How people use Claude for support, advice, and companionship

Key Findings

• — Only 2.9% of Claude.ai interactions are affective conversations; companionship and roleplay combined are under 0.5%.
• — Users bring practical, emotional, and existential concerns: career, relationships, persistent loneliness, and questions of meaning.
• — Claude refuses user requests in supportive contexts less than 10% of the time, with pushback concentrated on safety grounds (e.g., dangerous weight-loss advice, self-harm support).
• — Expressed user sentiment tends to shift slightly more positive over the course of affective conversations, with no clear negative spirals observed — though the authors caution this does not establish lasting emotional benefit.

Methodology Notes

~4.5M conversations from Claude.ai Free/Pro accounts screened down to 131,484 affective conversations; automated privacy-preserving analysis via Clio with multiple anonymization layers; classification validated against opt-in user feedback data. Limitations: expressed language only (no psychological outcomes), no longitudinal data on dependency, snapshot in time, text-only, and Claude is not designed for emotional support — limiting generalization to purpose-built companion platforms.

Sources