area do not match on launch day. Whoever overlooks this plans campaigns that do not run that way here.

In the EEA, ads at launch are exclusively contextual. Used are:

  • the current conversation topic

  • the approximate location

  • the device type

  • the time of day

  • the language

Explicitly not used are earlier chats and saved memories. Users in the EEA and Switzerland do not get personalized ads at first.

The reason is legal and it matters: for contextual delivery OpenAI relies on legitimate interest. Personalized advertising would need consent. On top of that comes the Digital Services Act, which prohibits targeting based on protected categories such as ethnic origin, religion, political views or sexual orientation, and forbids profile-based advertising to minors.

For your planning this means:

  • Audience lists (Custom Audiences) are a platform feature, but on launch day in the EEA they are not the means you work with here. There is also a practical hurdle: a Custom Audience needs at least 25,000 matched users, which excludes most mid-sized lists anyway.

  • Retargeting in the usual sense is not available on launch day.

  • Your most important lever is therefore not the audience but the context: how well you describe the conversational situation in which your offer fits.

This is not a restriction you optimize away. It is the frame you work within here, and it shifts the work from audience selection to understanding the buying situation.

4. Which industries ChatGPT Ads are blocked or restricted for

ChatGPT Ads are not open to every industry. OpenAI reviews every ad in a two-stage process, first by machine, and for edge cases additionally by people. Under OpenAI's ad policy, currently blocked or heavily restricted are:

  • Gambling, betting, lotteries and cash-prize gaming

  • Alcohol above 0.5 percent as well as tobacco and nicotine, including vaping

  • Adult content: dating, sexual services and products, companion apps, nudity, sexual health

  • Weapons and ammunition

  • Drugs

  • Financial and legal services as regulated fields

  • Health and health claims

  • Political advertising

  • Cryptocurrencies

The distinction matters: regulated does not automatically mean forbidden. In several of these fields ads are possible, but need approval or proof. Whoever works in one of these areas clears the admissibility before building the campaign, not after. And because ChatGPT Ads are new, this list can change.

5. Which campaign goals are documented

At the platform level, currently documented are:

Goal

Billing

For what

Reach

CPM

Visibility in relevant conversations

Clicks

CPC

Traffic to a landing page

Conversions

oCPC

Delivery is optimized toward clicks more likely to lead to a nominated conversion event; billing is still per valid click

Added to that are country and platform targeting, product feed campaigns, Custom Audiences, and the OpenAI pixel and the Conversions API. Which of these building blocks are practically usable in the EEA on launch day follows from section 3.

6. Context Hints instead of exact-match keywords

Context Hints are the real break with search advertising. They describe topics, needs, situations and use cases. They help the system understand when an ad fits. They are not exact targeting rules, but broad thematic signals.

An example for a coffee grinder:

User compares quiet espresso grinders for a small kitchen, needs fine adjustability and has a budget of at most 500 euros.

That is closer to a buying-decision question than to a keyword like "buy espresso grinder". It describes a situation with requirement, constraint and budget.

The rule here is simple and often violated: if you need Context Hints for clearly different products, audiences or use cases, build separate ad groups rather than packing everything into one. A group meant to cover beginners and professionals, kitchen and gastronomy at the same time no longer describes any situation in the end.

7. Preparation in seven steps

  1. Capture decision situations, sorted by Discover, Explore, Compare and Decide.

  2. Collect recurring requirements, constraints and objections, in the customers' words, not marketing's.

  3. Form separate ad groups per product, audience and use case.

  4. Develop several genuinely different messages per group, not three variants of the same sentence.

  5. Provide a matching landing page or answer layer for each context. Whoever sends to the homepage loses the situation they just won.

  6. Prepare conversion events, consents and server-side measurement.

  7. Record the organic AI visibility as a baseline before the campaign starts.

The seventh step is the one you cannot make up later. Once paid delivery is running, you can no longer cleanly separate what the campaign caused from what would have happened anyway. Whoever measures today can answer that question later. That is exactly what LLM Discovery Intelligence is meant for.

8. What OpenAI can measure, and what not

The pixel and the Conversions API capture actions after an ad click: purchases, leads, registrations. An event is attributed when it matches a configured campaign goal, lies within the attribution window, and can be connected to a valid click.

That is solid platform attribution. But it does not answer a set of questions that decide success:

  • Was the user already ready to buy, or did the ad create the need?

  • Which question blocked the decision?

  • Which answer on the landing page was the relevant one?

  • Would the conversion have come without the ad?

  • How do paid and organic recommendations develop side by side?

These questions lie outside what an advertising platform can know about itself. A platform measures its own click. It does not measure whether your offer fit the question.

9. The honest division of labor

We think it is wrong to promise a product now that the platform does not yet allow. In the publicly available documents there is a Conversions API, bulk uploads and a technology-partner access, but no generally available interface to fully create, steer and read back campaigns.

That is why we explicitly do not say that private ChatGPT prompts could be viewed, that ad impressions could be connected to organic answers at the user level, that campaigns could be steered via a public ads API, or that every conversion could be causally attributed to the ad. None of that is true today.

What remains is a clear division of labor:

OpenAI measures the ad click. Klariton measures whether need, answer and business outcome line up.

In practice that means: the buying situations that good Context Hints are built from already sit in your material and in your customers' questions. Making them visible is the work that comes before the first campaign and that holds even when the platform changes. Owned Intelligence and Buying Answers are built for that, and Safe Guard holds the resulting statements against your product data, so you claim nothing in an ad that your material does not support.

The biggest mistake would be to treat ChatGPT Ads as another channel into which you tip existing search ads. The channel does not ask for your keywords. It asks whether you can describe your customer's situation.

What ChatGPT selects the ad from, and why that weighs differently in the EEA

Added 1 September 2026.

OpenAI's help lists the signals that count when an ad is selected. The sentence is worth quoting, because it explains the whole preparation:

"It considers multiple signals, including the context and intent of your

current conversation, the ad's landing page, title, copy, advertiser-provided

context hints and targeting selections, and, when ads personalization is

enabled, select signals from your broader ChatGPT experience."

Five of those are in your hands: the landing page, the title, the copy, the context hints and the targeting selections. The sixth, personalisation, belongs to the user and their settings.

And here is the part that concerns Europe. From the same help article:

"Personalized ads are not initially available in the European Economic Area

(EEA) or Switzerland."

The Ads Manager states the same from the other side: custom audiences are not supported for campaigns targeting the EEA or Switzerland, because personalised ads are not available there.

**So for the European market exactly the signal drops out that advertisers usually lean on hardest.** What remains is the conversation context, which you do not steer, and four things you do: landing page, title, copy, context hints.

In the US, context hints supplement targeting. In the EEA they are the targeting. Leave them empty and you hand the matching entirely to the model and to your ad copy.

Two practical lessons from a real account:

  • The field splits on commas. Prose pasted in gets broken into many separate

hints; one paragraph became 24 fragments for us, including single words like "Peru" or "filter coffee". Standing alone those say nothing about a buying situation. One hint per line, no commas.

  • The landing page is a selection signal, not an afterthought. It sits in

the same list as title and copy. What a fetcher cannot read there cannot contribute to the selection either.

Sources

Ask Klariton

Ask your question about Klariton.

Grounded in Klariton’s own knowledge, cited rather than invented.

Or ask your own question:
Next step

How visible is your brand to AI?

The free AI visibility check shows you in under a minute how AI assistants see your shop today.