The key question is: when can AI chatbots reuse news articles to answer questions? AG Szpunar makes a clear distinction with another relevant question, notably whether press publishers can object to AI training using their news articles.

Three takeaways from the Opinion:

1. Training is a separate question: AG Szpunar recommends leaving the training questions aside as hypothetical: training is not “online use” covered by the press publishers’ right under Article 15 of the DSM Directive. However, copyright may still apply, but the text and data mining (TDM) exception can cover it, provided the following conditions are met: lawful access, appropriate safeguards and no valid rights reservation.
2. Facts are free. Copying expression is not. An answer may infringe if an article’s wording, structure or arrangement remains recognisable enough to replace reading the original. Paraphrasing is no automatic escape; sharing the same facts in a genuinely different form is not enough. Systematically providing such substitutes may also amount to making the publication available to the public.
3. Access and design matter. Real-time retrieval of articles (RAG) is not covered by the TDM exception, but its technical copies may qualify for the temporary-copy exception if access and the resulting use are lawful. Bypassing effective paywalls or anti-scraping measures can defeat that protection. Free access alone is not consent to chatbot reuse. Providers may be responsible where infringement is built into their business model or safeguards fall short, not automatically for every user-driven slip-up.

The practical message: Publishers should review access controls and TDM opt-outs separately. AI businesses should check sources, outputs and safeguards, and consider licensing where needed.

What’s next: The CJEU will decide. On the evidence presented, the AG sees no apparent infringement by Google, subject to the national court’s findings.

The Opinion is available here