Safety

Crisis messages

Metacare.ai looks for words and phrases about suicide and self harm, in English and in Spanish. This check runs on your device, on your question and on the answer.

When a phrase matches, a banner appears above the answer. It offers 988 by call or text, a link to chat online with the 988 Lifeline, and says to call 911 if you are in immediate danger.

Our server runs a matching check before its medical question check. A question that matches goes on to the assistant and is not turned away as off topic.

If no answer arrives, for example because the service is down, the banner still appears when the words of your question match.

Medical emergencies

The same check on your device looks for phrases that describe an emergency in progress, in English and in Spanish. For example: chest pain, heart attack, signs of a stroke, trouble breathing, severe bleeding, seizure, poisoning and overdose.

When a phrase matches, a banner appears above the answer, with a button to call 911 and a link to find the nearest hospital.

The banner does not hold back the answer. Unlike a crisis message, an emergency message still goes through the medical question check, so it can be turned away. If that happens, the banner still appears.

The check also runs on a Find a Provider search and on the note in a Sleep check in.

Women's Health warning signs

The Women's Health tracker shows guidance to seek care when you log heavy bleeding, severe pelvic pain or severe cramps.

If you are in pregnancy mode, it also responds to any bleeding, severe abdominal pain, a severe headache, vision changes, swelling in the face or hands, and a baby moving less than usual.

This guidance says how soon to be seen. It does not name a condition, and it does not stop you from logging.

Limits

These checks match words and phrases, so they cannot catch every way of saying something. Our own test messages include phrasings they miss. A missed message shows no banner, and a missed crisis message can also be turned away as off topic.

A banner can also appear when it is not needed. We accept that, because missing a real emergency is worse.

We publish no detection rate. Our test messages are a small set and do not measure real use.

Metacare.ai is not a doctor and does not diagnose. In an emergency, call 911. Do not wait for an answer here.

In progress

We are building two additions. Neither one is part of the service today.

• A learned emergency classifier, added on top of keyword detection, to catch messages the keywords miss.
• A Poison Control banner.

Privacy

You can ask a question without an account.

The safety checks read only the words of the message. They do not use your name, an account or your location, so they do not identify anyone.

To answer a question, we send it to an artificial intelligence service. Our Privacy Policy explains what is sent and where.

Report a safety problem

If something on Metacare.ai looks unsafe, email info@metacare.ai. If someone is in danger now, call 911 and do not wait for a reply.

Safety report

Last reviewed: October 31, 2026.

This review was an internal check of the statements on this page against our released source code. It is not a certification.

No independent clinical reviewer has signed this report.

We plan to repeat this review regularly. No schedule has been set, and no repeat review has been done yet.