New research coordinated by the European Broadcasting Union (EBU) and led by the BBC found that AI assistants misrepresent news content across multiple languages and platforms. The study, involving 22 public service media organizations in 18 countries, identified systemic inaccuracies in AI-generated news responses. Professional journalists evaluated over 3,000 responses from four AI tools: ChatGPT, Copilot, Gemini, and Perplexity.
Key findings revealed that 45% of AI answers contained at least one significant issue. Among these, 31% exhibited serious sourcing problems, such as missing or misleading attributions. Additionally, 20% of responses had major accuracy issues, including hallucinated details and outdated information.
Gemini was identified as the least reliable tool, with significant issues in 76% of its responses due to poor sourcing. Comparisons with earlier BBC studies indicate some improvements but show persistent, high levels of error in AI-generated news.
The Reuters Institute’s Digital News Report 2025 states that 7% of online news consumers utilize AI assistants for news, increasing to 15% among those under 25. This trend raises concerns about public trust, as EBU Media Director Jean Philip De Tender noted that these systemic failings could lead to a decline in democratic participation if users lack trust in news sources.
Peter Archer, BBC Programme Director, emphasized the necessity for accurate news content, acknowledging the potential benefits of AI while recognizing existing issues. The research team also introduced a “News Integrity in AI Assistants Toolkit” aimed at improving response quality and enhancing media literacy.
This toolkit addresses critical questions regarding effective AI responses and outlines the issues needing resolution. Additionally, the EBU and its members are advocating for the enforcement of existing regulations related to information integrity and media pluralism.
The study builds on previous research from February 2025, confirming that the challenges faced by AI assistants are not confined to specific languages or markets. Participating broadcasters represented countries including Belgium, Canada, France, Germany, Italy, and the United Kingdom, among others.
Separate BBC research showed a significant portion of the public, particularly younger demographics, tends to trust AI for accurate news summaries. Over a third of UK adults expressed this trust, with nearly half among those under 35. These findings present a dilemma, as inaccuracies in AI-generated news may lead users to misattribute errors to traditional news providers, potentially damaging their trust in credible news sources.
