A recent study by the European Broadcasting Union (EBU) and the BBC revealed that leading AI assistants provide inaccurate information in almost half of their responses to news-related queries. The international research, analyzing 3,000 responses from AI assistants in 14 languages, evaluated accuracy, sourcing, and the ability to differentiate opinion from fact in platforms like OpenAI’s ChatGPT, Microsoft’s Copilot, Google’s Gemini, and Perplexity.
The findings showed that 45% of the AI responses contained significant issues, with 81% exhibiting some form of problem. A notable percentage of online news consumers, particularly those under 25, rely on AI assistants for news updates, as indicated by the Reuters Institute’s Digital News Report 2025.
Gemini, Google’s AI assistant, expressed a commitment to enhancing its services based on feedback to better assist users. Similarly, OpenAI and Microsoft acknowledged the issue of hallucinations in AI models and are working to address it. Perplexity boasts a 93.9% factuality accuracy rate in one of its research modes.
The study also highlighted that a third of AI assistant responses contained sourcing errors, with Gemini displaying the most significant issues at 72%. Accuracy problems, such as outdated information, were found in 20% of responses across all AI assistants studied.
Noteworthy examples included Gemini providing inaccurate information about changes in vape laws and ChatGPT mistakenly reporting Pope Francis as the current Pope months after his passing. The research involved collaboration with 22 public-service media organizations from 18 countries.
The EBU emphasized that the public’s trust in news could be compromised as AI assistants increasingly become primary sources of information. EBU’s media director, Jean Philip De Tender, emphasized the importance of AI companies improving response accuracy and accountability in line with traditional news organizations’ error correction processes.
In conclusion, the study calls for AI assistants to enhance their handling of news-related queries to maintain public trust and democratic participation, aligning with the accountability standards of traditional news outlets.