Google‘s use of press stories in its Gemini chatbot to respond to queries from internet users can infringe copyright rules in certain circumstances, an adviser to Europe’s highest court said on Tuesday. Advocate General Maciej Szpunar was advising the Luxembourg-based Court of Justice of the European Union in a case brought by Hungarian website operator Like Company against Google’s artificial intelligence chatbot Gemini. A ruling against Google, forcing the tech giant to pay fees for the use of publishers’ content, could send a lifeline to online media outlets such as Like Company, which have seen traffic dwindle as news summaries written by AI chatbots like Gemini increasingly replace the traditional search engine results that include hyperlinks back to their sites. The case sits at the center of a wider conflict between publishers and AI developers, one that has been building for several years and that now appears to be reaching a point of legal reckoning in Europe.
Szpunar’s opinion was carefully calibrated. He said that when a 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. The qualifiers matter. He did not declare that Gemini’s behavior is unlawful in all cases, and he did not hand publishers the sweeping victory some had hoped for. Instead, he identified specific conditions under which infringement would occur. Gemini’s answers fall under the publishers’ right of reproduction, he said, when chatbots get access to publications by unlawfully circumventing paywalls or other technical barriers, or when they reuse the form of expression of that publication in a recognizable manner. That distinction between accessing content improperly and reproducing its expression is the analytical core of the opinion, and it reflects a long-standing principle in European copyright law.
Copyright rules do not protect the mere facts reported by press publications, Szpunar added, but only the specific form of expression. It is a familiar boundary, one that allows a news summary to convey the substance of a story without appropriating the language in which it was told. A chatbot that reads an article about a corporate earnings report and answers a user’s question with the key figures would not, on that reasoning, be reproducing protected expression. A chatbot that reproduces the article’s distinctive phrasing, its structure or its stylistic choices in a way that a reader would recognize as coming from that source would be operating in different territory. The practical difficulty, of course, is that large language models are designed to absorb and imitate patterns of expression, and drawing a clean line between conveying facts and reproducing form is considerably harder in practice than it is in the abstract.
On the other side of the ledger, Szpunar said the use of material protected by copyright to train large language models is freely permitted under the text and data mining exception. That conclusion is significant because it preserves the legal basis on which AI developers in Europe have built their models. Training and output, in other words, are treated differently. The act of ingesting vast quantities of text to build a model falls within an established exception, while the act of generating a response that reproduces protected expression may not. For publishers, that split offers a partial remedy rather than a comprehensive one. It does not give them leverage over the training process, which is where much of the value in their archives is extracted, but it does give them a potential claim when the output of that training competes directly with their own product.
The stakes are difficult to overstate. Online media has spent two decades adjusting to the economics of search engines, which sent traffic to their sites in exchange for the right to index their content. AI chatbots disrupt that bargain. A user who asks Gemini for a summary of the day’s news receives an answer without clicking through to the publisher’s website, which means the advertising revenue that once sustained newsrooms does not materialize. The resulting decline in referral traffic has been severe for many outlets, and it has prompted a wave of litigation and licensing negotiations around the world. Some publishers have chosen to sign commercial deals with AI companies rather than pursue legal remedies, calculating that a predictable payment is preferable to an uncertain judgment. Others have concluded that the only way to establish the principle is through the courts, and the Like Company case is among the most closely watched of those efforts.
Opinions of the advocate general are not binding, but they are usually followed by the court’s judges, which gives Tuesday’s statement considerable predictive weight. A final ruling is expected in the coming months, and its influence will extend well beyond the parties involved. If the court adopts Szpunar’s reasoning, AI developers will need to examine how their systems retrieve and reproduce content, particularly where paywalls and recognizable expression are concerned. Publishers will gain a legal foothold but not a comprehensive solution to their economic troubles. What the case will not resolve is the deeper question of how journalism should be funded in an era when machines can summarize the news without ever sending a reader to the source. The law can assign liability, but it cannot restore an advertising model that technology has fundamentally altered.
