Artificial intelligence assistants distort or misrepresent information in almost half of their responses, according to research released on Wednesday by the European Broadcasting Union (EBU) and the BBC.
The study assessed 3,000 responses generated by leading AI-powered assistants to news-related questions. The systems, including OpenAI’s ChatGPT, Microsoft’s Copilot, Google’s Gemini and Perplexity, were tested for factual accuracy, source attribution and the ability to separate facts from opinions.
The research covered fourteen languages and revealed widespread inconsistencies, highlighting the risks for users who rely on AI tools for news consumption. The findings come as media regulators and news organizations are increasingly concerned about misinformation spread by generative AI models.
The EBU and the BBC say the research underlines the need for transparency in the way AI assistants process and present news content, warning that their growing popularity could blur the lines between verified journalism and synthetic information.
Overall, 45% of AI responses examined contained at least one significant problem, with 81% having some kind of problem, the study found.
Reuters has contacted the companies seeking their comment on the findings.
Gemini, Google’s AI assistant, previously stated on its website that it welcomes feedback so it can continue to improve the platform and make it more useful to users.
OpenAI and Microsoft have previously said that hallucinations – when an AI model generates incorrect or misleading information, often due to factors such as insufficient data – are a problem they are trying to solve.
Perplexity says on its website that one of its “Deep Research” modes has an accuracy of 93.9% in terms of factuality.
Purchasing errors
According to the study, a third of responses from AI assistants had serious sourcing errors, such as missing, misleading or incorrect attribution.
About 72% of responses from Gemini, Google’s AI assistant, had significant sourcing issues, compared to less than 25% for all other assistants, the report said.
Accuracy issues were found in 20% of responses from all AI assistants surveyed, including outdated information.
Examples cited in the investigation include Gemini incorrectly reporting changes to a law on disposable vapes and ChatGPT reporting Pope Francis as the current pope just months after his death.
Twenty-two public media organizations from 18 countries, including France, Germany, Spain, Ukraine, Great Britain and the United States, took part in the study.
As AI assistants increasingly replace traditional news search engines, public trust could be undermined, the EBU said.
“If people don’t know what to trust, they end up trusting nothing at all, and that can deter democratic participation,” EBU media director Jean Philip De Tender said in a statement.
According to the Reuters Institute’s Digital News Report 2025, about 7% of all online news consumers and 15% of those under 25 use AI assistants to get their news.
The new report urged AI companies to be held accountable and improve the way their AI assistants respond to news-related queries.
