A German judge just made Google responsible for what its AI says
Munich court holds Google liable for AI Overviews defamation, a legal first with broad implications for search, publishers, and AI-generated content worldwide.
A Munich civil court issued what may be the most consequential judicial ruling in the short history of AI-generated search results. On May 28, 2026, the Landgericht Munchen I held Google directly and primarily liable for false and defamatory claims generated by its AI Overviews feature. The judgment, under case reference 26 O 869/26, came from the court’s 26th civil chamber following an oral hearing on April 23. Google must bear 80 percent of court and legal costs. Violations of the injunction carry fines of up to EUR 250,000 per instance or custodial sanctions.
The case, reported by PPC Land on June 10, was brought by two publishing companies forming part of a Munich-based media group operating 12 branded imprints. The second plaintiff is a subsidiary publishing books and magazines under the GeraMond brand, concentrating on technology and history. The triggering queries combined each company’s name with the German word “Betrugsmasche,” meaning fraud scheme. AI Overviews generated false and defamatory text in response to those queries. The harmful responses were first documented on January 20, 2026. Formal notice was sent to Google on February 2. The oral hearing took place on April 23. The injunction was issued May 28, nearly four months after the publishers first identified the problem.
What makes this ruling structurally different from every prior European case is the direct liability finding. Until May 28, courts examining AI Overviews had either dismissed cases or framed potential liability in conditional terms. A Frankfurt Regional Court ruled in September 2025 that AI Overviews can materially harm competition and reduce click-through rates, but dismissed the plastic surgeons who brought that case because the specific AI summary was found “ultimately not false” when read in complete context. The Frankfurt judges did state that an objectively incorrect AI Overview “could constitute an unreasonable obstruction” under German competition law. Munich has now taken that conditional possibility and made it a judgment. The conditionality is gone.
The legal architecture matters precisely because of how Google has contested every prior case. Google’s consistent argument is that it functions as an intermediary, surfacing and distributing information rather than authoring it. That argument has some grounding in the history of platform liability: search engines were not historically treated as publishers of the content they indexed. AI Overviews does not index. It generates. The Munich court applied that distinction directly. When a system produces new text, the operator of that system bears authorial responsibility for the accuracy of what it produces. Hosting a link to a false claim is different from generating the false claim. The court found the latter is not intermediation.
The practical scale of the problem is documented across multiple research bodies. Ahrefs published data in February 2026 showing AI Overviews correlate with a 58 percent reduction in click-through rates for top-ranked pages in affected query categories. Earlier Ahrefs data from March 2025 measured a 34.5 percent organic click reduction for pages appearing in results with an AI Overview present. A separate study cited in prior European proceedings found zero-click searches increased from 56 to 69 percent after AI Overviews launched in May 2024. Organic search traffic to news publishers fell from 51 percent to 27 percent of their total traffic between 2023 and 2025, with AI Overviews contributing substantially. These are not merely editorial concerns. They are the measurement basis for advertising revenue models built on organic traffic.
The legal pressure has been building along multiple axes simultaneously. In July 2025, the Independent Publishers Alliance filed a formal antitrust complaint with the European Commission, alleging that Google’s AI Overviews misuse publisher content without compensation or opt-out, causing measurable traffic and revenue losses. The European Commission opened separate formal proceedings in December 2025 examining whether Google breached Article 102 of the Treaty on the Functioning of the European Union by using publisher content to build AI Overviews without appropriate compensation or choice. In the United States, Penske Media Corporation, owner of Rolling Stone, Variety, and Billboard, filed a federal antitrust case in September 2025 arguing Google coerces publishers into supplying content for AI training while diverting their readers. A Minnesota solar company filed a separate US suit over specific false AI-generated claims about its business. The Helena World Chronicle case, which a US federal judge dismissed in April 2026, turned on antitrust framing rather than content liability, and its dismissal left the direct defamation route untested in American courts.
The Munich ruling is a preliminary injunction, not a final judgment. It addresses a specific set of false claims about specific plaintiffs, not the AI Overviews product category in general. Google has both the right and the resources to contest it through the full appeals process. But it establishes that a European civil court, applying German defamation law, will hold the operator of an AI-generated search feature directly responsible for the accuracy of what that feature produces, on financial penalty terms. That is a materially new operating condition for any platform generating summaries at the scale AI Overviews operates. Publishers, legal teams, and search practitioners who have been watching this litigation arc since the Frankfurt decision now have a first actual injunction to point to. Any platform deploying AI-generated content summaries at scale, not only Google, now has a European precedent in which the generative nature of the output was the basis for holding the operator directly liable.
The Munich ruling landed on the same day as a dense set of programmatic and platform developments that together sketch the shape of the market heading into the second half of 2026.
Samsung Ads announced on June 10 that its Smart TV home screen placements are now available through programmatic channels for the first time. The Trade Desk and Google DV360 are the two named DSPs at launch, with Magnite’s SpringServe ad server powering the supply side. Global rollout begins in Q3 2026, with additional DSP partners expected. Home screen placements are structurally distinct from mid-roll or in-app inventory: they appear before any app opens, before any content plays, at the moment a viewer switches on the set. Until this announcement, those placements were only accessible through direct deals with Samsung Ads. Magnite Group SVP Mike Laband described the move as making “high-impact CTV placements more seamless, scalable, and measurable.” How third-party measurement companies classify and verify this surface is still being worked out. Samsung and The Trade Desk are actively educating buyers on how to think about the new offering. The IAB’s relevant CTV standards remain in development, and Samsung said it chose not to wait for the standards process to complete.
A DV360 platform roadmap last updated May 22 details more than 12 changes arriving between now and August. The June wave includes the deprecation of legacy audience expansion for YouTube campaigns, which is replaced by lookalike audiences requiring manual migration with no automatic transition of existing configurations. The Data Onboarder dimension renames from “AdAdvisor by Neustar” to “AdAdvisor by TransUnion,” following a corporate rebrand at the data provider. Floodlight and Inventory Availability reports begin a phased migration from offline to instant reporting, creating a window where data behaviour may differ depending on timing. The August wave is structurally heavier: digital content labels, the brand safety exclusion taxonomy in continuous use since DV360’s early years, are deprecated. Sensitive category exclusions for display, video, CTV, and audio go with them. Inventory Modes and content themes replace both, but there is no automatic mapping from legacy label structures to the new ones. Advertiser-level category exclusions including News and Politics also retire in August. Deal creation changed in April as well: manual non-programmatic guaranteed deal creation was deprecated, with the Deal Sync API now the sole method for creating new deals. On June 10 specifically, the DV360 API completed its rollout of full Demand Gen support for all partners, closing a gap that had required manual UI work for certain Demand Gen campaign operations since the format was introduced.
AppsFlyer’s State of Fraud for Marketers 2026, published June 10, draws on 106.4 billion installs across 246,000 apps from Q1 2025 through Q1 2026. Organic traffic now accounts for 52 percent of all fraudulent mobile installs, more than any paid channel. That headline figure matters not as a standalone number but because of what it does to the rest of mobile measurement: every mobile marketing team uses organic install rates as the internal benchmark against which paid channels are evaluated. Inflate that organic number, whether through deliberate fake installs designed to look like self-discovered traffic or through attribution failures that route paid installs into the organic bucket, and every paid channel’s measured performance is systematically overstated. The fraud is not just wasted spend. It corrupts the baseline against which every other metric is read.
DSP fraud grew 59 percent year over year. Owned media fraud, covering email and push, grew 221 percent. Spoofing, which fabricates devices, users, and behavioral event sequences from scratch rather than hijacking genuine installs, outpaced total install growth every quarter between Q3 2025 and Q1 2026. Its specific danger is detection resistance: clean-looking signals, not anomalies, are generated. Device fingerprints, behavioral patterns, and event timing are all constructed to pass standard verification. AppsFlyer notes this makes spoofing the most likely type of fraud to be undercounted in any dataset, including the one underpinning the 2026 report itself. In Finance on Android, Real Users Lift, the metric measuring what share of reported installs reflect actual human users, has been locked between 50 and 53 percent for five consecutive quarters with no improvement. In Social Media on iOS, it reached 275 percent in Q2 2025, meaning three in four installs were fraudulent for an entire quarter. Every ROAS figure, cohort analysis, and growth metric built on Social Media iOS data during Q2 2025 was measuring an audience that was largely absent. The affiliate-to-self-reporting-network fraud gap stands at 36 times, a figure that concentrates the practical risk squarely on media plans that include affiliate channels with limited third-party oversight.
LiveRamp connected its Conversions API Hub to ChatGPT on June 10, announced by Travis Clinger, Chief Connectivity and Ecosystem Officer. The CAPI architecture routes conversion event data from an advertiser’s own server infrastructure directly to the destination platform, bypassing browser-based tracking. ChatGPT’s conversational interface creates fragmented signal conditions: users move between sessions, devices, and interaction modes in ways that make client-side tracking structurally unreliable. Server-to-server connections bypass those constraints. Meta’s own data shows CAPI users see a 17.8 percent lower cost per result on average compared to those relying solely on client-side tracking. LiveRamp’s Hub already connects DIRECTV, which became the first MVPD to integrate in April 2026 ahead of the upfront season. The ChatGPT integration arrives as the ad platform accelerates: self-serve Ads Manager opened May 5 with CPC bidding at a recommended $3 to $5 per click, daily budgets and ZIP-code geo-targeting were added May 22, and conversion-optimised campaigns launched June 5 with eligibility gated on having at least one active conversion event flowing through the pixel or CAPI by June 1.
Minerva launched publicly on June 9 with $20 million in seed funding from The General Partnership, 8VC, Lingotto Innovation, Topology Ventures, and NBA Investments. The Brooklyn-based company was founded by Jackson Engles, Daniel Saedi, and Matthew Joseph, who worked at Lazard, Bridgewater, and Citadel respectively before building a platform targeting broken first-party data inside marketing teams. The product combines an Agentic Data Engineer for data unification, a proprietary identity graph with over 1,000 consumer attributes, campaign creation and optimisation automation, and performance reporting. The OpenAI collaboration uses GPT-5.5, released on April 23, 2026, across two product workstreams. The agents execute defined tasks without requiring machine learning expertise to supervise them. Approximately three dozen customers were already signed before the public launch. The Trade Desk launched its own in-platform AI agent, Koa Agents, in April 2026 with Stagwell as pilot partner. Adobe introduced AI agents for digital marketing the same month. Amazon has run an Ads Agent for natural-language campaign management since November 2025. Minerva’s argument is that the limiting factor across all of these is not the quality of the model but the quality of the first-party data the model acts on.
The EU ordered Meta on June 10 to restore free API access to WhatsApp for competing AI assistants, citing serious risk to competition in the AI market. Third-party AI systems must access WhatsApp on the same technical terms as Meta’s own AI products. The order sits alongside the Munich ruling as a second European regulatory action in a single day against AI-related platform behaviour, and sets a precedent for how dominant messaging platforms can structure access controls in regulated markets.
Affinity joined AdCP as a Founding Member on June 10, pushing the Ad Commerce Protocol’s work on agentic advertising standards to cover OEM, browser, app store, and TV OS surfaces that programmatic has historically not reached. Both the Samsung home screen announcement and this AdCP expansion point in the same structural direction: inventory that never sat inside the programmatic stack is being pulled into it, either through existing DSP integrations or through new governance frameworks purpose-built for agentic buying.
Acxiom’s InfoBase Geo-Based audience segments went live in The Trade Desk’s data marketplace on June 10, covering the UK and Germany, extending location-anchored programmatic targeting in two major European markets simultaneously. Amazon Ads unified reporting reached general availability the same day, consolidating campaign data across accounts in a single interface and retiring two legacy Ads Console tools on December 31, 2026. Adelaide added AU pre-bid targeting to Amazon DSP, letting buyers screen Display, Online Video, and Streaming TV inventory by attention quality score before placing a bid rather than optimising around attention only in post-campaign analysis.
Fubo and NBCUniversal signed a distribution deal on June 10 restoring NBC, Bravo, Telemundo, four NBC Sports regional sports networks, NBCSN, and new FAST channels to Fubo’s subscriber base, adding ad-supported live sports and entertainment inventory that had been absent from the platform. The Trade Desk named David Haddad, former president of Warner Bros. Games, to its board on June 10, adding media and entertainment expertise as the company rebuilds after a Nasdaq compliance notice issued during a period of stock price pressure in late 2025.
Apple updated its Applebot documentation on June 8, with PPC Land reporting on June 10. The update adds AI training controls, nosnippet directives, X-Robots-Tag support, and new crawl behaviour detail tied to how Siri processes publisher content for AI-generated responses. Publishers who have not audited their robots directives since AI crawling became standard face potential exposure. iOS 27 features announced June 10 include Tap to Share for contactless data exchange at retail checkout, Local Lists in Maps, and Visual Intelligence bill-splitting, none of which are ad products but each of which introduces a behaviour pattern that will become relevant to local and retail campaign planning.
Google released Gemini 3.5 Live Translate on June 10, a speech-to-speech model supporting over 70 languages inside Google Meet, the Translate app, and the Gemini API. Real-time speech translation at that language breadth, delivered through standard enterprise tools, changes the cost structure for multilingual content and international customer communication operations at scale. Knicks versus Spurs became the top US search query on June 10 as Google Trends data showed vintage jersey searches at all-time highs and Victor Wembanyama at a career search peak, a demand signal with direct operational relevance for retail, apparel, and sports-adjacent paid search campaigns running through this week and beyond.
Also noted
June 10, 2026: APA and Edison Research surveys put 2025 US audiobook revenue at $2.43 billion, up 9 percent year over year, with 157 million active American listeners, over 750,000 active titles, and rising YouTube piracy. A sharp drop in willingness to purchase AI-narrated titles was also recorded, with the 2026 survey showing significantly lower consumer acceptance of AI narration compared to the prior year’s data.
June 10, 2026: Novi’s analysis of 10.7 million ChatGPT citations across more than 98,000 sources found Reddit ranking above Sephora and Allure when ChatGPT recommends beauty, skincare, and fragrance products. The dataset points to a structural gap between where beauty brands have historically invested in owned content and where AI systems find the sources they trust enough to cite, with community-generated content from Reddit scoring higher than brand-controlled editorial and retail properties.
June 10, 2026: Semrush analyzed 89,000 LinkedIn URLs cited by AI search tools, identifying content structure, employee advocacy patterns, and B2B brand visibility signals that correlate with AI citation rates. The analysis provides the first granular picture of how professional network content earns placement in AI-generated responses, with implications for B2B marketers optimising for generative engine visibility rather than traditional organic search rankings.
June 10, 2026: Kinsta launched free Bot Protection for all WordPress hosting plans, giving site owners dashboard control over AI crawlers and automated traffic inside MyKinsta without touching server configuration files. The launch follows Apple’s Applebot documentation update and the broader industry push toward publisher-controlled AI crawl governance.
June 10, 2026: EMARKETER projects US AI ad spending will more than double to $68.25 billion by 2030, while placing ChatGPT’s US chatbot advertising ceiling at $5 billion by the same year, with over 80 percent of total AI ad spend flowing beside AI-generated content rather than inside chatbot interfaces. The forecast frames the distinction between advertising adjacent to AI and advertising inside AI as the defining budget allocation question of the next four years.
June 10, 2026: Google DeepMind selected 15 European robotics startups for a three-month equity-free accelerator with access to Gemini robotics model infrastructure and direct mentorship from the DeepMind team, as Google continues its push to position Gemini as the foundational model layer across both digital and physical AI applications.