Millions of people now ask AI chatbots what is happening in the world, and the answers they receive are often built from reporting that someone else paid to produce. A 2026 Reuters Institute survey of more than 90,000 people worldwide found that 10% use AI chatbots for news each week, a share that is significantly higher among younger audiences. That shift matters because explaining events, why they matter, and what they mean has traditionally been the work of journalism: reporters in the field, investigations, and research that audiences rely on.
The economics of that work are breaking down. Julia Becker, publisher of Funke Media Group, said the old exchange in which search engines indexed articles and sent traffic back in return "has broken down." DW Director General Barbara Massing made a sharper point: AI companies draw on quality reporting to generate answers, yet the reporting itself often remains largely invisible by the end of the process. When the source disappears from the answer and the visitor never arrives, the newsroom loses both visibility and revenue while the tool keeps the value.
Those who build these systems should pay for what they use. Professor Martin Senftleben of the University of Amsterdam makes a useful distinction between training models and generating answers to individual questions. Training is a more complicated legal question. The answer-generation problem is more urgent, and he argues that existing European copyright law may already let publishers and journalists claim payment collectively. His proposal is a collecting system in which every AI response drawing on journalism triggers a payment, distributed in the way music royalties are. Germany, he said, should introduce it without waiting.
Google says it is willing to negotiate with German media companies, and its "AI Contribution" pilot rewards some publishers whose content plays a significant role in answers from Gemini and AI Overviews. But the pilot is an early experiment open only to selected publishers, not a general payment system. Voluntary goodwill and limited pilots are not a substitute for a rule that applies to everyone. Microsoft, Anthropic, and OpenAI have backed clearer standardized signals about whether content may be used for training, which is a useful start but still leaves payment unresolved.
Jörg Heidrich of Heise is right that the rules must not simply wall off European companies or squeeze out startups, independent developers, and research institutions that also depend on data. Still, his warning that Europe has already fallen behind should not become an argument for letting the strongest platforms keep taking reporting for free. A workable regime can distinguish between large commercial answer engines and smaller research uses. Germany and the EU should start there, pay publishers collectively for the answers built on their work, and let the rates they set become the baseline for individual deals.