Advertising in AI chat: Why a “Sponsored” notice often isn’t enough under German law

So far, you have known advertising on the internet in fairly clear forms: a banner at the edge of the page, an ad above the search results, a paid post in the feed. AI assistants add a format that cannot be locked into a frame: the conversation. When an answer in a chat suddenly contains an offer, when a brand's "assistant" replies, or when a product is recommended, a question arises that applies to every advertising company in Germany: Is it recognizable that this is advertising?

This article deals with a timeless topic, not a piece of news. The occasion is two sources from late summer 2026. One is a legal article by Prof. Christian Solmecke (WBS.LEGAL) dated 27 August 2026 on labeling advertising in AI chatbots. The other is a report by The Next Web dated 16 September 2026 on an OpenAI test of "Sponsored Agents" in ChatGPT. Together, they show what companies should prepare for if they plan to have their advertising steered by agents in the future.

Important up front: this article is not legal advice. We report what the cited sources write and put it into context for the Mittelstand. For your specific case, you need a lawyer who specializes in competition law.

What OpenAI is testing: Sponsored Agents

On 16 September 2026, The Next Web reported on an OpenAI test. According to the report, after clicking an ad in ChatGPT, users can choose to start a conversation with an agent sponsored by a company. There they can ask follow-up questions and follow a link to the company's website. According to the report, OpenAI describes it this way: a user can start "a clearly labeled conversation with a business-sponsored agent in ChatGPT".

According to this source, the status is clearly limited. OpenAI is testing the format with selected advertisers and is not rolling it out widely. The test is restricted to the USA. The report makes no statement about availability in the EU or in Germany. In the background, it merely mentions that ChatGPT ads were reportedly to be expanded to 31 European countries in August, but it does not name Germany explicitly. It also says nothing about prices, the exact design of the labeling or the number of test participants.

For your planning, this means: Sponsored Agents are not something DACH companies can book and use today. They are a signal of the direction AI advertising is moving in. We therefore treat them as "something that is coming toward companies" and not as an available tool.

What the separation requirement (Trennungsgebot) demands

Regardless of whether and when Sponsored Agents come to Germany, the legal question already applies today to any advertising that appears in an AI chat interface. Prof. Solmecke classifies it as follows in his article of 27 August 2026.

According to the article, the central provision is § 5a (6) of the German Unfair Competition Act (UWG). It penalizes concealing the commercial purpose of a commercial practice. Users must therefore be able to tell at first glance whether they are looking at neutral information or a paid ad. The article describes the separation requirement (Trennungsgebot) as the cornerstone: it demands a strict visual and substantive separation of editorial or organic content from promotional statements.

According to WBS.LEGAL (27 August 2026), users must be able to tell at first glance whether they are looking at neutral information or a paid ad.

In a conversation, this is harder than with a banner. A banner has a frame; a chat history does not. When an assistant's answers and paid notices appear in the same flow, the boundary can blur. That is exactly why the format is legally delicate.

"Sponsored" is not enough: which labeling is required

The most practically important point of the article concerns the choice of words. According to Solmecke, German courts often classify terms such as "Sponsored" or "Promoted" as insufficient, because they are not immediately understandable as a commercial notice to all parts of the population. What is required, he says, are unambiguous German terms such as "Anzeige" or "Werbung".

The article is also clear about placement:

• A link or product image at the edge of the chat is not sufficient.
• The labeling should not be hidden in the small print or at the end of a long block of text.
• It must be located spatially directly next to the advertised product or service.
• Recommended, for example, are a set-off frame or a colored background.

This matters because many international platforms work with English terms. A company that places advertising on such a platform cannot simply rely on the platform's standard labeling being sufficient under German law. The article expressly states that advertisers cannot invoke the platform's responsibility across the board and have their own duties of review.

What is at stake in the event of a violation

On the consequences, the article says that advertising companies could be sued by competitors for injunctive relief and damages. It also warns of waves of cease-and-desist letters from consumer protection associations and competitors. According to the article, the Digital Services Act also obliges platform operators to disclose for every ad that it is an advertisement and to make transparent on whose behalf it is placed. Users should also learn on the basis of which main parameters the ad was selected for them. As a fine framework, the article names up to six percent of global annual turnover for platforms.

What the article does not contain: specific court decisions with case numbers, fine amounts under the UWG, and information on which labeling OpenAI actually uses. We therefore name no decisions and no amounts for companies. Austria and Switzerland have their own rules, which we have not checked here. The statements in this article refer to the German legal position as the cited article presents it.

What this means for the Mittelstand

An agent that plans, designs or places advertising usually optimizes for a goal: clicks, inquiries, sales. Whether the ad is sufficiently labeled under German law is not an optimization goal. It has to be a rule in the process, otherwise it never comes up. Three scenarios show where this becomes practical. They are examples for illustration, not cases from the sources.

Example 1: An advertising agent books a new channel

An agent is given the goal of increasing inquiries for your service and is allowed to test channels. As soon as a new AI chat channel offers advertising, it could integrate it. Whether the ad appears there in German and clearly recognizable as an "Anzeige" depends on the platform's format and on your settings. Without a review rule, you only notice when the cease-and-desist letter arrives.

Example 2: A brand agent in the chat

You have an agent answer customer questions. In its answers, it occasionally recommends your own products. This raises the question of when advice becomes advertising and how that is recognizable to the customer. The sources do not address this case in detail. But it shows that separating information from advertising does not only concern ad placements, but every conversation format in which a company speaks.

Example 3: A service provider places ads for you

An agency or a tool takes over your campaigns. The WBS.LEGAL article states that, as an advertiser, you cannot rely on the platform alone. By analogy, this also applies to service providers: you should clarify who checks the labeling, and put that in writing.

Which team belongs here: Daniel and Nadja

On amazing-agents.com, we assign this topic to the Research & Analytics area. Nadja is the Research Lead there and delivers topic research, competitor scans and research briefings. Daniel is the Analytics Agent, who evaluates performance data and builds reports. Lotte, as Trend Scout, watches new formats and channels.

For the question of a new advertising channel, a natural division of labor emerges. Research clarifies what the channel offers and which rules apply to it. The analysis later shows what the ads actually achieve. That this team looks at the labeling obligation before every new advertising channel is our proposal for the process, which we adapt with you in the audit. The legal assessment itself remains with your legal counsel.

From the social media team, we also describe on the website that Magdalena checks every content draft for brand fit, tone and formal correctness, and then Natascha approves it. Emails are created as drafts and not sent directly. We transfer this idea, that a person or a review instance looks before publication, to ads.

With and without an approval step: the comparison

The difference we emphasize with every topic is the one between an agent without a frame and an agent with limits and an approval step. For advertising in AI chat, it looks like this.

Without an approval step

• The agent books and words the ad solely according to performance goals.
• Nobody checks whether "Anzeige" or "Werbung" appears directly at the offer.
• The platform default is adopted, even if it is in English.
• The error only shows up through a cease-and-desist letter or a complaint.

With an approval step

• The agent prepares the ad as a draft and marks the labeling text.
• A person or a review instance confirms before going live that the advertising is recognizable as such.
• The decision is logged, so that it can later be traced who checked what.
• In case of doubt, the matter goes to your legal counsel before the ad goes live.

Honesty is part of this too: the approval step slows down placement. But it is considerably cheaper than a cease-and-desist declaration and its consequences. How strict the review should be depends on reach and format.

What you can prepare now

Because Sponsored Agents are so far only a US test, there is nothing to switch on. But you can prepare how your company deals with advertising in AI interfaces before it reaches you.

1. Write a labeling rule: Record that every ad carries a German label such as "Anzeige" or "Werbung" directly at the offer and is visually set off, as the cited article describes.
2. Keep a channel list: Note on which platforms you advertise or have advertising placed today, and who is responsible for the labeling there.
3. Ask your service providers: Clarify in writing with your agency or tool provider who checks the labeling before an ad goes live.
4. Define the approval before placement: Determine who sees new ad formats before they go live, and when your legal counsel is brought in.
5. Observe rather than book: Follow whether and when formats such as Sponsored Agents become available in your market. Until then: observe, do not book.

Frequently asked questions

Does this only apply to ChatGPT?

No. The cited article deals with ChatGPT, but the obligation to make advertising recognizable as such does not depend on the provider. It affects any advertising in a conversational or assistant interface in which information and ads can appear side by side.

Is it enough if the platform labels the ad?

Not across the board, according to the cited article. Advertisers have their own duties of review, and the platform's labeling must meet German requirements. Clarify this for your case with your legal counsel.

Does this also apply in Austria and Switzerland?

Those countries have their own rules, which we have not checked in this article. For your country, ask a local legal counsel.

Conclusion: an agent with limits and an approval step

Advertising in AI chat is not a legal vacuum in Germany. The WBS.LEGAL article describes that the separation requirement (Trennungsgebot) applies, that an English "Sponsored" is often not sufficient under German court practice, and that advertisers have their own duties of review. Sponsored Agents themselves are, according to the reporting, a US test with no statement about Germany. Together, the two give a clear line: prepare before the format arrives.

The difference that matters in the end is the one between an agent without a frame and an agent with limits and an approval step. The first optimizes for the goal and does not notice what is allowed. The second works more predictably, because it is clarified in advance what it may do, what a person checks and who decides before an ad goes live. Where would this check begin at your company?

Sources

• WBS.LEGAL (Prof. Christian Solmecke): "Werbung im KI-Chatbot: Die strengen Vorgaben zur Werbekennzeichnung bei ChatGPT", 27 August 2026: wbs.legal
• The Next Web: "'A clearly labeled conversation': OpenAI tests Sponsored Agents in ChatGPT", 16 September 2026: thenextweb.com
• Amazing Agents: description of the agents Nadja, Daniel, Lotte, Magdalena and Natascha on amazing-agents.com