What the EU opinion proposes about Gemini and copyright

On November 3, 2026, Advocate General Maciej Szpunar advised the Court of Justice of the European Union (CJEU) that a chatbot’s retrieval and reuse of press articles may infringe publishers’ rights in some circumstances. His nonbinding opinion in Case C-250/25 considers how an article was accessed and whether an answer reuses its protected expression so extensively that readers no longer need to consult the original. It does not decide the dispute between Google Ireland Ltd. and Hungarian publisher Like Company. The CJEU’s summary of the opinion sets out the proposed analysis.

What the opinion proposes for chatbot answers

Szpunar’s proposed analysis has two central considerations. One is whether the chatbot accessed a press article unlawfully, including by bypassing effective technical measures such as a paywall. The other is whether its answer recognizably reuses the article’s particular expression so extensively that the user is entirely dispensed from consulting the original.

That distinction matters because the press-publisher right in Article 15(1) of the EU Copyright in the Digital Single Market Directive protects the specific expression of press publications, not the facts they report. It does not give publishers a monopoly over news itself. A chatbot can draw on a fact without necessarily reusing the journalist’s expression; Szpunar’s proposed analysis focuses on both access and the form and extent of reuse.

The dispute concerns Gemini’s retrieval and alleged reuse

The case began in Hungary, where Like Company sued Google Ireland Ltd. Like Company alleges that Google repeatedly reproduced and made its articles available through Google Gemini between June 13, 2023, and May 31, 2024.

The CJEU describes Gemini as using retrieval-augmented generation, also called “grounding,” to collect web information in real time. That mechanism is relevant to the dispute because a chatbot may retrieve press articles while answering news-related questions. The alleged repeated use remains a claim by Like Company, not a court finding.

The opinion also addresses the provider’s role

Szpunar’s advice goes beyond what appears in an individual answer. He said potential acts of use may be attributable to a chatbot provider when they are inherent to its business model, or when the provider has failed to put appropriate countermeasures in place credibly and effectively. That is part of his proposed legal analysis, not a finding that Google is liable.

Training is a separate question

Like Company also alleged that its articles were used to train Gemini. Szpunar advised that the training questions in the case be declared inadmissible: the case contained no evidence that Like Company’s articles had been used for training, and he considered LLM training not to be “online use” of press publications under the right at issue.

Separately, he observed that, as a rule, protected works may be used to train large language models under the text-and-data-mining exception in Article 4 of the directive. That general observation does not establish that Like Company’s articles were used to train Gemini.

The courts still have separate roles

The CJEU will interpret EU law; the Budapest Regional Court will resolve the underlying dispute under the CJEU’s eventual ruling. Szpunar’s opinion is nonbinding, so it does not determine infringement or compensation. The CJEU said its judgment would follow later.