The Gemini news copyright dispute has reached the European Union’s highest court, where an Advocate General says real-time chatbot answers may infringe press-publisher rights in defined circumstances. The opinion addresses Google Gemini’s retrieval and reuse of online news, rather than declaring every AI summary unlawful.

 

The opinion draws four legal boundaries:

  • Facts themselves do not receive press-publisher protection.
  • Recognizable reuse of expression may qualify as reproduction.
  • Bypassing paywalls or scraping controls can make access unlawful.
  • The opinion is advisory and does not decide the case.

 

More on This Story

 

Gemini News Copyright Test Targets Real-Time Retrieval

First Advocate General Maciej Szpunar delivered the opinion on November 3 in Case C-250/25, Like Company. The Hungarian publisher sued Google before the Budapest Regional Court, which paused the national proceedings and referred four questions about EU copyright law to the Court of Justice.

 

Like Company alleged that Gemini reproduced and made its articles available through chatbot answers between June 13, 2023 and May 31, 2024. It also claimed that its articles were used to train Gemini’s large language models.

 

The case focuses on Gemini’s retrieval-augmented generation, sometimes described as grounding. This process lets a chatbot collect current information from the web and use it while composing an answer, including answers about news published after the underlying model was trained.

 

That distinction matters because a current retrieval request creates a separate chain of access, processing and output. The opinion asks whether those steps reproduce a protected press publication or make it available to a user in a way reserved to the publisher.

 

Recognizable Expression and Paywalls Shape the Test

Szpunar advised that a chatbot’s use of a publication can fall within the publisher’s reproduction right under two main conditions. One concerns how the service obtained the article; the other concerns what appears in the generated answer.

 

Access may be unlawful when the service circumvents effective technical measures designed to restrict reading or automated scraping, including a paywall. The opinion does not say that retrieving every freely accessible article is automatically unlawful.

 

The output question turns on expression. EU press-publisher rights protect the recognizable form in which reporting is written, while the underlying facts remain available for others to report. A chatbot may therefore summarize facts without necessarily reproducing the protected wording or structure of the source.

 

Risk rises when an answer reuses the publication’s expression recognizably enough that the user no longer needs to access the original article. The right of making a publication available to the public may also be implicated when this recognizable expression appears systematically across related queries.

 

The opinion places potential responsibility on the chatbot provider when the disputed use is inherent in its business model or when it fails to deploy credible and effective countermeasures. That approach would make product design, access controls and output safeguards relevant to liability.

 

Training Questions Were Treated as Hypothetical

The Hungarian court also asked whether using press publications to train a large language model falls within the publisher’s reproduction right or the EU text-and-data-mining exception. Szpunar advised the court to declare those questions inadmissible in this case.

 

The court record contained no evidence that Like Company’s articles had actually been used to train Gemini, according to the press release. The opinion also considered model training outside the specific online use of press publications covered by the publisher’s related right.

 

For completeness, Szpunar said protected works used to train an LLM are, as a rule, covered by the text-and-data-mining exception in Article 4 of the EU Copyright in the Digital Single Market Directive. That statement remains part of an advisory opinion, not a final judgment resolving all AI-training disputes.

 

The separation between training and retrieval is consequential. A company could face different legal questions when building a model from a dataset and when that model later fetches a particular article to answer a live query.

 

The Court’s Judgment Will Determine the Legal Effect

Advocates General independently propose legal solutions to help the Court of Justice decide cases. Their opinions are influential but do not bind the judges, and the court has not announced when it will deliver judgment in Like Company.

 

The Court of Justice will interpret the relevant EU rules, while the Hungarian court will ultimately resolve the underlying dispute using that interpretation. A final ruling would also guide other national courts hearing comparable questions about chatbot retrieval and press publications.

 

For publishers, the immediate issue is whether AI answers substitute for visits by reproducing protected expression. For chatbot operators, the practical questions include honoring access restrictions, limiting recognizable copying, preserving source links and documenting safeguards that counter repeated infringement.

 

The judgment may therefore shape how European AI services ground news answers without preventing them from reporting facts. Until it arrives, the November 3 opinion is a proposed legal framework rather than a ruling that Google infringed Like Company’s rights.

 

Image: Court of Justice of the European Union in Luxembourg, photographed by FrDr in September 2026 via Wikimedia Commons, licensed under CC BY-SA 4.0. Cropped to 16:9.