EU Court Adviser Raises Copyright Questions Over Google Gemini

An adviser to the European Union’s top court has said Google’s Gemini chatbot may infringe copyright when it retrieves and reuses press reporting in certain ways. The opinion, reported by Reuters on November 3, is not a final court judgment. It draws an important distinction between facts, which copyright does not protect, and protected expression, which an AI-generated answer could reproduce recognizably. The adviser also addressed situations involving paywalls and other technical restrictions.

The dispute highlights the tension between AI search answers and the economics of reporting. Publishers invest in gathering information and attracting readers to their own pages; a chatbot that answers a question directly can reduce the incentive to click through. Google maintains that Gemini produces original responses and points to controls available to publishers. The eventual court ruling, expected in the coming months, could clarify how those interests interact under EU law.

A legal line between retrieval and training

Advocate General Maciej Szpunar’s opinion in the dispute involving Hungarian publisher Like Company does not say that every AI summary is illegal. Instead, it distinguishes the process of using material to train a model from what the system displays when a user asks a question. Reuters reported that Szpunar considered EU text-and-data-mining exceptions applicable to training on copyrighted works, and that he found no evidence Google had trained Gemini on the publisher’s articles. That finding narrows the immediate dispute rather than resolving the broader debate about model training.

The output side presents a different legal problem. If Gemini reproduces a recognizable part of a publisher’s original expression, the fact that it appears in a generated response rather than a conventional copied webpage may not eliminate copyright concerns. Equally, reporting a fact learned from an article is not automatically copying protected expression. The distinction matters for developers trying to design answers that are useful without substituting for the original work.

Paywalls, search and publisher leverage

Szpunar also identified the possibility of infringement when a chatbot circumvents a paywall or other technical restriction to retrieve content. That issue is separate from whether a publisher allowed ordinary search-engine indexing. A news organization may welcome a search snippet that sends readers to its website but object to a detailed AI response that satisfies the reader without a visit. The European Publishers Council has pressed precisely that distinction, arguing that visibility in search should not require surrendering control over reuse in generative answers.

Google disputes the suggestion that its service simply copies articles. It describes Gemini as producing original responses based on learned language patterns and says publishers have controls. Those positions will be tested against the court’s interpretation of the applicable EU rules, rather than resolved by either side’s description of the technology.

What the court has not decided

An advocate general advises the Court of Justice of the European Union; the opinion is influential but not binding. Judges are expected to issue a ruling in the coming months. The opinion does not announce damages, require Google to shut down Gemini or establish that every publisher can demand payment for every factual answer. It does, however, put concrete boundaries—protected wording, restricted access and the difference between training and display—at the center of the coming decision. For publishers and AI companies alike, the result could determine whether they need to change how retrieval-backed answers are assembled and presented.

This report is not legal, financial, or health advice. See the disclaimer.