When an AI Search Engine Mentions Your Brand, Who Is Actually Speaking?

As AI-powered search becomes increasingly integrated into everyday online activity, courts are beginning to confront a question that would have seemed unimaginable only a few years ago: when an AI-generated answer refers to a trade mark, who is actually responsible for that use?

A recent decision from the Regional Court of Berlin II provides one of the first judicial attempts to answer that question. The case arose from AI-generated search summaries that appeared in response to queries about perfume “dupes” lower-cost fragrances marketed as alternatives to well-known brands. A perfume company that owns several EU trade marks argued that the operator of a search engine had infringed its rights by generating answers that referred users to competing products.

The search engine had introduced two AI-based features. One generated an overview above traditional search results, summarising information gathered from websites across the internet. The other allowed users to ask questions and receive AI-generated answers based on information the system considered relevant. When users searched for perfume dupes or alternatives to specific branded fragrances, the AI-generated responses explained the concept of dupes and pointed users towards third-party fragrance providers, often accompanied by links, snippets and preview images.

The claimant argued that this was not simply a case of a search engine indexing information. By selecting, generating and presenting content within its own platform, the company behind the search engine was allegedly using the trade marks as part of its own commercial communication. In other words, the AI-generated response was no longer just a gateway to other websites but a piece of content created and displayed by the platform itself.

The court disagreed. Drawing heavily on established case law from the Court of Justice of the European Union, particularly Google France and Louboutin, it emphasised that trade mark use requires active conduct and a sufficient degree of control over the use of the sign. A trade mark must form part of the alleged infringer’s own commercial communication. Simply providing the technical means through which others use a trade mark is not enough.

In the court’s view, that is precisely what was happening here. Although the search engine generated the text, it was merely summarising information already available on third-party websites. The references to branded perfumes originated from those external sources rather than from the search engine operator itself. As a result, the trade marks did not appear as part of the operator’s own advertising or commercial message.

A key part of the court’s reasoning centred on how users perceive AI-generated search results. It concluded that the average internet user would understand these summaries as a new way of presenting search results rather than as original editorial content created by the search engine. The presence of links, snippets and preview images reinforced the connection to external sources and made it clear that the information was derived from third-party content. The court even went so far as to say that an express disclaimer distancing the search engine from the content was unnecessary.

The claimant also argued that generative AI fundamentally changes the nature of search. Traditional search engines simply point users towards information, whereas AI-generated responses actively synthesise and present information as a direct answer. Yet the court remained unconvinced. It considered the decisive factor to be the underlying third-party information on which the response was based. The search engine may have processed and reorganised that information, but it did not exercise sufficient control over the specific use of the trade marks to become a trade mark user itself.

Nor was the court persuaded by the fact that sponsored advertisements sometimes appeared alongside the AI-generated responses. Following the logic of earlier decisions concerning keyword advertising, it held that earning revenue from advertising does not automatically mean that a search engine operator adopts every trade mark that appears within search results or AI-generated summaries as part of its own commercial communication.

The claimant’s attempt to rely on German unfair competition law was equally unsuccessful. The court noted that the parties operated in entirely different markets. One sold perfumes, while the other provided search services. Any commercial benefit gained from improving the attractiveness of a search engine did not create a sufficient competitive relationship with the perfume manufacturer.

The judgment is significant because it treats AI-generated search summaries as an evolution of traditional search results rather than as a distinct form of publisher-created content. That approach offers considerable protection to search engine operators and avoids exposing them to trade mark liability every time an AI-generated answer references a brand, a comparison or a competing product.

At the same time, the decision leaves important questions unresolved. Modern AI search systems do much more than retrieve information. They crawl websites, rank sources, determine which information is considered reliable, generate summaries and decide how those summaries are presented to users. These design choices can have a profound effect on the content that ultimately appears on screen. Whether that degree of influence remains merely technical assistance or becomes legally relevant control is likely to become one of the defining questions of AI-related trade mark litigation.

For now, the Berlin court has taken a relatively cautious approach. Trade mark owners seeking to challenge references to their brands in AI-generated search results may need to focus on the websites supplying the underlying content rather than the platforms generating the summaries. As AI search tools continue to evolve, however, courts will inevitably be forced to revisit the issue.

The question is unlikely to disappear any time soon. If an AI system presents information in its own voice, decides what to include and delivers a direct answer to a user’s question, it becomes increasingly difficult to determine where the platform ends and the content begins. When an AI-generated answer appears on screen, deciding who is actually speaking may prove to be one of the most important legal challenges of the AI era.