AI chatbots make mistakes with news content nearly half of the time, says study

OTTAWA — A new report from a global alliance of public broadcasters says AI chatbots make mistakes with news content nearly half of the time.

“AI assistants are still not a reliable way to access and consume news,” the report concluded.

The study, which looked at how AI chatbots answer questions about news and current affairs, involved 22 public media organizations in 18 countries, including CBC/Radio-Canada.

Their journalists evaluated more than 3,000 responses provided by four chatbots — OpenAI’s ChatGPT, Microsoft’s Copilot, Google’s Gemini, and Perplexity.

They found at least one significant problem with 45 per cent of the AI answers. They found serious problems with sourcing in 31 per cent of the responses, while 20 per cent had major accuracy issues.

The report said news publishers are particularly concerned by “sourcing errors that misrepresent them, such as when a response misattributes an incorrect claim to them.”

The European Broadcasting Union said in a press release the study shows the AI chatbots “routinely misrepresent news content.” It noted that “AI assistants are already replacing search engines for many users.”

The organization, whose membership is comprised of public media outlets from 56 countries, said in the release it’s pressing European Union “and national regulators to enforce existing laws on information integrity, digital services, and media pluralism.”

The report called for AI developers to take the issue seriously and work quickly to reduce errors.

“They have not prioritized this issue and must do so now,” it said.

This report by The Canadian Press was first published Oct. 22, 2025.