TL;DR
- First Advocate General Maciej Szpunar delivered his opinion in Case C-250/25 Like Company v Google Ireland on 3 November 2026; CJEU press release No 155/26 calls it the first generative AI case at the Court.
- Infringement test for chatbot answers: unlawful access (such as getting past a paywall or anti-scraping measure) or an answer that reuses an article's expression so closely the reader can skip the original. Facts stay free.
- Szpunar advised that the two AI training questions are inadmissible, but suggested training is, as a rule, permitted under the EU text and data mining exception; PPC Land reports he treats robots.txt as a valid opt-out.
- The opinion does not bind the Court of Justice; judges began deliberating on 3 November 2026 and no judgment date was given.
On 3 November 2026, First Advocate General Maciej Szpunar told the Court of Justice of the European Union that an AI chatbot such as Google Gemini can infringe press publishers' rights when it retrieves a news article and reuses it to answer a user, but only in certain circumstances: if the chatbot got past a paywall or another technical block to reach the article, or if its answer copies the article's way of telling the story so closely that the reader no longer needs the original. He also suggested that training an AI model on protected works is, as a rule, allowed under the EU's text and data mining exception. The opinion in Case C-250/25, Like Company v Google Ireland, is not binding, and the Court's judgment will come later.
What the Advocate General said on 3 November 2026
The Court of Justice published the opinion with a summary page and a video explainer on its CURIA website, alongside press release No 155/26. The press release calls Like Company "the first generative AI case to reach the Court of Justice". That label matters because it means the EU's top court is, for the first time, being asked how existing copyright rules apply to a chatbot that writes its own answers.
The press release states that Gemini uses a technology called Retrieval-Augmented Generation, also called grounding. In plain terms, the chatbot collects information from the web in real time to improve its answers, and when users ask about the news it often retrieves (or "scrapes") press articles that are available online. The case is about what happens to those articles between retrieval and answer.
According to the CURIA press release, Szpunar's central points are these:
- The press publishers' right protects only the specific form of expression of a publication, not the facts it reports. Its purpose, the Court's press office wrote, is to protect the investment in making the publication, not to give publishers a monopoly over the news itself.
- Whether a chatbot answer reproduces a publication depends on two factors. First, whether the chatbot accessed the article unlawfully, in particular by circumventing effective technical measures such as a paywall or a measure designed to restrict web scraping. Second, whether the answer only reports facts, or reuses the article's form of expression so recognisably that the user can skip the original entirely.
- The separate right of making a publication available to the public is, as a rule, infringed where the article's form of expression is systematically recognisable, in that substitutive way, in the chatbot's answers to related queries.
- The chatbot provider is liable where such uses are inherent to its business model, or where it failed to put in place appropriate measures to counter them credibly and effectively.
How the case reached Luxembourg
Like Company is a Hungarian operator of several web portals. It sued Google Ireland before the Budapest Regional Court, relying on the press publishers' right in Article 15(1) of the 2019 Copyright in the Digital Single Market (CDSM) Directive. The press release says Like Company argued that Google used its articles to train the Gemini models and repeatedly reproduced them and made them available to the public through the chatbot between 13 June 2023 and 31 May 2024.
The Hungarian court paused the case and sent four questions to the Court of Justice. Two concern training: whether using press publications to train a large language model falls under the publisher's reproduction right, or is freely permitted under the text and data mining (TDM) exception in Article 4 of the CDSM Directive. The other two concern the chatbot's answers to users.
PPC Land, which published two detailed write-ups of the opinion on 3 and 5 November, adds background drawn from the opinion text. It reports that one of Like Company's sites, balatonkornyeke.hu, is free to read and funded by advertising, and that the company's main example was a Gemini summary of a 21 July 2023 article about a Hungarian singer's plan to bring dolphins to Lake Balaton. According to PPC Land, that summary included details that were not in the article at all, such as the dimensions of a pool, which Google described as model hallucinations. PPC Land also reports that the hearing took place on 10 March 2026 and that the European Commission and nine national governments submitted written observations.
The answer-side test, in one table
The table below sets out the parts of the opinion that the Court of Justice's own press release states directly, so each row can be read on its own.
| Question | Advocate General's view (press release No 155/26, 3 Nov 2026) |
|---|---|
| Are reported facts protected? | No. Only the specific form of expression of a press publication is protected. |
| When can a chatbot answer reproduce an article? | When access was unlawful, such as circumventing a paywall or anti-scraping measure, or when the answer reuses the article's expression so recognisably that the user can skip the original. |
| Who is liable? | The chatbot provider, where the use is inherent to its business model or it failed to take appropriate counter-measures. |
| Is AI training covered? | Training questions should be declared inadmissible; for completeness, training is as a rule permitted under the TDM exception. |
| Is the opinion final? | No. It does not bind the Court; judges are now deliberating and judgment will follow later. |
Training: inadmissible, then answered anyway
Szpunar advised the Court to declare the two training questions inadmissible. The press release gives two reasons: there is no evidence that Like Company's articles were used to train Gemini, and training a model is not a form of "online use" of press publications, which is the only use the press publishers' right covers. So, in his view, those questions are hypothetical.
He then gave his view anyway "for the sake of completeness": using works protected by copyright and related rights to train an LLM is, as a rule, freely permitted under the TDM exception. PPC Land's second write-up, focused on training, reports the conditions the opinion attaches to that view. In PPC Land's reading, the works must be lawfully accessible, which means not scraping obviously pirate sources and not circumventing paywalls or bot blocks; rightholders can opt out with a machine-readable reservation; and where a model could routinely produce content resembling specific training material, the developer should be able to show measures against memorisation, such as deduplication and output filters, or a national court should refuse the exception.
The same outlet reports that the opinion treats the robots.txt protocol as a valid machine-readable opt-out, along with Google-Extended (Google's robots.txt-based control) and metadata embedded in files, and that it does not rule out a plain-language reservation in a site's terms. It also quotes the opinion's line that a trained LLM is not a database, meaning the model's parameters are not themselves stored copies of the works.
Why this bears on AI search and robots.txt decisions
This is where the case touches marketing directly. PPC Land reports that the opinion itself notes the technology also powers Google's AI Overviews and AI Mode, and that the opinion says Gemini answers sometimes, but not always, include links to their sources.
The opinion gives technical access controls real legal weight. Under the test in the press release, a chatbot that circumvents an effective paywall or anti-scraping measure fails the first factor. Our analysis, not a statement in the opinion: if the Court follows this test, a site's crawler rules become part of a publisher's legal position in the EU, not only a technical setting.
PPC Land also reports a point that will interest anyone managing search visibility. According to its reading, the opinion rejects Google's argument that leaving articles freely accessible amounts to implied consent to chatbot reuse. The opinion notes that the bots used for grounding are not fundamentally different from those used for search indexing, and for Google appear to be the same bots, so blocking them could affect how a site appears in search. Google told the Court that Google-Extended lets sites block training and grounding without losing Search visibility; PPC Land reports the opinion said this "would need to be verified".
PPC Land further notes that the record before the Court predates two later developments it has covered: a Search Console setting Google introduced in June 2026 that excludes a site from AI Overviews and AI Mode grounding, and an announcement on 30 October 2026 that Google, Microsoft, OpenAI and Anthropic would honour a single robots.txt token for AI training. The opinion mentions neither, according to PPC Land.
What the opinion means for content owners
Read together, the press release and the secondary coverage point to a practical split. Facts are free for a chatbot to repeat. Distinctive expression is protected against close, substitutive reuse. Technical access controls, where they exist and are effective, must not be circumvented.
PPC Land reports that the opinion makes the distinction sharper: long, in-depth analysis is by nature more recognisable than a purely informative report of events, so it is better protected. A short factual news item, written in commonplace language, offers little to recognise. For brands that publish content, that is a useful framing: original analysis, proprietary data and a distinctive structure are the kinds of material this reading of the law protects best, while generic summaries are the kind an AI answer can restate freely.
On the specific Hungarian dispute, PPC Land reports that the opinion's guidance is unfavourable to Like Company: the publisher does not appear to have used any technical measures, Gemini's answer showed essentially the same facts in different words, and Google appears to have taken appropriate measures. The Budapest court, not the Court of Justice, will decide that.
What the source did not say
- The opinion is not a ruling. The press release states that it does not bind the Court of Justice and that judgment will be given at a later date. No date was given.
- Nothing in the press release addresses countries outside the EU, including Thailand.
- The press release does not mention robots.txt by name; that detail comes from PPC Land's reading of the full opinion.
- The opinion concerns conduct between June 2023 and May 2024. It does not say how Gemini, AI Overviews or AI Mode work today.
What this means for Thai marketers
EU law does not govern Thai websites directly, and the Court of Justice has not ruled yet. Still, two practical points follow for Thai brands and publishers. First, crawler and AI-bot rules in robots.txt are now being argued in court as legal signals, not only technical ones, so they deserve a deliberate decision rather than a default. Second, as our own analysis rather than anything the opinion says, the line Szpunar draws between facts and expression fits the logic of generative engine optimisation: facts are what an AI answer can restate freely, while distinctive analysis and original data are the material this reading of the law protects best. Teams that sell into the EU, or run EU-facing sites, should watch for the judgment.
FAQ: the Like Company v Google opinion
Is the Advocate General's opinion binding?
No, the opinion does not bind the Court of Justice. The Court's press release of 3 November 2026 says the judges are now beginning their deliberations and judgment will be given later; Advocates General propose solutions independently, and the Court may or may not follow them.
Can Gemini or other chatbots summarise news articles in the EU?
Yes, in Szpunar's view a chatbot can freely report the facts in an article. It risks infringing only if it reached the article by circumventing a paywall or anti-scraping measure, or if its answer reuses the article's form of expression so recognisably that the reader no longer needs the original.
Does the opinion say AI training on web content is legal?
Mostly, yes: Szpunar suggested training is, as a rule, permitted under the EU text and data mining exception, while advising the Court not to answer those questions at all. PPC Land reports that his view depends on conditions, including respecting machine-readable opt-outs such as robots.txt and avoiding pirate sources.
Does this apply to websites in Thailand?
Not directly; the case interprets EU law, and the Court's press release says nothing about non-EU countries. It is relevant to Thai businesses that publish for EU audiences, and as a signal of how courts may treat AI opt-outs.
Should I change my robots.txt because of this?
Not because of the opinion alone, since it is not a final ruling. It is a good prompt to check which AI crawlers your robots.txt currently allows or blocks, and to make that a deliberate business decision rather than an accident of defaults.
If you want help reviewing how your site is set up for AI crawlers and AI answers, Relevant Audience's GEO team can walk through the options with you.
