Gemini answers may infringe EU press publishers' rights, CJEU adviser says
Gemini answers that retrieve and reuse online news articles in real time can infringe the press publishers' right that EU law grants to news organizations, according to an Opinion delivered on 3 November 2026 by Advocate General Maciej Szpunar at the Court of Justice of the European Union. The Opinion concerns Case C-250/25, Like Company, the first case on generative AI to reach the Luxembourg court, and it was published by the Court in press release No 155/26.
The dispute was brought by Like Company, a Hungarian operator of web portals, against Google Ireland before the Budapest Regional Court. According to the Court's summary, the publisher claims that Google used its press articles to train the large language models behind Gemini and that the chatbot reproduced those articles and made them available to users between June 2023 and May 2024, the period in which Google's chatbot was first branded Bard and then Gemini. The Hungarian court stayed its proceedings and asked the Court of Justice four questions about how the 2019 Copyright in the Digital Single Market Directive applies to this kind of system.
Szpunar draws a sharp line between the two halves of the case. On training, he proposes that the Court declare the questions inadmissible, because there is no evidence that Like Company's articles were used to train the model and because training is not an "online use" of press publications. He adds, for completeness, that training a large language model is as a rule permitted under the text and data mining exception. On chatbot answers, his view is less favorable to Google: a provider can infringe the press publishers' right when its chatbot fetches an article, reuses it to write a reply and hands that reply to a user, and the provider is liable when that use is inherent to its business model or when it has not taken credible and effective measures to counter it.
The Opinion is advisory. The judges have begun their deliberations and the judgment will be delivered at a later date, with no date announced.
··········
LIKE COMPANY V GOOGLE IRELAND: THE FOUR QUESTIONS BEFORE THE COURT.
The case turns on Article 15 of Directive 2019/790, the provision that created a neighbouring right for publishers of press publications in the EU. That right covers the online use of press publications by information society service providers, and it protects the publisher's investment in producing the publication. Article 15 leaves out hyperlinking, individual words and very short extracts, and the protection expires two years after publication, counted from 1 January of the following year. Article 4 of the same directive contains the text and data mining exception that AI developers rely on, which applies unless the rightholder has reserved its rights in an appropriate manner, such as machine-readable means for content published online.
According to the Court's press release, the Budapest court referred two questions about training and two about the chatbot's answers. The first pair asks whether training a model on press publications falls under the publisher's reproduction right or is covered by the Article 4 exception. The second pair asks whether the answers a chatbot gives to users can fall under the reproduction right or under the right of making available to the public, both part of the Article 15(1) protection.
The technical background described by the Court is central to the Advocate General's reasoning. Gemini uses retrieval-augmented generation, a design in which the system pulls information from the web at the moment a question is asked and feeds it to the model, so it frequently retrieves news articles when users ask about current events. Szpunar's analysis treats this real-time retrieval as the activity that can engage the publisher's right, separately from whatever data the model saw during training.
··········
HOW THE ADVOCATE GENERAL SORTS GEMINI'S USES OF NEWS ARTICLES.
The Opinion does not treat every contact between an AI system and a news article in the same way. The table below summarizes the positions set out in the Court's press release.
........
Activity | Advocate General's position | Deciding factor |
Training a large language model | Questions inadmissible; training generally allowed under the TDM exception | No evidence of training; training is not online use |
Retrieving an article to answer a user | Can infringe the reproduction right | Unlawful access, or answer recognisably reuses the article's expression |
Answers that systematically echo an article | Making available to the public is infringed as a rule | Expression systematically recognisable in answers to related queries |
Reporting facts or the news itself | Outside the press publishers' right | The right protects form of expression, not facts |
Provider liability | Provider is liable | Use inherent to the business model, or no credible countermeasures |
........
Two factors decide whether a chatbot reply infringes the reproduction right. The first is how the chatbot reached the article: Szpunar points to unlawful access, in particular the circumvention of a paywall or of other technical protection measures. The second is what the reply contains. An answer that only reports the facts in an article stays outside the right, while an answer that reuses the article's form of expression so recognisably that the user no longer needs to visit the original falls within it.
The threshold for the right of making available to the public is set higher. In the Advocate General's view, that right is generally infringed when the form of expression of an article is systematically recognisable in the chatbot's answers to related queries. On that reading, a single reply that borrows a sentence would not meet the description, while a retrieval pipeline that routinely turns a publisher's reporting into substitute answers could.
The Opinion repeatedly anchors the right in its purpose. The press publishers' right, Szpunar writes according to the press release, protects the investment involved in producing a publication and does not give publishers a monopoly over the news. Facts reported in an article remain free for a chatbot to state.
··········
WHAT THE OPINION MEANS FOR GEMINI USERS AND EU PUBLISHERS.
For people who use Gemini in Europe, nothing changes today. An Advocate General's Opinion has no direct legal effect, the Court of Justice has not ruled, and the national case in Budapest will only be decided after the Luxembourg judgment. Gemini's ability to answer questions about current events from web sources therefore continues as before, and the Opinion gives no indication that answers summarizing facts from news reporting would be affected even if the Court follows it.
The practical stakes sit with Google and the publishers whose work Gemini retrieves. If the Court adopts the Advocate General's approach, a provider's exposure in the EU would depend on concrete design choices: whether its retrieval respects paywalls and technical restrictions, how closely its answers track the wording and structure of the source, and whether it can show credible measures against substitute answers. Because the Court of Justice interprets EU law for all member states, the reasoning would apply beyond Hungary and beyond Gemini, to any AI chatbot that grounds its answers in press content.
Publishers who want to limit how Google's AI uses their pages already have one control documented by Google. The Google-Extended token in robots.txt lets a site manage whether content Google crawls may be used for training future Gemini models and for grounding answers in Gemini Apps and in Grounding with Google Search on Vertex AI. Google states that the token does not affect a site's inclusion in Google Search and is not used as a ranking signal. Google's crawler documentation also says it plans to support a shared AI training preference signal, written as user-agent: __usage_ai-training, in early 2027, a development covered in our report on the shared robots.txt token for AI training opt-outs.
........
Publisher control | What it covers | Effect on Google Search |
Google-Extended (robots.txt) | Gemini model training and grounding in Gemini Apps and Vertex AI | None, per Google |
Paywall or technical protection | Access to the article itself | Depends on the publisher's setup |
Shared AI training signal (planned) | AI training preference, from early 2027 | Not yet available |
........
For readers comparing Gemini's consumer tiers, the case does not touch pricing or plan limits; our guide to Gemini plans and pricing covers what each subscription includes. Regulators are examining the publisher question from other directions as well, as in the UK competition inquiry described in our article on the CMA investigation into Meta's terms for publisher content.
··········
THE TRAINING QUESTION AND THE TEXT AND DATA MINING EXCEPTION.
The part of the Opinion on training will attract attention across the AI industry even though Szpunar recommends that the Court not answer it. His primary position is procedural: the Court only answers questions that are relevant to the dispute before the national judge, and on the facts described there is no evidence that Like Company's articles were used to train Gemini. He also considers that training is not an online use of press publications, which places it outside the scope of the Article 15 right.
His secondary remark, offered for completeness, is that training a large language model is as a rule permitted under the text and data mining exception of Article 4. That exception allows reproductions and extractions for text and data mining when the developer has lawful access to the content and the rightholder has not reserved its rights, for online content through machine-readable means. The Advocate General's phrasing, "as a rule", leaves room for exceptions that the press release does not detail.
The practical result is that the Opinion separates two questions that publishers have often raised together. Whether a developer may learn from news articles during training is treated as a matter for the text and data mining framework, with its opt-out mechanism. Whether a chatbot may reuse a specific article when answering a user is treated as a matter for the press publishers' right, judged answer by answer and system by system.
··········
WHAT HAPPENS NEXT IN CASE C-250/25.
........
Step | Status on 4 November 2026 |
Reference from the Budapest Regional Court | Made; national proceedings stayed |
Advocate General's Opinion | Delivered on 3 November 2026 |
Court of Justice judgment | Deliberations begun; no date announced |
Decision on Like Company's claim | For the Budapest court, after the judgment |
........
Under the preliminary reference procedure, the Court of Justice does not decide the Hungarian dispute. It interprets EU law, and the Budapest court then applies that interpretation to the facts, including the question of whether Gemini's answers about Like Company's articles met the thresholds the Court sets. The interpretation binds other national courts that face similar questions, which is why the eventual judgment will matter for AI providers operating across the EU.
··········
GEMINI AND THE EU PRESS PUBLISHERS' RIGHT: CONFIRMED FACTS AND OPEN QUESTIONS.
The verified record is limited to what the Court has published. Advocate General Szpunar delivered his Opinion in Case C-250/25 on 3 November 2026; he proposes that the training questions be declared inadmissible while stating that training is generally permitted under the text and data mining exception; he considers that chatbot answers built on real-time retrieval can infringe the reproduction right and, when an article's expression is systematically recognisable, the right of making available to the public; and he places liability on the provider when such use is inherent to its business model or is not credibly countered.
Several points remain open. The Court of Justice may depart from the Opinion, and the timing of its judgment is unknown. The press release does not say how the Advocate General applies his tests to the specific Gemini answers cited by Like Company, which is in any case a question for the Budapest court. The press release does not record any reaction from Google to the Opinion.
The direction set by the Opinion is still specific enough to act on. Retrieval that respects paywalls and technical restrictions, answers that report facts instead of reproducing a publisher's expression, and documented safeguards against substitute answers are the conditions under which, on Szpunar's reading, a chatbot such as Gemini stays within EU law.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
datastudios.org
