Judge reprimands search engines over AI summaries

The way search engines present search results is changing. Whereas search engine results used to consist of a list of links, a ready-made answer is increasingly appearing at the top of the page: the so-called AI summary. On Google, this is powered by Gemini. Other major search engines, such as Microsoft Bing (via Copilot), also offer similar AI summaries.

These AI summaries synthesise sources, draw conclusions and provide users with an immediate answer, without them having to click any further. For businesses that rely on online visibility, this is already a cause for concern. But what if that AI-generated summary is factually incorrect and wrongly links a company to fraud or scams? A recent ruling by a German court makes it clear that Google is liable for this.

What happened?

Two German publishers discovered that Google’s AI-generated summary, when users searched for their company names, made categorical claims: they were allegedly known for fraudulent and unfair commercial practices, luring customers into subscription traps, and having links to other, unrelated companies. These claims were demonstrably false. The publishers lodged a complaint with Google, but Google took no action. They then took the matter to court and sought an injunction against these statements through summary proceedings.

The Munich Regional Court largely upheld the claim. This ruling goes beyond the specific case and sets out a judgment of principle regarding the status of AI-generated search results.

No longer a mere conduit, but a publisher

The crux of the ruling centres on a fundamental distinction. For its search engine services, Google can rely on the protection afforded by the Digital Services Act (DSA) to platforms that act as passive intermediaries transmitting third-party content. According to Article 6 of the DSA, a hosting provider is only liable if it knew that the shared content was illegal, or if it failed to remove the information after being notified that the content was illegal.

The court found that Google played an active role with its AI summary and was therefore not merely a conduit. The AI summary not only contains a summary of what is on individual web pages, but Google also structures, evaluates and draws independent conclusions based on the information from the search results in the AI summary. Furthermore, in the AI summary, Google formulates (or, to put it more bluntly, fabricates) answers that simply cannot be found in the underlying sources. In this case, Google claimed in the AI summary that the publishers changed their name and URL to make identification more difficult, a claim that did not appear in any of the cited sources. Google is thus creating its own statements, for which it bears full responsibility.

Also noteworthy is the ruling on verifiability. The judge points out that Google could technically have verified the AI output against its own source pages, which serve as the basis for the summary, without needing to contact external parties. This renders the argument that proactive verification is impossible untenable. Precisely because the output is verifiable via internal sources, Google cannot hide behind practical objections.

Can an AI have an opinion?

The court classifies some of the AI’s statements as expressions of opinion, which under German law enjoy a certain degree of constitutional protection. The judge does, however, note that the opinion expressed was generated by AI and is therefore not an expression of a person’s conviction. According to the judge, it is therefore primarily an expression of commercial activity and ‘at most secondarily’ an opinion. That ‘opinion’ thus merits considerably less protection than a genuine human expression of opinion, particularly when it is based on incorrect facts.

These considerations by the German court raise the question: can AI hold an opinion? The idea that AI can hold an opinion presupposes a subjectivity that, by definition, is absent in an algorithm. Nevertheless, this classification carries legal weight, as it determines which legal framework applies. The outcome in this case is clear: an AI-generated ‘opinion’ that is factually incorrect offers Google no protection.

How might this case play out in the Netherlands?

The ruling has direct practical relevance for any business operating online. AI-generated summaries from Google and other search engines also appear in the Netherlands and are based on similar algorithms. The German court’s reasoning is broadly applicable and aligns seamlessly with the DSA, which, as a European regulation, also applies in the Netherlands. It is realistic to expect that Dutch courts would reach the same conclusion in similar cases.

In practical terms, this means that Google or another provider of an AI system can be held liable if an AI summary makes factually incorrect statements about a person or business, or statements that qualify as misleading commercial practices. The fact that the output was generated by an algorithm is therefore not a valid defence.

What can you do if this happens to you?

The first step is to document the disputed AI output, including the date of the search, the search terms entered and a printout or screenshot of the full summary. Next, you should formally notify Google or another provider of the unlawfulness of the content in the AI overview, preferably via a registered letter of formal notice clearly setting out which statements are incorrect and why. If Google fails to respond or responds inadequately, summary proceedings offer the possibility of quickly seeking an injunction, as the publishers in Munich have successfully done.

TK has extensive experience in dealing with unlawful online statements, reputational damage and the practical application of the DSA. See also the article in the Financiële Dagblad on this subject, in which our colleague Lex Keukens shares his views on this ruling.

Would you like more information?

Please contact Maurits or one of the other professionals from the team for an initial exploratory discussion. We’d be happy to work with you to identify the most effective approach for your situation.