EU Advocate General says AI chatbots can infringe press publishers' rights in some cases

The opinion

On Tuesday, 3 November 2026, Advocate General Maciej Szpunar of the Court of Justice of the European Union delivered his opinion in a dispute between Like Company and Google. The case is the first before the Court concerning generative artificial intelligence. An opinion of the Advocate General is a proposal to the judges rather than a binding decision, and the Court will issue its judgment at a later date. The dispute started in Hungary, where Like Company, which operates several news portals, sued Google before the Regional Court in Budapest. That court referred questions on EU law to Luxembourg, and the Hungarian court will ultimately decide the case within the limits set by the Court's answers.

The publisher's claims

Like Company says Google's Gemini chatbot produced a summary of an article from one of its portals that was detailed enough for a user to have no need to visit the site. According to the company, this cut traffic to its portal and its advertising income. Gemini uses retrieval-augmented generation, which allows the model to gather information from the web in real time, and the chatbot often pulls press articles available online when answering questions about news. The company also alleges that Google used its articles to train Gemini and that the chatbot made them available in answers between 13 June 2023 and 31 May 2024. Szpunar noted that the file contains no evidence that Like Company's articles were used to train the Gemini models.

Training questions

Szpunar recommends that the Court declare inadmissible the two questions the Hungarian court asked about training AI models. He considers them hypothetical, since there is no evidence that the articles of the plaintiff were used to train Gemini. He also takes the view that training a large language model does not amount to online use of press publications. He indicated that, as a rule, using protected material to train a model falls under the EU exception for text and data mining. He asks the Court to concentrate instead on the questions about chatbot answers, which he considers relevant to the main proceedings.

When answers can infringe

The central point of the opinion concerns the real-time retrieval of articles and their reuse in chatbot answers. Szpunar says the publisher's right does not protect the facts reported in news, only the specific form of expression and the investment needed to produce a publication, and it is not meant to give publishers a monopoly on news. Whether copying falls within the reproduction right depends on two factors. The first is whether the chatbot accessed the article unlawfully, for example by bypassing a paywall or by scraping the page. The second is whether the answer merely reports facts or reuses the form of expression in a recognisable way, so that a reader has no reason to open the original. According to Szpunar, the right of making content available to the public is generally infringed where that expression is systematically recognisable in the answers.

What comes next

The Court has not yet ruled, and its judgment will be delivered at a later date. Opinions are not binding, so the judges may follow the proposed approach or depart from it. Once the Court has answered the questions, the Budapest court will decide the dispute itself, bound by the Court's answers on EU law. The outcome will determine how far the publisher's right reaches when a chatbot retrieves and summarises press content in real time.