Who Owns an AI Overview? Google, Liability, and Sources

Who is legally responsible when a Google AI Overview says something false? In the space of nine weeks, three separate events gave three different answers. A Munich court ruled on May 28, 2026 that an AI Overview is Google’s own statement and Google is directly liable for false ones. A U.S. federal judge ruled on July 22, 2026 that plain search-result output is not a work protected under the Copyright Act. And on July 30, 2026, Google itself pointed at the web pages it sourced.

That last event is today’s news hook. After a controversial AI Overview answer drew public criticism on X, Google’s official account responded that the output was “picking up particular web pages for this specific search” — attributing the answer to its sources rather than to its own generation process. Search Engine Roundtable, which reported the exchange on July 30, 2026, noted the obvious tension: Google has separately argued in court that its search results, including AI Overview responses, are protectable works.

This post is not about the underlying answer — we deliberately won’t characterize it beyond “a controversial answer.” It is about the liability mechanics now taking shape around AI Overviews, and what a publisher or brand can practically do when one gets it wrong: which court said what, on which date, which defenses have already failed, and the documentation-and-remedy playbook that follows from all three events.

The exchange itself was short. A political advisor and podcaster criticized an AI Overview answer on X, and Google’s official @NewsFromGoogle account replied publicly. Per Search Engine Roundtable’s July 30, 2026 report, Google’s response was: “Seems like this response is picking up particular web pages for this specific search.” Google added that the differing response was “not for all queries,” and that Gemini — Google’s standalone chatbot — does not produce the same answer to the equivalent prompt. It closed with: “Thanks for flagging, the teams are looking into it, will work to improve.”

Read those three statements as a liability position and they say something specific. First: the output is a function of what the sourced pages said, not of what Google generated. Second: Google’s own products disagree with each other on the same underlying query intent — which undercuts the idea that the answer was simply “what the web says.” Third: Google committed to fix it, which is a quality commitment, not a defense of the output as correct.

For publishers, the first statement is the one that matters. If Google’s public posture is that AI Overview errors originate in “particular web pages,” then the sites an AI Overview cites are being positioned — rhetorically, at least — as the responsible party for what the AI Overview says. That posture is exactly what the Munich court examined and rejected in May, and it lands eight days after that posture collided with Google’s own courtroom arguments about who owns this output. It is also part of a broader pattern of publisher-facing AI Overview friction — Google’s AI Mode rollout in France met organized publisher resistance on different but related grounds.

Trade coverage has mostly treated these as separate stories. Put them on one timeline and a tension emerges that no single ruling resolves: Google has pursued legal theories that treat its search and AI Overview output as protectable — with mixed success — while disclaiming authorship of that same output when it draws criticism.

The protectable-output leg comes from Google’s December 2025 lawsuit against SerpApi, a search-results scraping service. Google alleged violations of DMCA Section 1201’s anti-circumvention rules — SerpApi bypassing Google’s “SearchGuard” anti-bot barrier — with Google’s filings describing its search results, “including AI Overview responses,” as works protected under the Copyright Act. Search Engine Roundtable made that connection explicitly in its July 30 report on the blame event; the framing is the article’s own, not our inference.

The leg did not hold up well. On July 22, 2026, U.S. District Judge Yvonne Gonzalez Rogers (N.D. Cal.) dismissed Google’s DMCA claims, ruling that plain, aggregated search-result data — URLs, snippets, factual index data — are “publicly accessible facts” and not works protected under the Copyright Act. Those claims were dismissed without leave to amend. On the narrower question of third-party content Google displays (knowledge-panel material, for example), the judge found Google does not own the copyright to most of it, and dismissed those claims with leave to amend — Google has 21 days from the ruling to refile if it can show authorization from the actual copyright holders.

The dismissal was not a total loss for Google. The court affirmed that Google has standing to sue under the DMCA even without being the sole copyright owner, and agreed that spoofing browser fingerprints, rotating IPs, and solving CAPTCHAs to bypass SearchGuard constitutes “circumvention.” But circumvention alone isn’t illegal unless it guards copyrighted material with the rights-holder’s authorization — and on the core question, the court said plain search output is facts, not protected works.

The accurate way to state Google’s position, then, is not “Google claims copyright over AI Overviews” — the plain-facts version of that claim was dismissed. It is that Google has pursued IP-style protection over its output where that protects its business, while attributing the same output to source pages where authorship is a liability. Courts are now resolving that tension from both directions — and so far, neither direction favors Google’s framing of the moment.

The case that already answered the authorship question is Regional Court of Munich I (Landgericht München I), case no. 26 O 869/26, decided May 28, 2026. Two Munich-based publishing companies sought a preliminary injunction after an AI Overview falsely tied them to scams, subscription traps, and dubious business practices. The court granted it. The most detailed account came from The Decoder on June 9, 2026 (updated June 11), and the ruling is independently corroborated by the U.S. Library of Congress’s Global Legal Monitor in an item dated July 17, 2026.

The central holding goes directly at the “we just show search results” framing. The court held that an AI Overview is Google’s own content because the AI rewrites and judges results in its own words and according to its own structure, rather than merely linking to sources.

Two findings make this more than a technicality. First, the court found the AI Overview made claims “that are not even made in the search results” — none of the linked source pages drew the connections between the plaintiffs and the unrelated (genuinely dubious) companies that the AI Overview asserted. The court called these “the defendant’s own statements.” That is the factual pattern that matters most for brands: the harm wasn’t a bad source being quoted — it was the synthesis layer inventing a connection no source made. Blaming “particular web pages” has no purchase on that pattern.

Second, the remedy had teeth. The injunction bars Google from repeating specific categories of false claims — scam allegations, links to dubious companies, subscription traps, phone calls that never happened, false unavailability claims — with only two minor requests denied. Google was ordered to pay 80% of the legal costs in the case, with the two plaintiffs paying 10% each. Secondary trade coverage reports a threatened fine of up to €250,000 for violating the injunction; we have not verified that figure against a certified translation of the court order itself, so treat it as reported rather than primary-sourced.

Google’s on-record response, via a spokesperson to The Decoder (published June 11, 2026): “We invest deeply in the quality of AI Overviews to ensure that the overwhelming majority of responses provide accurate information, and they are designed to reflect the information that exists on the web... We’re carefully reviewing this decision, which is not yet final.” Google has indicated it intends to appeal — a stated position, not a confirmed appellate filing. Until an appeal resolves, this is a first-instance ruling: a live precedent in one jurisdiction, not settled European law.

The Munich ruling is most useful to publishers as a map of which arguments no longer work — at least before this court. Google ran the full defensive stack that has protected search engines for two decades, and the court worked through each layer. Per The Decoder’s account of the reasoning:

Two threads in the reasoning deserve emphasis. On the click-through defense, the court drew a parallel to German press law, where a publisher is liable for a misleading teaser even if the reader never opens the full article — and The Decoder notes the court’s skepticism is empirically grounded, citing Pew Research findings that users almost never click AI Overview source links. A confident, standalone answer is the product; the citations are decoration most users never touch.

The court also addressed free-speech balancing in a way that will likely be quoted for years: AI-generated “opinions” receive reduced protection because an AI’s output is “not the expression of an acquired conviction of the persons expressing it, but the result of an algorithm,” and offering AI-powered answers is “above all an expression of Google’s business activities.” Weighed against the plaintiffs’ rights — especially since the statements were untrue — Google’s interest lost. The reasoning rhymes with a broader 2026 pattern of holding AI operators accountable for what their systems assert; emerging disclosure standards for AI agent incidents are pulling in the same direction from the voluntary side.

Existing coverage is either pure legal analysis or pure SEO remediation advice. What a brand actually needs is one table that lines up the type of harm against where the law stands and what to do first. This is our original synthesis of the Munich reasoning, the SerpApi ruling, and the practical remediation consensus — as of July 30, 2026.

The fourth row is where most brands actually live, and it has no courtroom answer. For that, the work is visibility strategy: our full SEO strategy for the AI Overview era covers earning the cited position, and the zero-click data behind AI Overviews’ reach quantifies what is actually at stake when the answer absorbs the click.

The Munich plaintiffs won in part because they could show exactly what the AI Overview said, that no cited source said it, and that they had tried to get it fixed. That is a documentation posture any brand can adopt today, in any jurisdiction — and it doubles as the evidentiary record a lawyer would want if a Munich-style claim ever becomes relevant where you operate. The moment you spot a bad AI Overview about your brand, capture eight things:

1. The exact query text — verbatim, including any variant phrasings you tested.
2. Timestamp — AI Overviews regenerate; the same query may produce different output tomorrow.
3. A screenshot — of the full AI Overview as rendered, citations visible.
4. The verbatim AI Overview text — copied out, not just screenshotted, so it is searchable and quotable.
5. The cited source URLs — and whether each one actually supports the claim. The Munich court’s decisive finding was that none did.
6. Thumbs-down report filed (Y/N, date) — the “Report a problem” control beneath the AI Overview, with your short explanation saved.
7. Source-correction request sent (Y/N, date) — if a cited page carries the error, contact its publisher; if the cited page is yours, fix it.
8. A re-check date — put a calendar entry on it and re-run the query; log whether the output changed.

Why the source-side fix comes before anything else: SEO-industry consensus across multiple independent guides is that correcting the underlying cited page is the most reliable lever for changing AI Overview output, because AI Overviews regenerate from current source content rather than serving a cached, permanent snapshot. Google’s reporting channel, by contrast, comes with no guaranteed turnaround and no confirmation a correction will be made — public reporting and Google’s own community-support threads both confirm this. Knowing which pages AI Overviews tend to cite is half the battle; our citation-pattern study of 1,000 AI Overviews maps how those citations actually behave.

None of this would matter much if AI Overviews were an edge feature. They aren’t. Per Similarweb’s “2026 Generative AI Landscape” report (as covered by TechCrunch on July 27, 2026), AI Overviews now appear in roughly 43% of Google searches — up from about 15% of searches a year earlier. Semrush independently estimates the figure at closer to 48% of Google searches. The two vendors measure differently, so treat the truth as a range, not a single number — but under either estimate, nearly half of Google queries now open with a synthesized answer that some court, somewhere, has ruled is Google’s own statement.

That reach is also why AI Overviews sit at the center of Google’s business narrative — AI Overviews’ role in Alphabet’s latest earnings makes clear the feature is strategic, which cuts both ways: Google has every incentive to defend it, and regulators and courts have a large, visible target.

On the monitoring side, Semrush’s brand-misinformation framework (published May 8, 2026) is the most structured public methodology: track the accuracy and frequency of AI mentions separately — a brand can be mentioned often but described incorrectly, or rarely but critically — and watch for four recurring error patterns: outdated information, fabricated details, competitive misattribution, and missing-product invisibility. (Semrush states its AI Visibility Toolkit monitors 213 million prompts across ChatGPT, Google AI Overviews, and Perplexity — a vendor capability claim, not an independently audited figure.) Whatever tooling you use, the framework transfers: measure how often you appear, separately measure whether what appears is true, and route each failure into the incident record from Section 06. If you would rather have a team operationalize that loop — monitoring, correction, and the content work that earns the citation in the first place — that is exactly the shape of our agentic SEO service.

Looking forward, the trajectory matters more than any single ruling. The Munich decision is one court, first instance, under expected appeal — but it is the first time a court has squarely held that the synthesis layer, not the sources, authors the answer. If that holding survives appeal, or gets replicated in other EU jurisdictions, the notice-and-takedown era of search liability ends where AI answers begin: platforms would own their summaries the way publishers own their teasers. If it falls, the July 30 posture — sources are responsible for what the AI says about you — becomes the operative reality, and the only reliable defense is controlling what the citable web says about your brand. Either way, the publisher playbook is the same today: document, fix the source, report, and keep the record.