Recently, a German court issued an injunction against Google after its AI Overview generated a false statement, and Germany’s media regulator subsequently weighed in, examining how AI search is reshaping the visibility of news sources.
Both actions point to the same new form of front-end page. Imagine you search “whether a particular company is involved in fraud.” On a traditional search results page, the system returns a list of third-party webpage titles and links. As a user, you must click the links, read the source pages, and reach your own conclusion. The search page itself does not declare which conclusion holds.
Answer-based AI search changes this structure. It generates a conclusion in its own voice at the top of the search results (for example, “this company does have fraud issues”) while leaving only a small number of reference sources alongside the summary. This is a hypothetical search page used to illustrate the mechanism; it is not the actual query or factual findings in the Munich case.
It is precisely this change in page structure that triggers two distinct yet concurrent legal questions on the same page: first, who is deemed to have spoken the conclusion at the top of the page? Second, beside that conclusion, which sources can still remain visible to the user?
In traditional search, a user enters a keyword, and the system returns a list of third-party webpage titles and links through automated crawling and indexing. The user clicks a link to visit the third-party website and reads the source material, deciding the veracity of the content themselves.
This interaction determines the specific action of the search service provider: the system merely sends the user elsewhere, without rewriting or synthesizing third-party content, and without expressing an opinion on behalf of the webpage author. Faced with the vast volume of information on the internet, a search engine cannot verify the content of every webpage before presenting links. Prior to receiving a specific, clearly identifiable infringement notice, the service provider has no obligation to pre-screen third-party webpages; only after receiving notice, if the provider fails to address the relevant links within a reasonable period such that the infringing content remains retrievable, does it bear the corresponding duty to cease dissemination.
When a search page directly generates a conclusion at the top, this action changes: the page no longer merely offers entry points into third-party sites; instead, the service provider completes the statement directly on its own page.
On
28 May 2026, the Munich Regional Court I (LG München I), in emergency
proceedings under case number 26 O 869/26, prohibited
Google from continuing to disseminate false statements concerning two
publishing houses in its AI Overview.
In this ruling, the court treated the generated AI summary as content attributable to Google itself. The court’s reasoning was as follows: when a search platform extracts, reorganizes, and synthesizes multiple sources through an algorithm and presents an entirely new statement directly at the top of the search page, the system has gone beyond “providing entry points to third-party webpages” and the service provider has instead delivered a conclusion on its own page. Since the service provider has published this statement on its own page and in its own voice, it should bear direct content liability when the content is materially false.
This is a temporary injunction issued under emergency proceedings; it is not a nationwide final precedent, nor does it abstractly confer the status of “publisher” or “media company” on Google or on AI search. The publicly available text at present relies on a non-court-hosted transcript of the decision and professional reporting. Its significance lies in clarifying the boundary of one product action: when a service provider directly delivers a conclusion on its own page, it must take responsibility for that statement.
The court’s question was “who bears responsibility for a wrong answer”; yet even when the answer itself is entirely accurate, a second question emerges on that same page: which news sources can still remain within the user’s field of view?
On 14 July 2026, the Commission for Licensing and Supervision of the German Media Authorities (ZAK) published its first administrative decisions concerning Google AI Overviews and Perplexity, confirming for the first time that German media law applies to AI search and chatbots.
ZAK disaggregated the functions on the same page: the generated AI answer may constitute the service provider’s own content; while the system’s selection of sources, display of further reading, and provision of links and recommendations constitute a “Media Intermediaries” function, since these directly affect the discoverability of third-party content.
Under sections 91 to 96 of the German Interstate Media Treaty (Medienstaatsvertrag, MStV), media law’s concern with the intermediary function is not the truth or falsity of a single answer, but rather media plurality and distributional fairness. The MStV requires media intermediaries to disclose their criteria for access, retention, aggregation, selection, and presentation, and prohibits unjustified systematic discrimination against journalistic-editorial content over which they exercise particular influence.
The court dealt with the attribution of liability for a specific false statement, while the regulator is concerned with whether news sources receive fair presentation on the reconfigured search page. The full administrative orders of ZAK remain unpublished at present, and Google has stated that it plans to challenge ZAK’s position.
When the same search page both authors its own conclusion and rearranges sources, different national legal systems place their focal control points at different stages of the process:
In June 2026, the UK Competition and Markets Authority (CMA), in requirements directed at Google, focused on publisher control and fair ranking. The CMA required Google to give publishers effective control mechanisms to decide whether and how their content is used in search AI features, along with appropriate attribution; it also required that organic search, including generative AI search, follow objective, non-discriminatory ranking criteria.
In the United States, the discussion continues to revolve around Section 230 of the Communications Decency Act. Section 230 protects interactive computer service providers from being treated as the publisher or speaker of information provided by a third party, but if the service provider participates, in whole or in part, in the creation or development of the information, that protection no longer applies. Whether an AI summary’s extraction and reorganization of source material constitutes participation in the creation or development of information remains without definitive answer from any comparable AI search ruling in the United States.
The different jurisdictions have not arrived at a single unified universal rule. What they collectively reveal is one reality: when a search page shifts from “a list of links” to “an answer combined with sources,” it simultaneously implicates legal requirements across multiple dimensions — content attribution, source use, and distributional ranking.
Since an answer-based search page in practice splits into two distinct product actions, building the system and keeping logs naturally requires two corresponding and clearly delineated recording dimensions:
First, for the answer-generation action: record which web sources the system referenced, and through what logic the conclusion was synthesized. When the direct answer displayed on the page faces authenticity or attribution challenges, the generation process behind the answer can be clearly traced.
Second, for the source-presentation action: record which third-party sources were selected for display in the same search query, and how they were ranked and presented. This makes the selection process and presentation logic for sources transparent.
When a product shifts from “sending the user elsewhere” to “delivering conclusions and sources directly on the page,” treating these two page actions as distinct and keeping proper records for each can make the operating logic of the entire page clearer and more transparent.