Skip to content
Annuaire
Sections
News

Like Company v Google: AI chatbots and press publishers' rights

L’essentiel

The CJEU Advocate General considers that a chatbot reusing articles in real time can infringe publishers' related right. What this implies.

À retenir

The CJEU Advocate General considers that a chatbot reusing articles in real time can infringe publishers' related right. What this implies.

On 3 November 2026, the First Advocate General of the Court of Justice of the European Union, Maciej Szpunar, found that an AI assistant that fetches press articles in real time to answer its users can, under certain conditions, infringe press publishers’ related right. His opinion in the case Like Company v Google (C-250/25) is not binding on the Court. It nevertheless gives media outlets, chatbot developers and companies that publish online a first European interpretive framework.

A Hungarian publisher against Gemini

According to the page dedicated to the case on the CJEU website, Like Company operates news article portals in Hungary. The company claims that Gemini’s large language models were trained on its content and that Google’s assistant can produce summaries of it on demand. A Budapest court referred the matter to the Court in April 2025, recalls Digital Policy Alert, and the hearing took place on 10 March 2026.

The question raised is twofold: does this use constitute a “reproduction” and a “communication to the public” within the meaning of EU law? The technical heart of the case is retrieval-augmented generation (RAG): Gemini consults web pages at the moment the user asks a question, including the very latest news, then draws a summary from them. Publishers see this as taking their articles without authorisation or remuneration, diverting readers from their sites and cutting into their revenue.

What the Advocate General finds for real-time answers

According to the analysis published on 9 November by lawyer Kenza Declercq in Village de la Justice, the Advocate General draws a clear distinction between training the model and the search carried out at the moment of answering. For this second stage, the text and data mining exception does not apply. Temporary technical copies may fall under another exception, that of temporary reproduction.

The related right would be infringed in two cases:

  • if access to the content was unlawful, for example by circumventing a paywall or a bot block;
  • if the article’s own form of expression remains recognisable in the answer, to the point that the user no longer has any reason to open the original publication.

Reporting mere facts remains free. On the basis of a provisional assessment, these conditions do not appear to be met in the Like Company case, but it is for the Hungarian court to decide. The Advocate General also proposes easing the burden of proof: the court could presume use as soon as the content was accessible to the assistant and the publisher presents a reasonable sample of similar answers.

Model training, addressed in the alternative

The two questions concerning training should be declared inadmissible, Maciej Szpunar considers, since there is no evidence that Gemini was trained on Like Company’s articles, and training is not an “online” use within the meaning of the related right. He nonetheless examines them. His position: training in principle falls under the mining exception provided for in Article 4 of the 2019 Copyright Directive, unless the rightholder has opted out. He sets three conditions, including lawful access and a case-by-case assessment in which the developer must show that it has limited the risk of outputs too close to the original data. A model’s parameters are generally not copies of works, but an answer that reproduces an extract word for word can be one.

French publishers divided

On the press side, the reception is mixed. According to mind Media, publishers’ associations are torn between disappointment, caution and satisfaction. Mélanie Loubersac, managing director of DVP, the French press related rights collecting society, sees it as a “worrying signal for press publishers”: in her view, the conditions set are so restrictive that the summaries produced by AI-powered search engines would largely escape the related right.

What content publishers can do now

Without waiting for the judgment, the Village de la Justice article recommends several steps for any site that publishes valuable content, whether a media outlet or a company:

  • express a machine-readable reservation, via the robots.txt file, and back it up with a clear statement in the terms of use;
  • protect content technically (subscription, bot filtering), as a developer who circumvents these barriers could no longer rely on the exception;
  • keep dated evidence, as the reservation only applies going forward;
  • specify by contract who can exercise the opt-out and who can grant paid licences;
  • monitor the training data summaries that providers publish under the AI Act.

The limitation is clear: this opinion remains a proposal. The Court may depart from it, and the date of its judgment is not yet known.

Frequently asked questions

Is the Advocate General’s opinion binding on the Court?

No. It informs the judges, who may depart from it. The CJEU judgment, whose date is not known, will then be used by the Hungarian court to settle the dispute; according to Village de la Justice, it will also guide ongoing disputes in France.

Can a chatbot still summarise the news?

According to this interpretation, yes for facts, which are not protected. The risk arises when the answer reuses an article’s own wording to the point of replacing reading it, or when the content was obtained by getting past a paywall or a block.

Why has the question of Gemini’s training not been settled?

Because, for the Advocate General, nothing shows that the model learned from Like Company’s articles. He considers these questions inadmissible, while giving his opinion in the alternative: training would be covered by the mining exception, unless the rightholder has opted out.

Sources

Sur votre appareil

Comprendre cet article

L’analyse utilise l’intelligence locale du navigateur lorsqu’elle existe, sinon un résumé extractif. Le texte n’est envoyé à aucun service extérieur.

Facebook X LinkedIn

Ensuite A lire aussi

Free, no spam, one-click unsubscribe.