An adviser to Europe's highest court has sided, at least in part, with a news publisher taking on Google over its AI chatbot. Advocate General Maciej Szpunar told the Court of Justice of the European Union on Tuesday that Gemini's use of press stories to answer user queries can infringe copyright rules in certain circumstances, in one of the most consequential legal opinions yet on how AI answer engines treat news content.
The advisory opinion came in a case brought by Like Company, a Hungarian website operator, against Google's Gemini. For publishers across Europe — and for every AI company that summarizes the news — the stakes are existential. Traffic has dwindled as AI-written summaries increasingly replace the traditional search results that once sent readers, and revenue, to publishers' own pages. A ruling against Google that forces the tech giant to pay fees for using their content could send a financial lifeline to struggling online media. For a wider view of how governments are responding to similar pressures, see our AI policy news coverage.
What Szpunar Actually Said
The core of the opinion is deceptively simple. "When an LLM (large language model)-driven chatbot retrieves a publication from a press publisher's website, reuses that publication to generate an answer, and provides that answer to the user, the press publisher's right may be infringed in certain circumstances," Szpunar said, according to a court statement.
The phrase doing the heavy lifting is "in certain circumstances." Szpunar did not say chatbots automatically infringe every time they summarize an article. Instead, he identified specific conditions under which Gemini's answers fall under the publishers' right of reproduction.
When Chatbot Answers Cross the Line
Two situations stand out in the opinion. The first: chatbots that get access to publications by unlawfully circumventing paywalls or other technical barriers. The second: answers that reuse "the form of expression of that publication in a recognisable manner."
That second test may prove the most litigated in practice. Szpunar drew a sharp line between facts and expression: copyright rules do not protect the "mere facts" reported by press publications, but "only the specific form of expression." In other words, a chatbot that extracts the underlying facts of a news story and re-expresses them in its own words may be on safer ground than one that echoes a publisher's distinctive phrasing, structure, or framing closely enough that the source remains recognisable inside the answer.
For publishers, that distinction is both a shield and a frustration. It protects the creative work of journalism without granting ownership over the news itself. For AI companies, it sets a compliance target that is far fuzzier than a URL allowlist: they must now think not only about what information their models convey, but how recognizably they convey it.
Training Stays Outside the Line of Fire
Just as important is what the opinion did not do. Szpunar said the use of copyright-protected material to train large language models is "freely permitted under the text and data mining exception."
That carve-out separates the two frontiers of the AI-copyright fight. On the training side, where models ingest vast amounts of text to learn language patterns, the adviser sees the EU's existing text and data mining rules as sufficient. On the output side — where a deployed chatbot fetches a specific article and hands its substance to a user in exchange for nothing — the press publishers' right can bite.
The distinction mirrors the structure of the case itself. Like Company's complaint is not that Gemini read its article, but that Gemini's answer replaced the article, capturing the value of the publisher's work at the moment of consumption without a license.
A Non-Binding Opinion With Real Weight
Opinions of advocates general are not binding on the Court of Justice. But they are usually followed by the court's judges, and they routinely frame how lower courts, regulators, and companies read the eventual judgment. The court's final ruling in the Like Company case will follow in due course, and both sides will spend the intervening months pressing their readings of Szpunar's reasoning.
The timeline matters for an industry moving faster than litigation ever does. AI answer engines are already negotiating content licenses with major publishers, and deals signed today embed assumptions about what the law requires. If the court follows Tuesday's opinion, products built on the assumption that summarization is free — so long as training was lawful — may need to retrofit licensing, attribution, or technical controls for how they retrieve and re-express press content.
Google, for its part, now faces a European legal environment in which the most influential judicial voice on the question has blessed the core of the publishers' argument: that the press publishers' right, created to protect investment in news gathering, extends to the moment an AI chatbot turns a publisher's article into someone else's answer.
The judgment, when it comes, will not resolve every question — the line between protected expression and unprotected facts will keep lawyers busy for years. But the direction of travel is now clear, and it points toward a web in which AI answers carry a content bill of their own.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
