The study showed that 45% of the responses contained at least one serious error, 31% had shortcomings in source attribution, and 20% exhibited low accuracy. The worst results were shown by the Gemini model, where serious errors were identified in 76% of cases.
According to the Reuters Institute's Digital News Report 2025, 7% of internet users receive news through AI assistants, and among young people under 25, this figure is 15%.
Jean-Philippe de Tender, Director of Media at the European Broadcasting Union, noted: "Our research clearly demonstrates that these problems are not isolated. They are systemic, transnational, and multilingual, which jeopardizes public trust in the media."
EBU and its members urged EU regulators to ensure compliance with existing laws on information integrity and emphasized the need for independent monitoring of AI assistants' performance.
Additionally, the research group developed specialized tools to identify and address issues related to news distortion by artificial intelligence.
