Paywalls and Phrasing: The Two Gemini Copyright Tests
An adviser to the European Union’s top court has set two narrow conditions under which a Gemini copyright claim by a news publisher could succeed. Advocate General Maciej Szpunar delivered the opinion on 3 November 2026 in Case C-250/25. It does not bind the Court, which has set no ruling date.
The Two Tests the Adviser Set Out
Szpunar identified two circumstances in which an AI chatbot’s answer can engage the press publisher’s right, the protection European law gives news publishers over the online use of their articles.
| Test | What it turns on |
|---|---|
| How the chatbot reached the article | Whether access was unlawful, for example by getting past a paywall or a measure designed to block automated scraping |
| What the answer contains | Whether the reply reuses the article’s form of expression recognisably enough that a reader no longer needs to open the original |
The second test rests on a distinction that runs through European copyright law. Facts are not owned. The particular wording, structure and arrangement a journalist chooses are. A chatbot that reports what happened is on safe ground; one that reproduces how a specific article told the story is not.
There is a threshold attached. On the separate question of making a work available to the public, the adviser’s view is that the right is generally engaged where an article’s expression is systematically recognisable across answers to related queries. One borrowed sentence would not clear that bar.
Where those conditions are met, responsibility sits with the chatbot’s provider if the use is inherent to its business model, or if the provider has not put credible and effective measures in place to prevent it.
Why the Training Claim Was Set Aside
Like Company, the Hungarian web portal operator that brought the case, argued both that Google trained Gemini on its articles and that the chatbot reproduced them in answers. The adviser recommended the Court refuse to answer the training questions at all.
He gave two reasons. Nothing in the file showed that Like Company’s articles had in fact been used to train Gemini. And model training, in his reading, is not an “online use” of a press publication, so it falls outside the press publisher’s right the claim was built on.
He addressed training anyway, for completeness, and the conclusion favoured Google. Training a large language model on protected material is, as a rule, permitted under the text and data mining exception in Article 4 of the European Union’s 2019 copyright directive. Two conditions apply: the developer must have lawful access to the material, and the rightholder must not have reserved its rights, which for online content has to be done by machine-readable means. That reservation mechanism is the subject of a separate industry effort to standardise a single AI training opt-out in robots.txt.
How the Adviser Read These Particular Facts
This is the part most coverage of the opinion has left out. Having set the tests, Szpunar applied them provisionally to the dispute in front of him, and his reading did not favour the publisher.
- Access: no technical restriction on the articles appeared to be in place, so there was nothing to circumvent.
- Output: the article’s form of expression was not apparent in Gemini’s answer, which conveyed much the same facts in different words.
- Business model: the reproductions did not appear inherent to how Google operates the product.
- Mitigation: Google appeared to have measures in place of the kind the test requires.
That provisional view carries a caveat the adviser stated plainly: the findings of fact belong to the Budapest Regional Court, which referred the case and will decide the claim once the Court of Justice has interpreted the law. Whether Like Company’s portal even qualifies as a press publication is among the questions left to the national court.
What the Opinion Does Not Settle
As of 5 November 2026, nothing in European law has changed. An Advocate General’s opinion is advice to the Court of Justice of the European Union, not a judgment. The judges have begun deliberating, the Court has announced no date for its ruling, and reporting on the opinion puts the judgment “in the coming months” without a firmer timeframe.
The Court follows its advocates general in most cases but is free to depart from them, and this opinion runs to 240 paragraphs across questions no European court has answered before. It is the first generative AI copyright case to reach the Court of Justice, which is also why the outcome matters well beyond one Hungarian publisher.
What Changes for Publishers and Gemini Users
For publishers, the value of the opinion is leverage rather than law. It establishes that chatbot answers are not automatically outside the press publisher’s right, which strengthens a negotiating position in licensing talks even before any judgment lands. Some of those arrangements already exist in early form, including Google’s pilot paying a small number of sites whose pages shape its AI answers.
It also sets a ceiling. Because the protection covers expression rather than facts, a publisher cannot use this right to stop a chatbot from reporting the news it broke, only from retelling it in its own words. Szpunar’s reasoning gives publishers less than the headlines suggest.
For anyone using Gemini, there is no practical change. No feature has been withdrawn, no answer format has been altered, and Google has announced no product response. Google’s position, given to reporters on the day of the opinion, is that European text and data mining exceptions matter for the region’s competitiveness, and that “Gemini learns linguistic patterns to create original outputs rather than replicating content”. The company also points to the controls publishers already have to refuse certain uses of their material.
Like Company’s underlying complaint is commercial rather than textual. The publisher says its traffic has fallen as AI-written summaries take the place of search results that once linked to news sites, according to RTE’s report on the opinion. That grievance is shared widely across the industry, and this case does not resolve it.
What Happens Next
- Deliberation: the Court of Justice considers the opinion alongside the parties’ submissions. No ruling date has been set.
- Judgment: the Court answers the referred questions on how the press publisher’s right applies. Its interpretation binds every member state.
- Back to Budapest: the Regional Court applies that interpretation to the facts and decides whether Google infringed.
- Everything else: questions the opinion deliberately left open, including whether European copyright reaches model training carried out outside the European Union, wait for another case.
A comparable split is visible in other jurisdictions, where courts have treated training and output as separate problems. A United States appeals court took its own view of an AI training fair use defence earlier this autumn.
Frequently Asked Questions
Does This Opinion Stop Gemini Using News Articles?
No. It is advice to the Court of Justice, not a ruling, and it imposes no obligation on Google. Gemini continues to operate unchanged across the European Union.
Is the Gemini Copyright Opinion Binding?
No. Advocate General opinions are non-binding. The Court agrees with them in most cases but can reach a different conclusion, and only its judgment carries legal force.
When Will the EU Court Rule?
No date has been announced. Coverage of the opinion places the judgment within the coming months, which reflects the Court’s usual pace rather than any scheduled hearing.
Did the Adviser Say AI Training Is Legal?
He said training on protected works is, as a rule, covered by the text and data mining exception, provided the developer had lawful access and the rightholder did not reserve its rights in a machine-readable form. He also advised the Court to leave the training questions unanswered in this case.
Who Is Like Company?
A Hungarian operator of online news portals. It sued Google Ireland before the Budapest Regional Court, which referred four questions to the Court of Justice, making this the first generative AI copyright case the Court has taken.
What Period Does the Claim Cover?
Like Company’s allegations concern reproduction and availability of its articles through the chatbot between June 2023 and May 2024, a span in which Google’s assistant was renamed from Bard to Gemini.
