Real-World Use of a Mental Health AI Companion: Multiple Methods Study.

Overall attitudes toward AI among Headspace members remained neutral despite majority use of general AI tools.

• Cross-sectional survey included n=482 Headspace members assessing demographics, AI use, and the Artificial Intelligence Attitude Scale-4 (AIAS-4)
• AIAS-4 mean score was 5.7 (SD 2.2, range 1-10), indicating neutral overall attitudes toward AI
• The majority of survey respondents used and would continue to use general AI tools despite neutral attitudes

CAI tool 2.0 showed substantially higher 7-day retention compared to CAI tool 1.0.

• CAI tool 2.0: 77,894/153,249 (50.8%) completed 2 sessions within 7 days
• CAI tool 1.0: 68,701/240,720 (28.5%) completed 2 sessions within 7 days
• CAI tool 2.0 featured enhanced LLM conversational prompts, comprehensive memory, woven content recommendations, and more robust safety detection compared to 1.0

CAI tool 2.0 received higher positive conversation ratings than CAI tool 1.0.

• CAI tool 2.0: 37,819/40,449 (93.5%) positive conversation ratings
• CAI tool 1.0: 94,308/104,323 (90.4%) positive conversation ratings

Retained CAI tool 2.0 users showed greater session engagement compared to all CAI tool 2.0 users and CAI tool 1.0 users.

• Retained CAI tool 2.0 users averaged 6.1 sessions per user
• All CAI tool 2.0 users averaged 2.9 sessions per user
• CAI tool 1.0 users averaged 2.4 sessions per user

Real-world CAI tool use showed strong engagement across a large user base of 393,969 Headspace members.

• Engagement descriptive analysis assessed session and message counts, retention, and conversation themes
• Total sample of 393,969 Headspace members was analyzed for real-world engagement
• This was a descriptive analysis of real-world use rather than a controlled trial

Survey respondents viewed the CAI tool primarily as a guide to navigate mental health resources and provide in-the-moment support.

• Members saw the CAI tool as a guide to navigate to mental health resources and Headspace content
• Members emphasized the need for data safety and ethics transparency
• Members indicated the CAI tool should be a resource in addition to human-delivered mental health care, not a replacement
• Members emphasized the need for clinical guidelines structure

Diary study participants envisioned using the CAI tool during specific emotional states and daily routine moments.

• Diary study included n=15 Headspace members exploring the CAI tool's role within members' mental health journey
• Members imagined using the CAI tool when feeling stress or anxiety
• Anticipated use times included morning routines, commutes, or while winding down at night

The study identified three minimum viable safeguards necessary for purpose-built mental health AI products.

• Safeguard 1: Transparent labeling of intended use, benefits, and limitations
• Safeguard 2: Safety by design principles to monitor for overuse, detect risk, and flag needs for escalation
• Safeguard 3: Child and adolescent safeguards
• The authors contrasted purpose-built tools against general AI tools which lack safety guardrails, evidence-based practices, and medical regulation compliance

The CAI tool (Ebb) was purpose-built using motivational interviewing techniques with clinically backed safety mechanisms.

• Ebb was built by clinical psychologists and research experts
• The tool uses motivational interviewing techniques for subclinical guidance
• It incorporates clinically backed safety mechanisms
• The tool is designed for subclinical guidance rather than clinical treatment