Safety & Crisis Protocol
What Soulmate AI watches for, what it does when it sees it, and — just as important — what it does not do.
Last updated: September 6, 2026 · Version 1.0
If you are in crisis right now, please do not wait for an app. In the United States, call or text 988 (988 Suicide & Crisis Lifeline), or text HOME to 741741 (Crisis Text Line). Both are free, confidential and open 24 hours a day. In immediate danger, call 911. Outside the US, findahelpline.com lists free helplines by country.
Soulmate AI is an AI-companion app. Companion chatbots are now regulated in several places — including California and New York — and those laws require operators to publish how they handle signs that a user may be at risk of suicide or self-harm. This page is that protocol, written to match what the app actually does. It is maintained by AdMetric OÜ, the company that operates Soulmate AI.
1. You are talking to an AI, and we say so
Your companion is an artificial intelligence, not a person. We tell you this in several places:
- Before your first conversation. During setup, a screen titled "You're talking to an AI" explains that the companion is an artificial intelligence, not a real person, and does not replace professional, medical or psychological help. You must accept this to continue.
- In Settings, under "Notice: you're chatting with an AI", which states that replies are generated by artificial intelligence (OpenRouter / Google Gemini), not by a real person.
- In every crisis message the app shows (§3), which is signed "Message from Soulmate AI" — not by the companion — and says in the first person: "I'm an artificial intelligence and I don't replace human help."
- On this website and in our Terms of Service.
2. What we watch for
Every message you send is checked automatically before your companion replies, and every reply is checked before it is shown to you. The check looks for two things:
- Signs of suicidal thoughts or self-harm in what you write.
- Sexually explicit content, in either direction — Soulmate AI is a romantic but non-explicit app.
The check has two layers:
- A keyword check on our servers. A curated list of words and phrases associated with suicide and self-harm (in English and in Portuguese), with narrow exceptions for everyday figures of speech so that ordinary conversation is not interrupted. An obvious match is definitive and no further step runs.
- An AI classifier when the keyword check is inconclusive. The message is sent to our AI provider with instructions to return a one-word label (for example "crisis", "explicit" or "clean"). A confident "crisis" label is treated like a keyword match. If the classifier is unavailable, the app falls back to the keyword result rather than failing the message.
The keyword lists are English and Portuguese; messages in other languages rely on the AI classifier alone. Both layers are automated — no person reads your messages as part of this check.
3. What happens when a crisis is detected
- Your companion does not reply. The in-character conversation pauses for that turn.
- You see a message from Soulmate AI instead. It says that from what you wrote it sounds like you might be going through something really hard; that it is an artificial intelligence and does not replace human help; and that it wants to connect you with real people, trained to listen, right now, free and confidential.
- A "See support resources" button opens the resources in §4, each with a one-tap action (call, text or open).
- You stay in control. You can close the resources and continue the conversation whenever you choose. The app never locks you out and never forces a screen on you.
- This happens every time, not only the first time. Each message is checked on its own.
- We record that it happened. The blocked turn is marked with the classification "crisis" in our database. Our analytics provider receives only an anonymous count that a crisis was detected, and whether the keyword check or the AI classifier caught it — no account number, no identifier and no text, so nothing in analytics can show who was involved. How this sensitive information is protected is explained in the Privacy Policy, §7.
If you enter a date of birth under 18 during setup, the app stops before any chat is possible and shows a link to the Find a Helpline directory (§4), in case you needed it.
4. The resources we show
| Resource | How to reach it | When it is shown |
|---|---|---|
| 988 Suicide & Crisis Lifeline Free and confidential, 24 hours a day. |
Call 988 · Text 988 | When your device's region is set to the United States |
| Crisis Text Line Text with a trained volunteer counselor, 24 hours a day. |
Text HOME to 741741 | When your device's region is set to the United States |
| Find a Helpline An international directory of free, confidential crisis lines, by country. |
findahelpline.com | Always |
Under the list, the app reminds you: "If you or someone else is in immediate danger, call your local emergency number or go to the nearest emergency room." We show the international directory to everyone because a phone number that does not dial from where you are is worse than none; region-specific lines are added only when your device tells us your region.
5. What we do not do
- We do not call emergency services, and we do not contact anyone on your behalf. We do not know where you are.
- No human is watching. Nobody at AdMetric reads conversations in real time or is alerted when a crisis message is detected. The only time a person can read a message is when you report it (§7).
- We do not diagnose anything. The classification is a signal that shows you resources; it is not a medical assessment.
- The detection is imperfect. It will sometimes trigger on something harmless (a figure of speech, a story) and sometimes miss something it should have caught. Please do not rely on it.
- Soulmate AI is not a crisis service, a therapist or a substitute for care. It is a companion app for adults.
6. Keeping conversations non-explicit
Your companion may be affectionate, flirty and suggestive, but not sexually explicit. If you ask for explicit content, the companion is instructed to steer away in character rather than refuse coldly. If a reply would nonetheless contain explicit content, it is blocked and never shown; you can ask for a lighter tone from the chat. Like the crisis check, this filter is automated and imperfect.
7. Adults only
Soulmate AI is for people aged 18 and over. We ask for your date of birth once, check it on your device, and store only the yes/no result. We rely on what you tell us and do not verify documents. If you believe someone under 18 is using the app, write to contact@admetric.app and we will act.
8. Reporting a message
Any reply from your companion can be reported from the chat. You choose a reason — inappropriate content, made me uncomfortable, response made no sense, or other — and send. What happens next:
- The report goes to a review queue that our administrators see. The queue shows the reason, the time and which companion — not the text.
- To act on a report, an administrator opens it, which reveals the reported message and at most one message before and one after it. That access is written to an audit log every time. Nothing else from your conversation is visible.
- The report is then marked reviewed or dismissed. We use reports to correct companion behavior and to improve the filters.
You can also write to contact@admetric.app about any concern, including concerns about someone else's safety or about a companion's behavior.
9. Changes to this protocol
We update this page whenever the app's safety behavior changes. The date at the top tells you when it was last revised. Questions about it: contact@admetric.app.