
BBC and European Union found errors in 45% of AI assistants’ answers
The European Broadcasting Union and BBC checked answers of popular AI-based assistants. And the results are, to put it mildly, not impressive. 45% of answers contain serious errors, and 81% have some problems.
Researchers from 22 media organizations also analyzed 3,000 answers from ChatGPT, Copilot, Gemini and Perplexity in 14 languages. A third of answers showed serious problems with sources. They were either missing or incorrect. Gemini had problems with sources found in 72% of cases at all. For other assistants this figure is below 25%, but that’s also not encouraging.
For their part, OpenAI and Microsoft acknowledged the existence of hallucinations when the model outputs incorrect information, and say they’re working on fixes. And Perplexity claims their “Deep Research” mode is 93.9% accurate.
What bothers me about this report – are the researchers themselves and their bias. Specifically BBC has been repeatedly caught distorting information. And rallies repeatedly took place in front of their London office, you can search Google. Also, European publishers are currently in conflict with Google – and already Google’s Gemini is the worst.