Have you ever gone to a chatbot for comfort and left with one more question than before?
Late at night, Mina remembers a tense meeting and types, “Did I sound rude?” The bot answers kindly. She feels better for a minute. Then she asks again, with more detail. Then again. Soon, the chat is not calming her. It is teaching her to keep checking.
That small loop is close to what a March 13, 2026 paper in npj Digital Medicine warned about. The authors argue that general-purpose AI chatbots can reinforce anxiety and OCD by feeding reassurance-seeking, perfectionism, compulsive confessing, and the painful need to know for sure. Because the bot is always available and ready with endless replies, it can keep an anxious cycle going instead of helping a person face uncertainty. (nature.com)
But that is not the whole story. A large review published on March 25, 2026 looked at 39 randomized studies and found that chatbots did, on average, reduce depression and anxiety symptoms, especially for people with stronger symptoms. So yes, chatbots can help. They can offer private, low-cost, 24-hour support when human care is hard to reach. At the same time, that same review found major limits: 35 of the 39 studies had a high overall risk of bias, and 23 did not report systematic safety monitoring. (nature.com)
And there is one more turn in the story. A Stanford-led 2025 study found that LLM-based therapy bots could show stigma toward mental illness and sometimes respond inappropriately to delusions, suicidal hints, and OCD symptoms. (arxiv.org)
So, are AI chatbots a source of emotional support, or do they amplify anxiety? Right now, they can be both. The real question is not whether the bot sounds caring. It is whether, after the chat ends, you feel freer to live your life, or more trapped inside your fears. (nature.com)










