I. AI-Generated Search Results Under Increasing Legal Scrutiny
Generative AI is fundamentally changing how users access information online. Rather than displaying a list of links, modern search engines increasingly provide AI-generated summaries that synthesize information from multiple sources and present it as a direct answer to a user’s query.
This development raises an important legal question: to what extent should providers of AI-powered search services be responsible for the content of those generated responses?
Two recent German decisions illustrate the emerging debate. In its judgment of 28 May 2026 (docket no. 26 O 869/26), the Regional Court of Munich I (LG München I) held that a search engine operator could be responsible for allegedly false statements contained in an AI-generated search overview. Only days later, the Regional Court of Berlin II (LG Berlin II), in its decision of 1 June 2026 (docket no. 52 O 62/26 eV), rejected trademark and unfair competition claims relating to AI-generated search results.
Although both cases concerned AI-generated search functionality, they reflect markedly different views regarding the role of the search engine operator and the legal characterization of AI-generated outputs.
II. Munich: AI Summaries as Provider Content?
The Munich decision is likely to attract considerable attention because the court treated the disputed AI-generated overview as content attributable to the search engine provider itself.
According to the court, the AI-generated overview did not merely reproduce information available elsewhere on the internet. Rather, it selected, combined and structured information from various sources and presented it to users as a coherent response. From the perspective of an average user, the court considered the resulting text to be a self-contained statement generated by the search engine.
The court therefore distinguished AI-generated overviews from traditional search results and questioned whether the rationale underlying established search-engine liability case law can simply be transferred to generative AI systems.
At the same time, the judgment raises important questions. Search engines have long relied on automated processes to rank, organize and present information. Modern search functionality increasingly includes summarisation, contextualisation and other forms of algorithmic processing. Against this background, it remains open whether the mere fact that information is synthesized and presented in natural language should automatically result in attribution of the output to the provider.
The court’s approach to post-notice review obligations is equally significant. While it did not impose a general monitoring obligation, it suggested that, once notified of a potentially unlawful statement, a provider may be expected to verify whether the AI-generated output is supported by the underlying source material. Notably, the court was not prepared to limit such review obligations to cases involving obvious legal violations.
For search engine operators, this aspect of the judgment may ultimately be more consequential than the attribution analysis itself. If followed by other courts, the decision could lead to heightened expectations regarding complaint-handling procedures and the verification of challenged AI-generated outputs.
III. Berlin: Preserving the Intermediary Character of Search
The Berlin decision adopts a noticeably different perspective.
In proceedings concerning trademark and unfair competition claims, the court rejected the argument that AI-generated search summaries should be regarded as the search engine operator’s own commercial communication. Instead, it emphasized that users understand such outputs as summaries generated from information available on third-party websites.
The court further stressed that the search engine operator merely provides the technical framework through which information is processed and presented. As a result, it declined to attribute the generated statements to the provider in the same way as traditional advertising or editorial content.
From the perspective of search engine operators, the Berlin approach arguably reflects the practical realities of AI-assisted search more closely. AI-generated summaries are created dynamically in response to user prompts and are based on information obtained from external sources. Treating every generated response as a statement adopted by the provider could significantly expand liability exposure and create incentives for overly restrictive content moderation.
Importantly, the Berlin court’s reasoning also appears more closely aligned with the traditional distinction between content providers and intermediaries that has long shaped European internet law.
IV. The Code of Practice on Transparency of AI-Generated Content: Support for a Risk-Based Approach?
The recently published Code of Practice on Transparency of AI-Generated Content[1] may provide additional support for this intermediary-oriented perspective.
While the Code is not itself binding legislation, it offers valuable insight into the regulatory assumptions underlying the EU AI Act. Notably, the Code does not proceed on the basis that providers of general-purpose AI systems are the authors of every output generated by their models. Instead, it focuses on transparency, documentation, risk management and mitigation measures.
This regulatory approach appears broadly consistent with the reasoning adopted by the Berlin court. Both recognize that AI-generated outputs arise through complex interactions between models, prompts, users and external information sources. Neither assumes that every generated statement should automatically be attributed to the provider.
At the same time, the Code does not support complete immunity. Rather, it emphasizes that providers must implement appropriate governance structures and mechanisms to address risks associated with AI-generated content. In this respect, the Code also resonates with the Munich court’s concern that providers should respond appropriately once potentially harmful outputs are brought to their attention.
The emerging regulatory framework therefore appears to favour a model of managed responsibility rather than publisher-style liability.
V. Practical Implications for Providers and Deployers of Generative AI Systems
Although the legal framework remains in flux, the recent decisions provide several practical lessons for providers and deployers of generative AI systems, including AI-powered search tools, chatbots, virtual assistants and other applications that generate factual summaries or responses based on external information.
Particular attention should be paid to:
-
- mechanisms for identifying and, where appropriate, tracing AI-generated outputs to the information or sources on which they are based;
- procedures for investigating substantiated complaints regarding potentially unlawful or inaccurate outputs;
- safeguards against unsupported factual assertions, hallucinations and inaccurate source attribution;
- processes for correcting, suppressing or otherwise mitigating inaccurate outputs; and
- governance frameworks addressing high-risk use cases involving reputation, personality rights, consumer-facing information and other legally sensitive content.
The Munich decision suggests that courts may increasingly expect providers and deployers to assess whether challenged outputs are supported by the information on which the system relied. Organizations should therefore consider whether their existing review, escalation and remediation procedures are capable of carrying out such assessments efficiently and consistently once concerns regarding the accuracy or legality of an output are raised.
VI. Outlook
The Munich and Berlin decisions demonstrate that German courts are still defining the appropriate liability framework for generative AI systems.
While the Munich court placed greater emphasis on the provider’s role in generating and presenting synthesized information, the Berlin court focused on the intermediary character of AI-assisted information services and the user’s understanding that the output is derived from external sources. The recently published GPAI Code of Practice appears to lend support to the latter approach by emphasizing transparency, governance and risk management rather than automatic attribution of AI-generated outputs to providers.
Whether future courts will follow Munich’s attribution-focused reasoning or Berlin’s intermediary-oriented approach remains to be seen. For the time being, however, providers and deployers of generative AI systems should assume that courts may scrutinize both the design of AI functionalities and the effectiveness of the procedures used to address challenged outputs.
The key compliance challenge is therefore unlikely to be avoiding all liability for AI-generated content. Rather, it will be demonstrating that appropriate governance, review and remediation mechanisms are in place to address risks arising from AI-generated outputs, particularly once potential inaccuracies or legal concerns have been brought to the organization’s attention.