BBC Study Finds AI Chatbots Frequently Misrepresent News Content

Over half of AI-generated news summaries were found to contain significant inaccuracies, raising concerns about trust and misinformation.

The BBC tested ChatGPT, Copilot, Gemini, and Perplexity on 100 questions about its news content, with journalists reviewing their accuracy.
51% of AI-generated responses had significant issues, including factual errors in 19% of cases and altered or fabricated quotes in 13%.
Examples of errors included outdated information, misrepresentation of NHS vaping advice, and claims that former politicians were still in office.
The study highlighted AI's struggles with distinguishing fact from opinion, providing context, and avoiding editorialization in its responses.
BBC leadership called for greater collaboration with AI companies and regulatory oversight to ensure accuracy and trustworthiness in AI tools.