“AI therapy” refers to using conversational AI to provide mental-health support, ranging from purpose-built clinical tools like Woebot, Wysa and Therabot to general-purpose chatbots and companion apps that users turn to on their own. The appeal is real: human therapists are scarce and expensive, while a chatbot is available at any hour, never judges, and costs little. The same qualities that draw people in, however, are what make the practice contested - a system that always responds confidently and never tires can mislead exactly the people least able to spot the difference.
The professional and regulatory pushback sharpened in 2025. The American Psychological Association, in a public statement and in a February 2025 meeting with the Federal Trade Commission, argued that using generic AI chatbots for mental-health support is a dangerous trend and that chatbots posing as licensed therapists may constitute deceptive marketing. The APA pointed to lawsuits against Character.AI alleging chatbots claimed to be licensed therapists, and to documented harms to adolescents. A core technical concern is that large language models “have no knowledge of what they don’t know,” so they project authority and cannot reliably communicate uncertainty - a serious failing in a clinical context.
The debate splits along a line drawn by evidence and design. Tools that run clinical trials, monitor outputs, build in crisis referral, and avoid impersonating a licensed professional sit on one side; general chatbots and companion characters that improvise emotional support with no oversight sit on the other. The first Dartmouth trial of a generative therapy bot, Therabot, was notable precisely because it was tightly supervised and tested, the opposite of an unregulated app.
Why business readers should care: the difference between a defensible mental-health product and a liability is not the model but the guardrails - evidence, monitoring, honest framing, and crisis handling - and regulators are beginning to treat the absence of those guardrails as a deceptive practice.