The BBC and the European Broadcasting Union (EBU) released fresh data on Wednesday that shows that over half of the replies from leading AI assistants misrepresented news material.
3,000 answers to news-related queries from top AI assistants—software programs that employ AI to comprehend natural language instructions and carry out tasks for a user—were examined in an international study.
ChatGPT, Copilot, Gemini, and Perplexity were among the 14 languages in which it assessed AI assistants for accuracy, sourcing, and the capacity to discriminate between opinion and reality.
Overall, the study found that 81% of the AI responses had some kind of flaw, and 45% had at least one major issue.
As previously said on its website, Google’s AI assistant, Gemini, invites user feedback in order to keep improving the platform and making it more beneficial to consumers.
As previously stated, OpenAI and Microsoft are working to address the problem of hallucinations, which occur when an AI model produces inaccurate or deceptive information, frequently as a result of variables like inadequate data.
One of Perplexity’s “Deep Research” modes offers 93.9% factual accuracy, according to the company’s website.
The survey found that a third of AI assistants’ answers included significant source mistakes, including missing, inaccurate, or deceptive attribution.
Compared to less than 25% for all other assistants, Gemini, Google’s AI assistant, reported that 72% of its replies had serious source problems.
20% of replies from all AI assistants under study had accuracy problems, including out-of-date information, the report stated.
The study included the examples of Gemini inaccurate changes to a regulation pertaining to disposable vapes and ChatGPT identifying Pope Francis as the current pope months after his passing.
Participating in the research were twenty-two public-service media organizations from eighteen nations, including France, Germany, Spain, Ukraine, Britain, and the United States.
According to the EBU, public confidence may be weakened if AI assistants progressively replace traditional search engines for news.
EBU Media Director Jean Philip De Tender stated in a statement that when people are unsure about what to believe, they end up believing nothing at all, which might discourage them from participating in democracy.
According to the Reuters Institute’s Digital News Report 2025, 15% of those under 25 and 7% of all online news consumers use AI assistants to obtain their news.
The new research called on AI businesses to be held accountable and to improve how their AI assistants answer to news-related questions.
