KI Flash: When AI Answers Are No Longer Privileged

With two recent decisions, the Regional Courts of Munich I and Berlin II have, for the first time, taken a closer look at AI‑supported search and answer formats. Both cases concerned Google’s “AI Overview”.

Although the underlying facts differ, the core legal question in both decisions is essentially the same:

When does a platform have to treat the output of an AI feature as its own statement?

Search engines and many platform models typically act as aggregators and “technical tools” for finding third‑party content. As a rule, they benefit from liability privileges: they are usually only liable for third‑party content once they have actual knowledge of a legal violation and then fail to remove or block it (“notice and take‑down”). By contrast, they are generally directly liable for unlawful content that they themselves publish. This is precisely where the courts step in. 

Once platforms “adopt” third‑party content as their own, they are generally treated as if they had published it themselves.

The courts essentially apply this principle to AI‑generated outputs in these two decisions as well: if an AI output is understood as the provider’s own statement, courts no longer treat the service as a neutral intermediary. The special liability privileges enjoyed by classic search engines and host providers then apply only in a limited way, if at all. The provider is, in principle, liable as if it had authored the content itself. The two decisions take different approaches, but together they offer initial guidance on when courts tend in one direction or the other.

LG Munich I: “AI Overview” as the Provider’s Own Statement

In the case before the Regional Court of Munich I (judgment of 28 May 2026 – 26 O 869/26), a publishing company brought an action against Google in relation to the “Übersicht mit KI” (AI Overview) feature. When the company name is entered into the Google search bar, the autocomplete function already suggests, among other things, the term “Betrugsmasche” (“scam”). Once this suggestion is selected, an AI Overview appears above the conventional search results.

This overview consists of a continuous text in which the company is explicitly linked to allegations such as “unseriöse Geschäftspraktiken” (“untrustworthy business practices”), “Betrugsmasche” (“scam”) and “Abo‑Fallen” (“subscription traps”). The text is structured into several sections, includes links to third‑party websites, rephrases statements from sources in its own words and even contains concrete recommendations (“If you are dealing with …, be extremely cautious”, “If you have a subscription, try to cancel it in due time”, “If you receive unjustified demands, do not pay”). The overview also contains statements and conclusions that cannot be found in this form in the underlying sources.

The Munich court considers this AI Overview to be Google’s own substantive statement, not merely a technical display of third‑party content (paras. 33 et seq.). In particular, it stresses that:

• the AI generates a self‑contained narrative text that summarizes, structures and evaluates search results in its own words,
• it produces statements and links between pieces of information that are not contained in the underlying third‑party sources at all (so‑called “hallucinations”),
• from the perspective of a reasonable average user, the AI overview appears as an answer provided by Google to the search query, not as a neutral list of results; the advisory elements reinforce this impression.

On this basis, the court assumes that Google has “adopted” the AI content as its own. Google is therefore liable for unlawful AI overviews, in particular for untrue, reputation‑damaging factual allegations about the claimant.

In addition, the court finds that it is not sufficient simply to switch off the specific AI answer. Due to the AI’s “black box” character, similar content may be generated again at any time; in the court’s view, the risk of repetition remains. Overall, the decision shows that for newly generated, chat‑like answers, there is a relatively low threshold for assuming “appropriation” (“Zu‑Eigen‑Machen”) at least where the text does not merely summarize search results, but goes beyond them by creating its own content, giving concrete recommendations and, as in this case, partly relying on technical errors (hallucinations).

LG Berlin II: AI Answers as Search/Information Format in Trademark Law

In the case before the Regional Court of Berlin II (judgment of 1 June 2026 – 52 O 62/26), the focus was on AI‑generated texts that mention the claimant’s branded perfumes and at the same time highlight so‑called “scent twins” (“Duftzwillinge”) as cheaper alternatives, including links to the respective sellers. The AI texts described which vendors offer scent twins to the claimant’s branded perfumes and guided users via links straight to the websites of these vendors. In terms of substance, they largely stayed within what was reflected in the regular search results displayed below. The claimant regarded this as use of its trademarks by Google to promote knockoff products.

The court, however, denies that Google used the marks in its own right in the sense of trademark law (Art. 9 UMV). The starting point is the basic trademark use requirement: use only occurs where the sign is employed in the context of the user’s own commercial communication, i.e. to designate or promote that party’s own goods or services.

The court relies again on the user’s perspective: a “reasonably well‑informed and reasonably observant user” perceives the AI texts as a search and information format, not as advertising or Google’s own product communication. Such a user recognizes that the content is based on third‑party websites, that Google operates a search engine aggregating such content, and that Google itself does not sell perfumes. The court further emphasizes that the AI texts merely reflect the actual search results and that each statement in the “Übersicht mit KI” is backed by a link to the corresponding search result; there is no evidence of targeted selection or steering in favor of specific sellers.

Against this background, the court concludes that, although Google does display the marks within the AI answer, this display does not amount to trademark use by Google as part of its own commercial communication. In this context, the court therefore rejects an “appropriation” of the AI outputs.

Common Approach and Practical Takeaways

At first glance, the two decisions lead to different outcomes, but they are not necessarily contradictory. Both courts ultimately pose the same question: does the AI output still look like a search/result format that merely improves the user experience, or does it appear as an independent statement by the provider?

The answer depends mainly on the format, structure and content of the respective output. Taken together, these decisions draw an initial, soft line: it is not only “purely fictional” AI content that can trigger direct liability. Even a genuinely independent substantive processing of search results – going beyond a neutral presentation of sources – can already lead courts to treat the content as the provider’s own.

For all providers of AI‑supported search and answer systems, the concept of “appropriation” thus remains a central risk factor: the more an AI output appears as a distinct, evaluative statement and the further it moves beyond the underlying sources, the more likely it is that the provider will be treated as if it had authored the content itself.

Which concrete adjustments are advisable in any given case – whether in product design, answer logic, disclaimers or notice‑and‑action processes – depends on the specific architecture of the system. We would be pleased to support you in assessing existing AI functionalities from a legal perspective and in developing appropriate safeguards.