In the news · as of 7 November 2026
AI chatbots reusing web articles: what the EU Advocate General’s Gemini opinion means for your website
On 3 November 2026 the first generative AI case to reach the EU Court of Justice moved to its next stage. An Advocate General proposed when an AI chatbot that reads news articles in real time and answers from them infringes the publisher’s rights. The Court has not ruled yet.

Photo: Femy Praseeth, public domain (CC0)
In this guide
Short answer
As of 7 November 2026: In case C-250/25 Advocate General Szpunar proposed that a chatbot reusing online press articles may infringe the publisher’s rights if it got past measures such as a paywall, or if its answer reuses the form of the article so closely that the reader no longer needs the original, and only where the chatbot provider can be held responsible. Facts stay free. The opinion is not binding.
What the Like Company case is about
Like Company, a Hungarian company that runs news web portals, sued Google Ireland in Hungary. It claims that Google’s chatbot, first Bard and then Gemini, infringed its rights as a press publisher between 13 June 2023 and 31 May 2024 by reproducing its articles and making them available in answers. Gemini collects information from the web in real time to write its answers.
The Budapest court asked the Court of Justice four questions: two about training a language model on press articles, two about the chatbot’s answers. The right in question is the press publishers’ right in Article 15 of Directive (EU) 2019/790 on copyright in the Digital Single Market. It covers online use of press publications by online service providers, belongs to publishers established in the EU, lasts two years from 1 January after publication, and does not cover private use by individuals, hyperlinks, or individual words and very short extracts.
What the Advocate General proposed on chatbot answers
Advocate General Maciej Szpunar proposed that the two training questions be declared inadmissible: nothing showed that Like Company’s articles were used for training, and training is not an online use of a press publication. On the answers, he proposed three tests.
| Question | Proposed answer |
|---|---|
| When is an article reproduced? | If the chatbot accessed it unlawfully, for example by getting past a paywall or other measures against scraping, or if its form of expression is recognisable in the answer to the point that the user can do without the original entirely |
| When is it made available to the public? | If its form of expression is systematically recognisable in the answers to related user queries |
| When is the provider responsible? | If such use is inherent to its business model, or if it did not take the measures a reasonably diligent operator would take to prevent it |
The right protects the form of expression, meaning how an article is arranged, structured and phrased, not the facts it reports. But reproduction is not limited to word-for-word copying, otherwise the right would be too easy to get around.
Google argued that freely accessible articles may be used by its chatbot. The Advocate General disagrees: not having a paywall is a commercial choice, not consent. He notes that Google seems to use the same bots for its search index and for its chatbot, so, as Like Company explained, blocking them may cost search visibility. A publisher who keeps them to stay visible in search cannot be taken to have agreed to chatbot use.
On the facts he leans towards Google, subject to the Hungarian court’s findings: Like Company had no technical measures against scraping, its wording was not recognisable in the example answer Like Company submitted, and the use does not appear inherent to Google’s business, which seems to have measures against it.
Robots.txt, Google-Extended and AI training
In side remarks that are not part of his proposed answer, the Advocate General discusses training. In his view, training a language model is in principle covered by the text and data mining exception of Article 4 of the same directive, if the content was lawfully accessible and the rightholder has not reserved that use. Online, the reservation must be machine-readable. He leaves room for exceptional cases.
He names robots.txt as a widely used way to make that reservation and says it does not technically stop scraping. Google’s own method, the Google-Extended token in robots.txt, blocks use for training and for grounding Gemini answers; Google says it does not affect Google Search, which the Advocate General says would still need to be verified. He also mentions metadata, and does not exclude website terms written in plain language if machines can reliably understand them.
What the opinion means for AI chatbots and your company website
The directive does not mention company websites. A press publication must be mainly journalistic, regularly updated under one title, meant to inform the public, and published under editorial responsibility. An ordinary product or service site will usually not qualify, while a regularly updated company news or magazine section might. That is a reading, not settled law. Ordinary copyright applies either way.
- Decide what you acceptBeing read for search and AI answers, being used for training, or neither.
- Check robots.txtSee which AI crawlers and tokens it allows or blocks; our guide AI crawlers and robots.txt shows how.
- Block by token, not blindlySome AI uses ride on search crawlers. Use tokens such as Google-Extended rather than blocking a search crawler, so search indexing keeps working.
- Protect paid contentIf you publish news or paid articles, use measures that really restrict access, such as a paywall, and state your reservation in machine-readable form.
- Check your own assistantIf it fetches articles from other sites, let it summarise facts and link to the source rather than repeat long passages.
The last point is our reading. The opinion deals with Gemini and does not discuss assistants on company websites, but its tests are written for any language-model chatbot that retrieves press articles. We can review your crawler rules so they match what you want for search and AI answers; see AI search readiness. For assistants that answer from your own documents, see AI assistants.
This article explains the Advocate General’s opinion and the directive as they stand on 7 November 2026; it is not legal advice.
Questions we hear often
Want this sorted without doing it yourself?
Tell us what you are working on. A person replies within one working day.
