Therabot Study Flawed and Biased

Getting your Trinity Audio player ready...

To the editor,

As a psychologist and Chair of Psychotherapy Action Network (PsiAN), I’m writing about your recent story on Therabot, a generative AI therapy chatbot, and a newly published RCT that assesses its performance (Meet Therabot, the first AI-powered therapy chatbot to prove its worth in a controlled trial, published 4/1/25). Your report was glowing, as were effusive headlines in other publications discussing Therabot and this “landmark study.”

However, a quick look at the methodology and inherent bias in the study give ample cause for concern. First, the individual who developed Therabot is the one who conducted the research study – a clear sign of researcher bias. Second, regarding research design, the study compared one group of individuals using Therabot, for only 4 weeks, with an optional additional 4 weeks, against a group that did nothing (waitlist control). Doing something, especially talking about one’s problems, usually helps people feel better, even if they’re talking to a friend, a pet, or their journal.

In addition, while it’s typical for efficacy research studies to measure symptom levels at the end of the study, it’s a far cry from what people say they want from therapy. People want not only symptom relief, but they want to achieve important goals such as getting to the root of their issues, understanding themselves, learning how to make different choices in their lives, improving their relationships, and increasing their self esteem. These are hard-won gains that come over time, and in collaboration with a trusted therapist. These goals and expectations have been documented in PsiAN’s two nationwide quantitative surveys.

I have no doubt that AI therapy models will continue to improve. Access to effective care needs to increase. The demand for professional therapy continues to outpace the supply of therapists, particularly those who participate in-network with insurance plans. With below-market reimbursement rates and excessive administrative burdens, insurers have driven therapists out-of-network and thus contributed to a “therapist shortage,” largely of their own making. Vast insurance reform is needed, and technology solutions may be helpful. What isn’t helpful is the dissemination of methodologically weak studies infused with bias testing variables that hold minimal meaning for people who are struggling and suffering in their lives.

Linda Michaels, PsyD MBA
Chicago, IL