We investigated the intervention effect of an AI robot based on cognitive behavioral therapy
(CBT) and it’s specificity for different depression severity. Methods: A total of 160 participants who
scored between 6 and 19 in Patient Health Questionnaire-9 (PHQ-9) were randomly assigned to the
AI intervention group or the control group, with 71 and 53 subjects were included in the final analysis,
respectively. The AI intervention group completed 7-week chatbot-delivered CBT including weekly
sessions with different themes and daily exercises through the PSYBOT Application. The control group
(waitlist) were required to give their mood ratings and read mental health science articles daily through
the same platform for 7 weeks. All outcomes were measured through the PHQ-9 at baseline, and at the
end of the third, fifth and seventh week of the intervention. Results: Participants in the AI group showed
improvements in depressive symptoms compared with the control group after the 7-week intervention,
as indicated by decreased scores in the PHQ-9 (p<0.05, d=-0.44). At the end of the 3rd week of the chat
bot-delivered CBT intervention, significant effects were found among participants with moderate and
moderately severe depression (p<0.001, d>2) but not in mildly depressed subjects. The effects remained
constant and were further in the 5th and the 7th week of intervention. Conclusions: Chatbot-delivered
CBT may serve as a possible modality in alleviating the depressive symptoms of people with moderate
and moderately severe depression. Multi-dimensional dependent variable indicators need to be used for
further verification.