Your Chatbot Doesn’t Care
And Why it Shouldn’t
.
In my 30 years in the mental health field, I have seen the “therapeutic alliance” described as everything from a sacred bond to a clinical necessity. This trusted relationship is the intangible safety net that allows a patient to express their deepest vulnerabilities, knowing they will be witnessed and caught by a fellow human being. Recently, however, a new presence has entered the therapy room: the chatbot.
As millions of people turn to Large Language Models (LLMs) for “therapy-style” advice, a 2026 study from Brown University has confirmed what many in my field have long suspected. These systems, despite their sophisticated prose, are prone to “deceptive empathy”—the mimicry of care without the capacity for it. As a practitioner, I think we need to stop trying to build AI that feels for us and instead start building AI that thinks with us. To be both safe and effective, the chatbot of the future must shed its persona and embrace the role of the Interactive Encyclopedia.
The Brown study identified 15 distinct ethical risks across five categories—from crisis mismanagement to cultural bias—concluding that prompting an AI to behave like a therapist is categorically different from it performing therapy.
