What actually happened
On 3 November 2026, Advocate General Maciej Szpunar issued an opinion in Case C-250/25, Like Company v Google, the first generative-AI case before the EU Court of Justice. A Hungarian publisher says Gemini reproduced material from its articles. The key question is what happens when a chatbot retrieves reporting in real time and turns it into an answer.
Szpunar suggested infringement could arise where a chatbot circumvents effective access or scraping restrictions, or reproduces a publication's recognisable expression so extensively that the user has no need to read the original. Facts alone are not protected in the same way. Rephrasing an article does not automatically remove the issue if its protected expression remains recognisable.
He recommended setting aside the questions about model training because there was no evidence that Like Company's articles trained Gemini. Separately, he said training is, as a rule, permitted under the EU text-and-data-mining exception. That is not a blanket permission overriding the exception's conditions, including valid rights reservations.
Why a small team cares
Consider a WordPress publisher selling specialist research behind a subscription. If an AI service gets past a real paywall and serves an answer that substitutes for the paid article, the opinion gives a reason to distinguish that conduct from ordinary search indexing or factual summaries. The publisher still needs to check whether access controls are technically enforced, rather than just hiding text in the browser.
Training and live retrieval are also separate commercial decisions. A publisher might want its free explainers cited in AI answers but not its paid research reproduced. As Monday's robots.txt story explained, vendor rules for training and grounding are still inconsistent. A robots.txt directive is a crawler preference, not the same thing as an enforced paywall.
Hype vs useful
This is a non-binding Advocate General opinion, not a finding that Google infringed copyright. The Court's judgment will come later. The case also concerns the EU's specific rights for qualifying press publishers: an ordinary blog does not automatically receive those rights simply because it publishes content.
Buying an AI-copyright monitoring platform now would be premature. The useful distinction is between public content you want discovered, paid content you restrict and answers that repeat recognisable expression rather than merely report facts. Blocking every bot may reduce useful citations as well as unwanted reuse. Start with knowing what your site actually exposes, and watch the ruling before making a costly policy change.
