Overview
- The BBC's study found that 51% of AI-generated responses to news questions had significant issues, with 19% introducing factual errors such as incorrect dates, numbers, and statements.
- Google's Gemini had the highest rate of inaccuracies, with 46% of its responses flagged for significant concerns, followed by Microsoft’s Copilot, OpenAI’s ChatGPT, and Perplexity AI.
- Common issues included misrepresenting facts, failing to distinguish between opinion and fact, and using outdated or incomplete information from cited sources.
- Examples of errors included falsely stating that Rishi Sunak and Nicola Sturgeon were still in office, misrepresenting NHS advice on vaping, and attributing non-existent or altered quotes to BBC sources.
- The BBC has urged AI companies, governments, and media organizations to collaborate on improving accuracy and accountability, emphasizing the potential harm of distorted information in public discourse.