Google's AI search summaries have again been ruled Google's own responsibility in Germany: on October 18, 2026, German tech outlet The Decoder reported that the Aachen Regional Court had already, in a decision dated September 24, barred Google from continuing to spread false statements generated by its AI, using the same reasoning as the Munich Regional Court's ruling in May this year — AI summaries are not neutral relays but Google's own statements. The same report revealed that the appeal Google publicly announced in the Munich case never reached a decision; the case ended in a settlement instead.
The Munich case: accusations that appear in none of the sources
In late May 2026, Munich Regional Court I held Google directly liable for the content of its AI Overviews. Two Munich-based publishers had found that, for searches of their company names, the AI summaries linked them to fraud schemes, subscription traps and dubious business practices. The court found that the AI had stitched information about other, genuinely problematic companies onto the plaintiffs, and that several accusations appeared nowhere in the source links the summaries displayed. The judges' classification is the crux: the AI rewrites, combines and evaluates information in its own words, producing "independent, new, and substantive statements" — nothing like a traditional results page that merely lists third-party links. Google designed the system and controls the algorithms, and only Google can correct the errors. Google's defense, that the summaries carry a warning that content may be inaccurate and should be verified, was not accepted, and the court issued a preliminary injunction.
The Aachen case: someone else's record pinned on a doctor
The Aachen case is more jarring. A physician served as chief physician and professor of neuroradiology in Aachen from 2010 until late June 2026. On August 28, 2026, a colleague alerted him that Google's AI Mode, when explaining his departure, claimed he had sexually assaulted his secretary; follow-up searches with terms such as "suspension" and "court proceedings" returned long, detailed fabrications complete with an invented legal aftermath. In reality, the criminal case involved a different former university professor and had nothing to do with the doctor — something the court could confirm from its own records. The doctor explained the mix-up in detail in late August and asked for removal; in mid-September Google said it would take no action, a formal legal demand went unanswered, and the false texts stayed online. The Aachen Regional Court therefore banned further spread of the three disputed text variants under an emergency procedure, without hearing Google first, with fines of up to 250,000 euros per violation. The court wrote that whoever provides AI-generated search is also responsible for the answers generated in that context, and added that Google could have checked the output against the doctor's objections with its own AI and would have caught the errors itself. The decision is not yet final; Google can still challenge the injunction.
A settlement resolves the case, but the precedent stays undecided
The Decoder learned that the appeal Google loudly announced in the Munich case never produced a higher-court decision: the case was settled, on undisclosed terms, with Google and the affected publisher both declining to comment. That means the Munich reasoning was never tested on appeal — yet the Aachen court has already cited it directly. With two courts aligned, the view that generative answers count as the platform's own words is hardening into a consistent German position. Settlement has a distorting side effect, as The Decoder's analysis points out: specialized law firms and affected parties could deliberately probe AI search with targeted queries until the model stitches together defamatory statements, then file for injunctions and negotiate settlements — a business model built on individual cases, while no binding higher-court precedent ever emerges because nobody litigates to the end.
What this means for AI search
Read together, the two decisions send every generative search product the same message: disclaimers and "we are just a platform" do not shield a company from responsibility for what its system generates. In practice, the courts have drawn a duty to review: since only the platform can change the algorithms and check outputs against sources, refusing to act once an error is reported turns a technical problem into a legal one. Earlier disputes on the publisher side were about whether content is cited at all and whether site owners can opt out — the question handled by Google's global AI search controls for site owners. The German cases go one layer further: when a summary says something false, the platform answers for it as its own statement. The next step most often discussed is publisher compensation — if AI answers count as Google's original content, the publishers whose work feeds those answers may assert claims of their own, which would rewrite the cost structure of AI search entirely. For users and businesses, the actionable steps are concrete: keep screenshots and the exact queries when an AI summary mixes you up with someone else, demand removal in writing and keep proof of delivery. AI hallucination itself is nothing new; what is new is that German courts have started sending the bill to the platform — and two of them have now done so.
