The First Big Legal Crack in AI Answers — and What It Means for GEO
An EU court adviser just drew a line: training on public content is fine, but reusing a publisher's article in real time to answer a user may infringe their rights. If it holds, the economics of AI answers change.
Most of the AI-and-copyright fight so far has been about training — whether models can learn from scraped content. On November 3, 2026, an EU court adviser moved the fight to a more dangerous place for AI-answer products: what happens at answer time. And the line he drew should reshape how anyone thinking about generative engine optimization plans for the next few years.
What was decided (and what wasn't)
Advocate General Maciej Szpunar delivered a non-binding opinion in Case C-250/25, Like Company v Google Ireland. The core of it, as reported:
- Answer-time reuse is exposed. When a chatbot retrieves a press article in real time and reuses it to answer a user, the publisher's rights "may, in certain circumstances" be infringed.
- Training is not (here). Training models on publicly available material was treated as "freely permitted under the text-and-data-mining exception." The adviser called the training questions largely hypothetical in this case.
Crucially, this is an opinion, not a judgment. The Court of Justice of the EU still has to rule, and the Budapest court applies the result. AG opinions are influential but not binding — so read this as the clearest signal yet of where European law is heading, not as the destination.
Why the distinction is the whole story
The split between training and answer-time reuse is the most important line in AI law right now, because it separates two very different business activities:
Like this decoded read? Get the next one in your inbox.
One sharp take on the latest in AI for PMs, founders and operators. Every few days. Free.
Free forever. Unsubscribe anytime. No spam.
- Learning from the web — ingesting public text to build a model's general capability. Treated as permitted.
- Reselling the web — reproducing a specific publisher's article, in substance, to satisfy a user so they never visit the source. This is where the right may bite.
That second behavior is exactly what makes AI answers commercially threatening to publishers, and it's precisely the behavior now under legal scrutiny. It also maps onto a business model under pressure elsewhere — the EU Commission has separately questioned publishers about Google's AI opt-out, and the European Publishers Council filed an antitrust complaint over AI Overviews and AI Mode.
What this means if you do GEO
We've argued that the SEO funnel moved inside the model and that the game is now being the source the model cites. This opinion adds a legal dimension to that same shift:
- If you publish content, publisher rights and opt-out tools are gaining legal teeth. That strengthens your hand — the model that strips your article and gives nothing back is the exact behavior regulators are circling. Structure content to be cited and linked, and know your rights are becoming more enforceable, not less.
- If you build AI answers, the safe design is the one that points to sources rather than substituting for them — real attribution, real links out, and licensing where you're reproducing substance. An answer engine that cannibalizes the source is now carrying legal risk, not just reputational risk.
The deeper point: the AI-answer products that survive won't be the ones that absorb the web most completely. They'll be the ones that found a sustainable relationship with the people who make the content — through attribution, traffic, or licensing — before the courts forced one.
The takeaway
A single non-binding opinion won't rewrite AI overnight. But it drew the line the whole industry will now argue over: learn from the web, yes; resell it at answer time, maybe not. For content teams, it's a reason to lean harder into original, citable work and to treat your publishing rights as an asset. For builders of AI answers, it's an early warning to design for attribution now — because "the model just summarized it" is starting to sound, in a European courtroom, a lot like "the model reproduced it."
Sources
- Courthouse News: EU legal adviser finds limited uses of Google AI could run afoul of publishers' rights
- Silicon UK: EU adviser on Google AI and press publishers' rights
- The Decoder: EU publishers claim Google's AI Overviews reduce their traffic and revenue
- FourWeekMBA: EU adviser says Gemini may breach press rights in first GenAI case
Want more sharp takes on AI and search? Browse the AI Decoded archive.