Google AI Overviews: German court makes Google liable for false answers

News · 13 Jun 2026 · Daniel Reid

Google AI Overviews now carry legal weight in Europe after a German court ruled that the summaries Google’s AI writes above the search results are Google’s own words, not a neutral list of links, and that Google can be held liable when they state something false. The Regional Court of Munich (Landgericht München I) issued a preliminary injunction on 28 May 2026 under case reference 26 O 869/26, and Google confirmed on Friday 12 June 2026 that it will appeal. For UK readers the ruling lands on a question that has been building all week: when an AI summary invents a damaging claim about a person, a business or a product, who is accountable, and what is your recourse?

This is one of the first decisions anywhere to put a search company on the hook for what its model makes up, and it arrives just as UK shoppers and researchers are leaning on AI answers more than ever. The detail matters, so here is what the court actually decided and why it changes the accountability question for everyone who reads an AI summary rather than the pages beneath it.

What the Munich court actually ruled

The legal hinge is deceptively simple. A traditional search engine surfaces a ranked list of other people’s pages, and European law has long treated that intermediary role generously: the platform is not usually the author of what it links to. The Munich judges decided that an AI Overview is different in kind. When Google’s model reads several sources and writes a fresh paragraph in its own voice, the court found, it produces “independent, new, and substantive statements” that are not present as such in any of the linked results. That makes Google the speaker, not a noticeboard, and a speaker can be sued for defamation when the statement is false.

In this case, according to The Decoder’s 10 June 2026 report, the summaries fabricated associations between two Munich publishers and scams and “subscription traps,” connections that were not even made in the underlying search results. The court issued a preliminary injunction rather than a final damages award, banning Google from repeating those specific claims and reserving penalties of up to EUR 250,000 per breach. Google was ordered to cover 80% of the costs. None of this is a finding that AI Overviews are unlawful as a feature; it is a finding that Google owns the output and must answer for the falsehoods in it.

It is worth being precise about Google’s position, because the appeal will turn on it. As reported by Reuters and carried by MarketScreener on 12 June 2026, Google’s statement reads: “This case focuses on specific and narrow errors, not the foundational way AI Overviews displays web content. We disagree with the ruling and plan to appeal.” In other words, Google accepts the summaries can err but rejects the idea that an individual mistake makes it the legal author of every line its model writes. A higher court will now decide whether that distinction holds.

Why this matters for Google AI Overviews in the UK

A Munich injunction does not bind a British court. The German ruling rests on German civil procedure and the country’s strong personality-rights tradition, which protects individuals and companies against false factual assertions more readily than English law does. So nobody in Manchester can cite case 26 O 869/26 to force a takedown. What it does is establish a persuasive precedent and a template: a national court has accepted, in reasoned terms, that an AI summary is a publication by the company that runs the model. That framing will travel, and UK claimant lawyers will have read it closely.

The UK already turned off the regulatory pressure on this exact feature. We covered the CMA’s strategic market status work on Google’s AI search and the practical fallout of how to turn off Google AI Overviews when you would rather read the sources yourself. The Munich decision adds a third front to that picture. Competition regulators worry about market power, data regulators worry about accuracy, and now the courts have shown that ordinary defamation and personality-rights law can reach AI summaries directly. For a feature that Google has pushed to the top of the most-used webpage on earth, that is a meaningful expansion of legal exposure.

The shopping dimension is the one most UK readers feel daily. AI answers increasingly summarise prices, stock, delivery dates and product comparisons, and when those summaries are wrong the cost is concrete rather than reputational. A wrong “in stock at” line or an invented price drop sends you to a dead end, and the buying-advice failures of AI answers have been a top consumer complaint this week. If the Munich logic holds on appeal, the same reasoning that makes Google liable for a false defamation claim would make it the author of a false commercial claim too.

Your recourse when an AI summary gets it wrong

If an AI Overview states something false about you in the UK, you are not without options, though none is as fast as a German injunction. The first route is data protection. Under UK GDPR, you have a right to rectification of inaccurate personal data, and the Information Commissioner’s Office has repeatedly said the accuracy principle applies to AI outputs, not just to stored records. A summary that asserts a false fact about a named person is processing inaccurate personal data, and Google operates a removal request process you can escalate to the ICO if it is refused.

The second route is defamation. English defamation law can reach a false, damaging factual statement, but it requires you to show serious harm and to navigate the question of whether Google is a publisher or a mere conduit, which is precisely the point the Munich court resolved against Google. The third route, for false commercial or comparative claims, runs through consumer protection and the CMA rather than the courts. The practical reality is that all three are slower and more expensive than most people will accept for a single bad answer, which is exactly why the verification burden has quietly shifted onto you, the reader.

This is where the broader accuracy problem bites. We have written before about the Gemini privacy settings UK users should check and about whether Gemini is worth it in the UK, and the same caveat runs through both: an AI summary is a confident draft, not a verified fact. The German court has now said the company that publishes that draft must stand behind it. Until appeals settle, the safest assumption for any consequential claim, a price, a medical detail, a fact about a named person, is that you check the linked source before you act on it.

The verification burden Google has handed readers

There is a quieter cost in all of this that the Munich case throws into relief. AI Overviews were sold as a time saver, the answer without the click. But if the answer can be wrong in ways that only the underlying pages reveal, the time saving is conditional on you not needing the answer to be reliable. For trivial queries that is fine. For anything that matters, you end up doing the reading you were promised you could skip, which is the opposite of the value proposition.

Google keeps expanding what these summaries do, from build-your-own mini apps in Search to richer Gemini answers across its products, and each step folds more judgement into the model and away from the source list. That is genuinely useful when it works. Our coverage of NotebookLM in the UK shows the upside of grounding an AI in documents you control, where citations are tight and checkable. The contrast with an open-web AI Overview is the whole problem: NotebookLM tells you exactly which passage it drew from, while a Search summary blends sources into a single confident voice with no audit trail you can follow at a glance.

What UK users should do this week

Three practical moves. First, treat any AI Overview claim that you would act on, spend on, or repeat as a draft to verify, and click through to at least one named source before you rely on it. Second, if a summary states a false fact about you or your business, send Google a rectification request in writing, keep the timestamp, and escalate to the ICO if it is ignored; the accuracy principle is on your side even if the process is slow. Third, if you simply prefer the old links-first experience, you can suppress the feature, and our walkthrough on turning off AI Overviews still works for most UK accounts.

The bigger shift is cultural. For two decades the deal with a search engine was that it pointed and you decided. The deal with an AI Overview is that it decides and you trust. The Munich court has just told Google that if it is going to make the call, it has to own the call. That is the right principle, and the UK’s regulators, from the ICO to the work coming out of Google DeepMind in the UK, will be watching how the appeal resolves it.

MTW verdict

Our view is that the Munich court got the principle right even if the appeal narrows the remedy. If Google publishes a paragraph in its own voice at the top of the world’s most-used webpage, it cannot disown the sentences when they are wrong. The “specific and narrow errors” framing in Google’s appeal is doing a lot of work, because there is no clean line between a narrow error and a foundational one when the same system produces both. We do not expect a sudden wave of UK injunctions; English defamation and UK GDPR move slowly, and the realistic protection for now is your own scepticism. But the direction of travel is clear, accountability is moving towards the company that writes the summary, and that is overdue. Until then, read the source, not just the answer, on anything that matters.