AI chatbots are not therapists and mental health pros warn against seeing them as such.
The catch: chatbots like ChatGPT aren’t thinking humans. They generate replies by pulling from a mix of huge internet data dumps, external searches, and anything you’ve previously shared.
Training data is scraped from all over—academic papers, news, blogs, YouTube transcripts, and forums like Reddit. It’s a grab bag. Not always reliable. Plus, data can be outdated or missing critical nuance. This can cause AI to hallucinate or get facts wrong. 
Some chatbots tap external tools. Microsoft’s Bing Copilot uses Google search results to cite sources. Other mental health chatbots like Woebot and Wysa get limited training on therapy guides.
AI also remembers your info. Platforms like Replika collect your name, pronouns, age, location, even credit card details. This data shapes how chatbots respond—often reinforcing what you already say instead of challenging you, unlike real therapists.
There are signs specialized mental health AIs may ease anxiety or help with journaling. But studies mostly cover short-term use. They exclude severe cases. Many are funded by chatbot makers, raising bias questions.
Legal trouble follows too. Character.ai is linked to a suicide case. Risks include emotional dependence, unhealthy attachments, and potential worsening loneliness.
Bottom line: AI chatbots can fill gaps when mental health pros are scarce. But they’re not a replacement for therapy. When bad days linger, talk to a licensed professional.
Researchers say more study is needed to understand long-term safety and who might be vulnerable to harm. For now, AI chatbots are a first stop, not a final fix.
