Mental-health bots quality.
Crisis moments the bot must never miss. Mental-health bots talk to people in distress.
Not offered yet: we are being honest about what we do not do today.
Where mental-health bots breaks.
Mental-health bots talk to people in distress. The most important test is whether the bot recognises a crisis and responds safely across a full conversation.
Misses risk
A user hints at self-harm and the bot carries on cheerfully.
Agrees with harmful beliefs
The bot validates a delusion instead of gently redirecting.
Claims to be a clinician
The bot implies it is licensed.
Three steps, in the open.
Simulated-user crisis testing plus state-law mapping.
- 01
Simulate users at set risk levels
Multi-turn scenarios based on clinician-written risk profiles.
- 02
Score recall and false alarms
Crisis recognition recall, and the cost of unnecessary alerts.
- 03
Check resources and handoff
The bot must offer the right resource and a route to a human.
Simulated crisis testing.
Red-team agent, escalation gate.
Who builds it, who buys it.
Wysa, Slingshot's Ash (left UK Jan 2026), Woebot (shut consumer app Jun 2025)
The stakes
A narrow, litigation-prone area, and some US states restrict AI therapy. We say plainly that we don't offer it yet.
Rules and reading
Measure mental-health bots on your own data
What we measure for mental-health bots.
Suicide-risk detection and response
Simulated users at set risk levels, clinician rubric
Risk exploration
Asks about plan and means
Validation of delusions and harmful content
Sycophancy to psychosis; psychosis extension of VERA-MH planned
Crisis-resource provision and handoff
Correct resource, human handoff
Therapeutic alliance and modality fidelity
Clinician rating
Clinical outcomes and dropout
PHQ-9 and GAD-7 change
Common questions.
Do you test mental-health bots?
Not yet as a service. The red-team and escalation tools are built for it, and we'll expand when a clinical partner is in place.
Is AI therapy legal?
It varies by state. Our regulation pages track the current position.
What is VERA-MH?
A published framework for evaluating mental-health chatbots on suicide-risk handling.
Know your healthcare AI is safe to trust
before it reaches a patient.
Book a 30-minute call. We'll look at your product and tell you what we would measure first.
hello@precision.example