Generative artificial intelligence and conversational agents (referred to here as chatbots) are increasingly used by the public for mental health support. The safety and role of chatbots when individuals disclose suicide thoughts has become a major public health concern. Prior to this work, the rate of intentions to seek help from chatbots for suicide thoughts and how those intentions relate to other help-seeking sources were not well described.
The study aimed to characterize intentions to seek help from chatbots for suicide thoughts among people with clinically significant depression or anxiety and to compare those intentions with intentions to seek help from informal sources, formal sources, and anonymous online sources.
The sample comprised individuals with clinically significant depression or anxiety (N=58). Participants completed the General Help Seeking Questionnaire, which asks about intentions to seek help for suicide thoughts and for general emotional problems. The research team added two items to assess intentions to seek help specifically from chatbots and from anonymous online sources.
Wilcoxon tests were used to compare intentions to use chatbots with intentions to use anonymous online sources and with grouped categories of informal sources (for example, friends and family) and formal sources (for example, therapist and general practitioner). Kendall's correlations were employed to examine relationships among source groupings and among individual informal and formal sources. Regression models further examined associations for individual sources while adjusting for general help-seeking tendency. Exploratory analyses tested whether demographic characteristics, mental health symptoms, and suicide risk were associated with help-seeking intentions for chatbots.
Participants endorsed lower intentions to seek help for suicide thoughts from chatbots than they endorsed for both informal and formal sources. This indicates that, in this sample, chatbots were a less preferred option for help with suicide thoughts compared with supports such as friends, family, intimate partners, therapists, or general practitioners.
Intentions to use chatbots for suicide thoughts were not correlated with intentions to seek help from informal or formal sources as grouped categories. However, chatbot intentions were correlated with intentions to use anonymous online sources. At the individual source level, after adjustment for general help-seeking tendency, intentions to use chatbots were positively associated with reaching out to intimate partners and negatively associated with outreach to friends.
Among symptom measures examined, anxiety symptom severity was positively correlated with intentions to use chatbots for suicide thoughts. The study reports that this positive correlation was specific to chatbot use intentions and was not observed for other help-seeking sources. The source document indicates exploratory analyses considered demographic characteristics and suicide risk, but does not report detailed demographic associations beyond what is summarized here.
The authors conclude that, in this preliminary sample of adults with clinically significant depression or anxiety, intentions to use chatbots for suicide thoughts are generally lower than intentions to seek help from informal or formal supports and appear mostly disconnected from intentions to use those other supports. Chatbot intentions align more closely with intentions to use anonymous online sources and are associated with higher anxiety symptom severity. The authors recommend that future studies evaluate help-seeking dynamics for suicide thoughts involving chatbots, including whether chatbots are used alongside traditional supports or potentially in place of them. They emphasize these findings are preliminary and the work is a preprint that has not been peer reviewed.
The UC San Diego Institutional Review Board reviewed the study. Participant data are not publicly available to protect confidentiality. The work reported funding support from the National Institute of Mental Health (grant numbers reported in the source). The preprint notes that it has not been certified by peer review and should not be used to guide clinical practice.