Understanding user preferences for gaining trust, when utilising conversational agents for mental health data disclosures

Encouraging humans to disclose personal information is a complex process that is built upon trust, and this is especially true when related to sensitive topics such as mental health. Currently, this data is collected through trained professionals but COVID-19 has seen an increasing demand for support. This paper looks at maximising trust in mental health conversational agents. The study collected data from 177 participants, using survey questionnaires, to examine what human-like features help cultivate and encourage trust. Analysis suggests respondents prefer something that reflects themselves. For example, 78% stated a conversational agent should display a static avatar they can shape to their own preferences. Other factors found to have an impact were friendly greetings (preferred by 76%) and patience (99%). This initial study establishes that humans believe mental health conversational agents can, and should, exhibit a range of human-like features. Some preferences are largely universal across all demographics, whereas others are more specific. This study then delivers a framework of desirable attributes, traits and characteristics, which will be used to test if these features are more successful at establishing trust than standard online forms.