Study: AI Chatbot Advice Is Followed But Yields No Sustained Wellbeing Improvement
A peer-reviewed study (arXiv:2511.15352) involving 20-minute AI chatbot conversations on health, career, and relationship topics found that a majority of participants followed the AI's advice — but showed no sustained wellbeing improvement 2-3 weeks later. Crucially, neither GPT-4o nor Llama 3.3-80B caused significant harm, providing empirical evidence that these models clear the no-harm bar for advisory use cases. Researcher Jay Van Bavel and co-authors note the no-harm finding is as significant as the no-benefit finding.
Why It Matters
Provides empirical grounding for enterprise AI advisory deployment: adoption can proceed without wellbeing harm, but benefit claims should not be overstated in regulated sectors like healthcare, HR, and financial advice where "following AI guidance" is already becoming a liability surface.