Advocate General Maciej Szpunar told the Court of Justice of the European Union on 3 November that Google’s Gemini chatbot may infringe press publishers’ rights when it uses their publications to answer users. His opinion in the dispute between Hungarian website operator Like Company and Google was reported by CNA. The opinion is advisory: the court’s judges will decide the case. Szpunar drew a distinction between a chatbot’s use of a publication in an answer and the use of copyrighted material to train a large language model.
Key points
- Szpunar said a chatbot may infringe a publisher’s right when it retrieves and reuses a press publication in an answer.
- His distinction turns on protected expression, rather than the facts reported in an article.
- He said training on copyrighted material is permitted under the text-and-data-mining exception.
- The opinion in Like Company’s case against Google is non-binding.
Article 15 and Gemini’s answers
Like Company alleges that Gemini reproduced and communicated substantial parts of its protected articles without permission between June 2023 and February 2024, Crypto Briefing reported. The case, C-250/25, concerns Article 15 of the EU’s Digital Single Market Directive, which gives press publishers rights over online uses of their publications. Like Company’s complaint concerns what users received from the chatbot, rather than the material used to train it.
Szpunar described a particular sequence: a chatbot retrieves a publication from a publisher’s website, uses it to form an answer and supplies that answer to a user. In those circumstances, he said, the publisher’s right may be infringed. That conditional view matters to the dispute. It addresses the use of a publication in a response, while leaving the judges to rule on the case before them. Like Company alleges substantial reproduction by Gemini during the period covered by its complaint.
Szpunar also distinguished the information in a news report from the language used to convey it. Copyright does not protect “mere facts”, he said, but “only the specific form of expression”. A chatbot answer about the same events as a published article therefore raises a different issue from an answer that reuses the article’s protected wording. His opinion places the press publisher’s right in that distinction, rather than treating every answer based on a news report as an infringement.
Szpunar separates training from answering
Szpunar took a different position on training. He said the use of copyrighted material to train large language models is “freely permitted under the text and data mining exception.” Text and data mining is the analysis of material by automated systems, including during model training. His account of that exception does not extend the same answer to a chatbot’s subsequent use of a publisher’s expression in a response to a user.
The distinction concerns two uses of material by an AI system. Training involves processing text for a model. The conduct described in the opinion involves a chatbot obtaining a particular publication and using it when answering a particular request. Szpunar treated the first as permitted under the exception and said the second may infringe a publisher’s right in certain circumstances. That is a narrower position than either a general prohibition on training with press material or a general permission to reproduce it in answers.
Publishers and AI companies are also making arrangements for news responses outside this case. AI Affairs reported a multi-year partnership between CNN and Google for AI news responses. The Like Company dispute concerns an allegation of reproduction without permission. Szpunar’s opinion addresses how a publisher’s legal right may apply to that alleged use, rather than the terms on which a publisher might agree to provide its work.
Like Company’s case before the EU court
An advocate general advises the Court of Justice, but the judges are not obliged to adopt that advice. Opinions of advocates general are usually followed by the court’s judges, CNA reported. Szpunar’s view gives the judges a proposed way to distinguish training from answers under EU copyright rules. The judges’ ruling, rather than the opinion, will decide Like Company’s case against Google Ireland Ltd.
This is the first reference to the Court of Justice to address AI outputs and press publisher rights, according to Crypto Briefing. The dispute reached the court in April 2025. An oral hearing took place on 10 March 2026.
