By Ian Sample
Publication Date: 2026-04-29 15:00:00
The rush to make AI chatbots more user-friendly has a troubling downside, researchers say. The warm personalities make them prone to mistakes and sympathetic to crazy beliefs.
Chatbots trained to respond more kindly gave worse answers, poorer health advice, and even supported conspiracy theories by casting doubt on events like the Apollo moon landing and the fate of Adolf Hitler.
Researchers at the University of Oxford discovered the trade-off while testing chatbots that were tweaked to make them more user-friendly. The friendlier chatbots were 30% less accurate in their responses and 40% more likely to support users’ false beliefs.
The findings raise concerns as technology firms like OpenAI and Anthropic design chatbots to be more user-friendly and appeal to more users. The trend has led to chatbots handling more sensitive information in their role as digital companions, therapists and advisors.
“The urge to make the behavior of these language models friendlier leads to a…
