JMIR Publications

ABSTRACT

Background:

Baseline data from our survey of 527 self-referred users of the mental health chat- and voice bot Clare® indicate high psychological distress and barriers to accessing face-to-face care strong working alliance was established within 3–5 days (Working Alliance Inventory-Short Report, M = 3.76, SD = .72). The feasibility of sustained engagement and therapeutic bonding with Clare® in real-world use remains underexplored.

Objective:

This exploratory feasibility study evaluated engagement patterns, sustained therapeutic bonding, and preliminary mental health outcomes during 4- and 8-week use of the LLM-enabled voice and text chatbot Clare®.

Methods:

A single-group pre-post feasibility study was conducted with the English-speaking general population that self-referred and interacted with the voice- and chatbot Clare® for 4 weeks (n=53) or 8 weeks (n=21). Usage patterns, modes of engagement (hybrid, text-only, call-only), message volume, and call duration were examined. Users were further assessed for working alliance and changes in loneliness, depression, anxiety, distress, and life satisfaction at baseline, week 4 (t1), and week 8 (t2).

Results:

A total of 53 participants (73.6% women) engaged with the Clare® over 4 weeks (sample 1) and 21 participants (71.4% women) completed all assessments (sample 2), with both samples showing comparable demographic profiles. At baseline, participants reported moderate depression and anxiety, elevated social anxiety and high loneliness. Initial engagement peaked in the first week, with participants initiating an average of 1.77 calls and sending 10.02 messages, before declining steadily. On average participants initiated an average of 3.34 calls, sent 23.65 messages, and spent a total of 8.07 minutes in voice calls in 4 weeks. A comparable pattern was observed in the 8-week completer sample (n = 21). Working alliance increased over time, rising from 2.91 (SD 0.88) at baseline to 3.21 (SD 1.17) at mid-assessment and 3.34 (SD 1.03) at post-assessment (t3). There was no significant association with engagement intensity. Higher baseline distress was associated with fewer messages and lower alliance. Higher depressive symptoms were linked to fewer early calls and lower overall call frequency. Modest improvements were observed across loneliness, depression, anxiety, and distress. This was a feasibility study with substantial attrition and results should be interpreted with caution.

Conclusions:

Clare® appears feasible and acceptable for short-term community use, with early and stable bonding and preliminary signals of emotional and mental-health improvement. Declining engagement over time and the weak association between communication volume and alliance highlight the need for technology improvement and individualized symptom-oriented design strategies that support sustained and meaningful interaction.