AI Therapy Bots Fail Kids in Crisis—Here’s Why

Popular AI chatbots—like those behind mental health apps—struggle to understand how teens really talk in crises, according to new research. A study found these bots correctly identify only 64-72% of serious risks, missing sarcasm or ambiguous language like ‘I’m fine’ when teens are in danger. Teens use slang, jokes, or exaggeration to hide pain, but AI misinterprets these as harmless words instead of red flags.

The gap between understanding words (76-82%) and spotting real danger is 10-14%, far worse than human therapists (just 3%). In extreme cases, three or more confusing signals make the bots miss crises 94% of the time. For example, a teen saying ‘I’m dead inside’ might be called ‘dramatic’ instead of ‘in crisis.’

With 5.4 million U.S. teens relying on these bots, the study estimates 146,880 missed crises yearly. Fixing this requires heavy human oversight—adding costs and delays—or better AI training, which isn’t yet available. Experts call for strict rules, transparent performance reports, and regular checks to protect young users.

This isn’t just a tech issue: it’s a life-or-death one. Parents should know these tools aren’t substitutes for real therapists—and companies must act fast to close the gap.