AI advice followed by 79%, but well-being unchanged

The study addresses a growing trend of people using AI chatbots for personal advice on health, career, and relationships. Prior to this research, it was unclear whether people actually follow such advice and whether it benefits their mental health. The findings indicate a high compliance rate—74.6–79.3%—even for risky advice, which over 65% followed. This suggests a strong tendency to trust AI guidance regardless of stakes.

However, the lack of sustained well-being improvement raises questions about the value of AI advice for personal issues. The immediate drop in happiness after the chat, though temporary, hints at potential emotional costs. The observation that following any advice—including hobby-related—correlates with happiness suggests that the act of following advice, rather than the advice itself, might have effects.

This study does not prove harm, but it underscores that AI advice may not deliver the emotional benefits users might expect. It could imply that AI should be used cautiously for personal matters, or that such advice requires more careful design to actually improve well-being. The authors are from UK AI Security Institute and Limbic AI, and the paper is on arXiv.