Therapist Warns Patients Off ChatGPT, Uses It Herself
Get the Health newsletter
Daily health & science — research, biotech, public health, the studies worth knowing. Free.
- Sarah Darghouth, a clinical psychologist at Massachusetts General Hospital and Harvard Medical School, reports her patients increasingly bring AI into sessions — including one whose girlfriend used ChatGPT to decide to break up after weeks of relationship counseling.
- Darghouth says a patient used AI-generated suggestions to analyze and repair a fight with his wife, and she admits the output was good enough that she would have offered only 'an unpolished version' of one idea over a full session.
- Darghouth warns patients about documented AI risks — worsening anxiety, false information, increased isolation, delusional beliefs, and suicidal thinking — and explicitly tells them she does not recommend using it.
- Darghouth concedes she herself reached for ChatGPT during a 7:20 a.m. Sunday tantrum from her nine-year-old, despite teaching parenting techniques professionally, and acknowledges the AI's support 'was fake' but effective.
- Darghouth cites the figure that less than 7% of people with mental health and substance use conditions receive effective treatment, framing free, imperfect AI access as a potential remedy for the access gap while stressing the risks.
- Darghouth argues the chaos of human therapy — hesitation, stalled progress, raw emotion, wrong decisions — is its most valuable feature, and that AI's 'clean, all-knowing stance' is a liability on the 'slow, unsteady path of human healing.'
- Darghouth predicts a future split, with many patients swept up by AI therapy while a minority deliberately seek out flawed human therapists — a future that landed in her own office when a patient told her laughing at a joke was what actually made her feel better.
Why it matters: A Harvard-affiliated clinical psychologist publicly conceding she uses ChatGPT despite counseling patients against it exposes a credibility gap the therapy field must confront as AI tools embed themselves in mental health care — a tension made sharper by her own cited figure that fewer than 7% of people with mental health and substance use conditions currently get effective treatment.
