JMIR Publications

ABSTRACT

Background:

Psychological distress increases across adolescence and has been associated with a number of important health outcomes, with consequences that can extend into adulthood. One type of technological innovation that may serve as a unique intervention for youth experiencing psychological distress is the conversational agent (CA), otherwise known as a chatbot. Further research is needed on mental health chatbots - notably those designed for adolescents. The experimental mental health chatbot used in this study was designed to support adolescents experiencing psychological distress.

Objective:

The current study aimed to assess adolescents’ perceived emotional reactions to questions posed by a mental health chatbot and to evaluate adolescents’ preferences concerning the formulation of the chatbot’s questions.

Methods:

We recruited 21 adolescents aged 14 to 17 to participate in a pilot study with a 2x2x3 within-subjects factorial design. Each participant was sequentially presented with 96 chatbot questions for a duration of eight seconds per question. Following each presentation, participants were asked to indicate how likely they were to respond to the question, as well as their perceived affective reaction to the question. Demographic data and participant feedback were also collected.

Results:

Participants were an average of 15.4 years old (SD 1.05) and mostly female (60%; 12 females; 8 males). Logistic regressions showed that presence of GIFs (Graphics Interchange Format) predicted perceived emotional valence (β = -.40; P < .001), such that questions without GIFs were associated with a lower perceived emotional valence. Question type predicted emotional valence, such that yes/no questions (β = -.23; P = .03) and open-ended questions (β = -.26; P = .01) were associated with a lower perceived emotional valence compared to multiple response choice questions. Question type also predicted likelihood of response, such that yes/no questions were associated with a lower likelihood of response compared to multiple response choice questions (β = -.24; P = .03) and a higher likelihood of response compared to open-ended questions (β = .54; P < .001).

Conclusions:

The findings of this study add to the rapidly growing field of teen-computer interaction and contribute to our understanding of adolescent user experience in their interactions with a mental health chatbot. The insights gained from this study may be of assistance to developers and designers of mental health chatbots.