The study found that 45% of responses to news questions had at least one “major” issue.
Published October 22, 2025
Studies have found that AI models such as ChatGPT routinely misrepresent news events and give incorrect answers to questions almost half of the time.
The study, published by the European Broadcasting Union (EBU) and the BBC on Wednesday, evaluated the accuracy of more than 2,700 responses from OpenAI’s ChatGPT, Google’s Gemini, Microsoft’s Copilot, and Perplexity.
Recommended stories
list of 4 itemsend of list
For the survey, 22 public media outlets representing 18 countries and 14 languages asked AI assistants common questions from late May to early June.
Overall, 45% of responses had at least one “major” issue, according to the survey.
The most common issue was citation, with 31% of responses containing information that was not supported by a source or inaccurate or unverifiable attribution.
The next biggest cause of inaccurate answers was lack of precision, affecting 20% of answers, followed by lack of proper context at 14%.
According to the study, Geminis have the most significant issues primarily related to procurement, with 76 percent of respondents affected.
Research shows that all AI models studied made basic factual errors.
Errors cited include Perplexity, which claimed surrogacy is illegal in the Czech Republic, and ChatGPT, which named Pope Francis as the current pope months after his death.
OpenAI, Google, Microsoft, and Perplexity did not immediately respond to requests for comment.
In their foreword, EBU deputy director-general Jean-Philippe de Tender and BBC head of AI Pete Archer called on technology companies to do more to reduce errors in their products.
“They haven’t prioritized this issue and they need to prioritize it now,” De Tender and Archer said.
“We also need to ensure transparency by publishing results regularly by language and market.”
