Companies can self-book ads in ChatGPT in 31 European markets, including Germany and Austria.1 Less than 200 days after the initial pilot, OpenAI reported an annualized revenue run rate of $1 billion and tens of thousands of advertisers for **ChatGPT Ads**.2 For marketing managers and agencies, this is a nascent channel where the competition for placements is just beginning. This article demonstrates what the channel offers, for whom it is suitable, and how you can test it in a controlled manner.
What You Can Book in ChatGPT
OpenAI first tested advertising in the USA and then expanded it. Europe was added as the biggest step: 31 markets at once, with access available through the Ads Manager as a self-service option, via their own sales team, or through agency partners.1
The range of functions now corresponds to what you are familiar with from other channels. In addition to CPM and CPC bidding, there is optimization for conversions, along with geo-targeting and custom audiences.1 Furthermore, product feeds, a proprietary pixel, a Conversions API, and more than 50 technology and measurement partners are available. The Ads Manager is also open to small and medium-sized enterprises, which, according to OpenAI, account for a significant share of the business.2
You should be aware of three rules from the advertising principles, as they determine the mechanics. According to OpenAI, ads are clearly labeled and separated from the response, advertising does not influence the response itself, and advertisers do not gain access to user conversations.1 The ad system selects the ad based on the current conversation context, and depending on the country and settings, also based on further usage history.2
Why the Channel Is Worth Testing Now
People come to ChatGPT with a goal and describe it in full sentences: what they want to achieve, what is important to them, and what limitations apply. OpenAI explicitly cites the selection of software for a company as an example.2 This intent is more precise than three keywords and is temporally close to the decision.
Researchers at Princeton University measured how strongly a conversation influences a purchasing decision in two pre-registered experiments with 2,012 participants. The task was to select a title from an e-book catalog, with one-fifth of the titles marked as sponsored in the background. When the language model was instructed to promote these titles, 61.2 percent chose a sponsored book. In traditional search with highly ranked sponsored results, it was 22.4 percent. A visible "Sponsored" label did not significantly reduce the effect.5
This arrangement is not the same as ChatGPT Ads: there, according to OpenAI, advertising does not directly control the response.1 Nevertheless, the figures demonstrate how closely conversation and decision-making are linked. Appearing there with a labeled ad places you closer to the moment of decision than in any banner environment.
Additionally, consider the timing. Tens of thousands of advertisers and more than one billion weekly active users2 are operating in a market that has only existed for just over half a year. No one from outside knows how many competitors in your niche are bidding on the same placement because OpenAI does not publish competitive density. Our assessment: For many B2B niches, competition is currently lower than in established channels, and this changes with each quarter that self-service remains open. OpenAI's own examples, an online retailer with a threefold return on ad spend over 28 days and a partner with over 80 percent new customer share in ad traffic,2 are vendor statements and serve as indicators, not as a basis for planning. Your own test figures are the basis for planning.
The One Question That Decides the Test: Can You Reach Your Customers Here?
Ads are only visible to users of the free Free and Go plans. Plus, Pro, and Enterprise plans remain ad-free.1 This is the condition on which everything else depends, and it can be answered in one minute.
If you aim to reach managing directors or IT managers who use ChatGPT via a corporate subscription, you will not reach them here. Thus, for classic Enterprise B2B with long procurement cycles, this channel is currently unsuitable.
It is suitable if your customers come to ChatGPT privately or as a small business: freelancers, tradespeople, law firms and agencies with fewer than ten people, local service providers, consumer business. These groups primarily use the free plans, and they describe their situation there in full sentences before making a purchase.
Examine your last twenty inquiries: Did they come from individuals and small businesses or from purchasing departments? The answer determines whether the following sections are relevant to you.
The Other Half: Visibility Without a Budget
Advertising is only one part of how your brand appears in AI responses. The other part involves the sources that the models draw upon, and that doesn't cost any media budget.
Trellner Research surveyed the Perplexity models Sonar and Sonar Pro for the best products across 380 software categories and evaluated all 7,534 cited sources. 59.8 percent of the citations lead to domains outside the top 100,000 most visited websites, and 23.4 percent to domains outside the top million. Three sites, apparently under joint control, collectively published 215,128 machine-generated "best of" buying guides and appear among the ten most cited sources.6 The study covers only Perplexity, not ChatGPT. However, it illustrates how tenuous the source base behind an AI recommendation can be.
The straightforward path to achieving this is your own website: clear product pages with verifiable facts, a comparison page with honest criteria, current prices, and an imprint that builds trust. Exactly the kind of pages that a human buyer would also read. We have described what this means for the content work of agencies in the article on AI consulting for agencies and for marketing professionals.
Four Commitments That Make the Test Controlled
A test is controlled if four things are formally agreed upon in writing before the first euro is spent. All of this can fit on a single page.
1. A Measurable Goal. One offer, one market, one action: demo request, download, purchase. Along with the benchmark value against which the test must be measured, typically the cost per lead from your best channel over the last three months. Without this benchmark, you will end up with a number without a standard.
2. Measurement That Works Beforehand. Set up a pixel or Conversions API and verify with a test conversion that the count is actually received. Complete this step before starting, not afterwards. A campaign without reliable tracking provides impressions, not decisions.
3. A Cap and a Stop Criterion. A fixed amount for the entire test, which you provision as learning costs in advance. Additionally, the condition under which you will stop, for example: if the cost per lead after two weeks is three times the benchmark value, the test concludes. Both should be signed off jointly by Marketing and Management; otherwise, the test will turn into a continuous budget.
4. Three Figures and the Decision That Follows. Track weekly: Cost per conversion against the benchmark, proportion of new contacts among conversions, and lead quality based on feedback from sales. The third figure carries the most weight, as a cheap lead that sales does not want to call is not a result. In the end, one of three decisions is made: scale, if costs are right and sales confirms quality. Adjust and run a second round with the same cap if an ad variant performs significantly better. Stop if the stop criterion is met, and revisit the channel in six months. How such key figures can be continuously managed is detailed in Making AI Investments Measurable.
Two other things should be done in parallel. Once a month, ask ten typical customer questions to two or three AI assistants and note whether and with which sources your brand appears. This takes 30 minutes and serves as your baseline for the organic aspect.
And phrase your ads in a way that you can publicly stand behind them. The EU Commission has designated ChatGPT as a very large online search engine under the Digital Services Act.3 Therefore, Article 39 DSA applies here as well: a publicly searchable advertising repository containing content, advertiser, and ad period, which retains every ad for one year after its last display.4 The implementation deadline is four months after the designation, i.e., end of December 2026. This changes nothing for the test decision, but it does for the copywriting: what you book can be read by competitors and customers.
Agencies serving clients can offer this process as a package. The channel is new enough that the experience gained from the initial test itself is a selling point. We have outlined why the middle of the agency market is currently under pressure and where new services are emerging in Agencies in the AI Transformation.
Is the Channel Worthwhile for You?
In 30 minutes, we will review with you whether your customers are even reachable via the free plans, what benchmark you should set, and with which sources your brand currently appears in AI responses. No sales pitch, no slides.
Sources & References
- OpenAI, „ChatGPT Ads expands across Europe", 18.08.2026, Update 31.08.2026: openai.com
- OpenAI, „A milestone in expanding access to AI", 31.08.2026: openai.com
- Europäische Kommission, Vertretung in Deutschland, Pressemitteilung „Gesetz über digitale Dienste: Einstufung von ChatGPT, Reddit und Roblox", 31.08.2026: germany.representation.ec.europa.eu
- Verordnung (EU) 2022/2065 (Digital Services Act), Artikel 39 „Zusätzliche Transparenz der Online-Werbung": eur-lex.europa.eu
- Francesco Salvi, Alejandro Cuevas, Manoel Horta Ribeiro (Princeton University), „Commercial Persuasion in AI-Mediated Conversations", arXiv-Preprint 2604.04263, April 2026 (zwei vorab registrierte Experimente, N = 2.012, fünf Frontier-Modelle): arxiv.org · Einordnung und Autorenzitate: The Register, 09.04.2026: theregister.com
- Trellner Research, Report TR-2026-009 „Three sites made 215,128 'best software' pages for AI. Perplexity cites them" (760 Abfragen, 380 Kategorien, Stichtag 02.09.2026): trellner.com