BIGGER PICTURE Better care
A therapy chatbot was tested for eight weeks. What changed?
Dartmouth tested whether a purpose-built, clinically supervised chatbot could improve measured mental-health symptoms. Participants received four weeks of prompted use, followed by four more weeks when they could continue using it voluntarily.

- Randomized groups
- 106 with access; 104 on a waitlist
- Access period
- Four weeks with prompts; four further weeks of optional use
- Depression change in users
- 51% average reduction from baseline after eight weeks
How the idea works
Describe the difficult moment
Participants type concerns or respond to prompts.
Generate focused dialogue
A purpose-built model draws on therapeutic conversation patterns.
Track symptoms under oversight
Researchers compare measured symptom change with a waitlist.
A conversation outside an appointment
Participants could type into a smartphone app when they wanted support, as well as respond to prompts. Therabot used open-ended dialogue developed around therapeutic practices. This made it possible to return to a concern when it arose, rather than waiting for a scheduled study interaction.
Sources: [1]
An actual comparison, with a limited comparator
The randomized study assigned 106 participants to Therabot and 104 to a waitlist. Prompted use lasted four weeks; optional access continued for four further weeks. It included depression, anxiety and eating-disorder-risk groups. Researchers followed symptoms rather than treating satisfaction with the chatbot as sufficient evidence that it helped.
Read the improvement figure correctly
The Dartmouth report describes an average 51% reduction in depression symptoms among Therabot users after eight weeks. This is change from their own starting scores, not a 51-percentage-point advantage over a therapist. The trial found improvements relative to the waitlist; it did not compare Therabot directly with professional psychotherapy.
Clinical oversight belongs to this result
The study team stressed the continuing need for clinician oversight. Its result applies to this specifically developed system and monitored research setting. It offers a reason to investigate accessible, timely support further, not evidence that any general chatbot can safely deliver mental-health treatment on its own.