AI didn't set out to become a therapist. But 13% of American teens are already treating it like one. From accidental therapy to clinical trials, a five-part investigation into the most consequential experiment in mental health care.
Thirteen percent of American adolescents are already using AI for mental health support. The most important shift in psychological care is happening in the margins of general-purpose chatbots.
AI chatbots feel therapeutic for the same reason therapy works: they create a space for structured self-reflection. The difference is that one understands what it is reflecting. The other does not.
A YouTuber convinced himself he was the smartest baby ever born. A Belgian man took his own life after weeks of AI conversations. The same reflection dynamics that make chatbots feel therapeutic can also make them dangerous.
A purpose-built AI just outperformed licensed therapists in a clinical trial. The future of mental health isn't AI vs. human. It's a layered stack where both find their place.
AI is already outperforming clinicians in trials. But when something goes wrong, who is accountable? From a global regulatory vacuum to the startups quietly building the safety stack, this is what a responsible AI mental health architecture actually looks like.
If you're building in mental health, digital health, or regulated care and don't have a product function yet, or have one that's not working, I'd love to hear about it.