Study: AI suspicion in crisis counseling triples over 10 months
21.5% of suspicious clients explicitly prefer human counselors.
A new study from researchers at Stanford and collaborating institutions analyzed 75,777 real-world crisis counseling conversations from a human-staffed WhatsApp helpline in India to quantify how often clients suspect they are talking to an AI—and what happens when they do. Though no AI was actually used in these interactions, the proportion of conversations where clients voiced suspicion of AI involvement increased sharply from 0.8% in June 2024 to 2.6% in March 2025, a more than threefold rise in under a year. The study, published on arXiv, sheds light on a growing trust gap as mental health platforms explore AI integration.
Among conversations where suspicion was raised, 21.5% of clients explicitly stated a preference for speaking to a human. Suspicion typically surfaced early—68.3% of cases occurred in the first half of the conversation. When counselors tried to reassure clients (e.g., saying 'I assure you; this is not AI'), 17.6% of those clients either continued to press the issue or ended the conversation entirely. The findings signal that simply denying AI involvement may not be enough to rebuild trust. As AI tools increasingly assist counselors with triage, note-taking, or even generating responses, understanding these dynamics will be critical to designing systems that preserve the therapeutic relationship.
- Client suspicion of AI in crisis counseling rose from 0.8% to 2.6% over 10 months across 75,777 conversations.
- 21.5% of suspicious clients explicitly preferred human counselors; 68.3% of doubts emerged in the first half of messages.
- Reassurance from counselors backfired 17.6% of the time, leading to continued pressing or conversation termination.
Why It Matters
As AI integrates into mental health, this data reveals a growing trust gap that platforms must address.