Home News & ViewsAI and Copyright: The Advocate General’s Opinion Published in Landmark CJEU Case Like Company v. Google 19/11/2026 | Blog | Artificial Intelligence AI and Copyright: The Advocate General’s Opinion Published in Landmark CJEU Case Like Company v. Google Authors: Vilhelm Schröder, Jasmin Metwally, Joel Ettanen, and Veera Lahtinen Read time: 7min On 3 November 2026 Advocate General (“AG”) Szpunar delivered his opinion in Like Company v Google (Case C-250/25). This is the first ever generative AI case to reach the European Court of Justice (“CJEU”) and could provide important guidance for publishers and AI service providers to the extent adopted by the CJEU. While the Opinion is not binding on the CJEU, it provides important considerations for assessing whether the answers provided to the users of LLM-driven chatbots can be seen as infringing press publishers’ rights. The Questions Referred to the CJEU The questions referred to the CJEU concern both the use of press publications to train large language models and their use in generating chatbot responses. The substance of the questions can be summarised as follows: Does a chatbot’s response containing text partly identical to a press publisher’s online content amount to communication to the public, if the text is long enough to be protected by the press publishers’ right? If so, does it matter that the chatbot generates the response by predicting the next word based on observed patterns? Does training an LLM-based chatbot amount to reproduction, even where the model learns by observing and matching linguistic patterns If the answer to question 2 is yes, is that reproduction of lawfully accessible works covered by the text and data mining exception in Article 4 of the DSM Directive? If a user’s prompt matches or refers to text in a press publication, and the chatbot’s response displays some or all of that publication’s content, does this amount to reproduction by the chatbot service provider? The Questions on Training LLMs In the first part of his opinion, the AG takes the stance that questions 2 and 3 concerning training large language models (“LLM”) are hypothetical and, thus, should not be addressed by the CJEU. According to the AG, this is due to two reasons. First, there is no evidence supporting the assumption that Google’s Gemini was trained on articles by Like Company. Furthermore, while these two questions addressed to the CJEU concerned whether the use of press publications to train LLMs is subject to the exclusive right of reproduction granted to the publisher or whether such use is freely permitted under the text and data mining (“TDM”) exception, the training of LLMs cannot be seen as a form of “online use” and it, thus, falls outside the press publishers’ rights. The AG does, however, address questions 2 and 3 with the aim of drawing attention to the complexity of the issue, and urges the CJEU to proceed with caution on the matter. The AG concludes on a view that the use of copyright protected works and other subject matter for the purposes of training an LLM would, in principle, be permitted under the TDM exception in Article 4 of the DSM Directive, provided its conditions are met. These include lawful access and the absence of an appropriate reservation of rights by the rightsholder. The AG also recognises that exceptional cases may fail the three-step test, requiring a case-by-case assessment. Can a Chatbot’s Response Infringe the Press Publishers’ Right? The relevant part of the case, according to the AG, concerns essentially the question of whether the answers provided by LLM-driven chatbots to users’ queries infringe press publishers’ rights under 15 (1) of the DSM Directive, i.e., questions 1 and 4. This concerns retrieval-augmented generation (“RAG”), through which a chatbot retrieves online publications and uses them as context when generating an answer. RAG is distinct from training the underlying LLM. According to the AG, certain criteria need to be considered to answer the questions. The assessment must distinguish the technical copies made when retrieving a publication from the answer ultimately provided to the user. It must also be considered whether the relevant acts can be attributed to the chatbot provider. 1) Does the answer provided by the chatbot qualify as a reproduction of the press publication at issue or as making it available to the public? The AG suggests that this should be measured based on the usual infringement test conducted by the CJEU so as to remain technologically neutral: the relevant press publication must have been copied in order to create the content and at least some of the protected elements must be recognisable in the end result. According to the AG, scraping and reuse of a press publication through RAG involves copying, even where the original wording is not directly copied into the answer, However, recognisability must go beyond individual words or very short extracts. The decisive factor is whether the publication’s form of expression remains recognisable to such an extent that the answer directly substitutes the original article, allowing the user to dispense with reading the original article entirely. It is important to distinguish between protected and unprotected elements of a press publication. The AG points out that the press publishers’ right protects the “specific expression of information related to news”, not the reported facts. In this context, the AG further notes that as LLMs today can easily paraphrase content instead of replicating it, the concept of reproduction should not merely be limited to straight copying of press publications, to avoid excessively limiting the press publishers’ rights. Thus, the decisive question is whether the expression remains recognisable despite any paraphrasing. The AG also considers the separate right of making available to the public. According to the AG, making available may occur where the publication’s form of expression is systematically recognisable in answers to related user queries, to the same extent that those answers constitute direct substitutes for the works and other subject matter on which the model was trained. An isolated reproduction in an answer is therefore not enough to establish this form of use. 2) How has the chatbot accessed the press publication? The retrieval and analysis of a publication may involve technical copies even where the final answer does not reproduce its protected expression. According to the AG, the TDM exception does not cover the particular RAG process at issue, but the temporary reproduction exception in Article 5(1) of the InfoSoc Directive may apply if its conditions are met. In the absence of the publisher’s authorisation, this requires both lawful access to the publication and a lawful end use. Circumventing effective technical measures restricting access or scraping would prevent reliance on the exception. So would an answer in which the publication’s form of expression is recognisable to such an extent that it directly substitutes the original article. At the same time, making an article freely accessible online does not, in itself, imply consent to chatbot reuse. The AG recognises that blocking scraping bots may also affect a publisher’s visibility in search engines. 3) When can the acts be attributed to the chatbot provider? It is important to establish whether the acts can be regarded as carried out by the chatbot provider. This could be the case when such acts are inherent to the business model of the provider, or the provider has not installed adequate measures to avoid such acts. Such measures should address both compliance with access restrictions and the risk of reproducing protected expression in answers. The AG notes that his approach is “founded on a delicate balance”, which may result in a grey area between infringing and permissible chatbot responses. In his opinion, however, the key consideration is whether the chatbot provider has designed its service to respect copyright and related rights online and has put in place measures intended to prevent infringement. In his view, occasional accidents in this regard are not the defining factor. Applying this approach to Like Company’s case, the AG considers that the conditions for infringement do not appear to be met: the example does not appear to retain the article’s recognisable expression, and Google seems to have implemented appropriate safeguards. However, the final assessment of the facts remains for the referring court. Guidelines to be Drawn from the Opinion The Opinion highlights that using news facts is not the same as reproducing a publisher’s protected expression. For AI service providers, the practical focus is on lawful access, appropriate safeguards and whether potentially infringing uses are inherent in their business model. For publishers, lost traffic alone is not enough to establish infringement. Subject to the national court’s findings, the AG considers that the conditions for infringement do not appear to be met in Like Company’s case. If adopted by the CJEU, his proposed approach could provide important guidance for future disputes concerning AI and copyright. Contacts Vilhelm Schröder Specialist Partner vilhelm.schroder@hannessnellman.com +358 40 037 0585 Jasmin Metwally Associate jasmin.metwally@hannessnellman.com +358 45 133 4138 Joel Ettanen Associate joel.ettanen@hannessnellman.com +358 40 7152 597