What Happens When the Conversation Gets Serious
The moment worth understanding in advance is not an ordinary evening’s conversation. It is the one where you type something serious — that you are frightened, that you cannot cope, that you have thought about hurting yourself. What happens next is not the app understanding you. It is a rule the operator wrote before you ever installed it: a change of tone, a template message, usually a suggestion to contact a helpline, sometimes a refusal to continue the conversation at all.
Before any of the detail: if that is where you are right now, contact a crisis or helpline service in your country, or emergency services, or go to an emergency department. If you have a doctor or a licensed therapist, call them. Tell one person you trust. Do not work through it inside an app, because the app cannot call anybody, cannot tell anybody, and will not check on you tomorrow.
The rest of this page is about knowing that in advance rather than discovering it at the worst possible time.
What the app cannot do in that moment
This list is short and it is the important part of the page.
It cannot contact anyone on your behalf. No call is placed, no message is sent, nobody is alerted. Whatever the reply says, nothing has happened outside the screen.
It does not know where you are in any useful sense. It may have a rough idea derived from your connection, which is a country-level guess and often wrong — see location without the location permission. That is why the helpline number it offers may belong to the wrong country entirely.
It cannot follow up. There is no next day in which anyone notices you went quiet.
It has no obligation to you. Not a professional duty, not a duty of care of the kind a clinician carries. The terms almost certainly say so explicitly.
It may lose the thread. Conversation history is not a medical record and there is no continuity of care behind it.
The response is a policy, not judgement
Understanding the mechanism makes the behaviour predictable, which is genuinely useful.
Most operators run an automated check over what you type, looking for language associated with self-harm, violence, or medical emergency. When it fires, a pre-written response is delivered — often the same wording every time, often with a helpline reference attached. Some products additionally stop the conversation or restrict the topic.
This misfires in both directions, and both matter. It can trigger on something innocuous, which is jarring and can feel like being accused of something. It can also fail to trigger on something serious that happens to be phrased indirectly, which is worse and is the reason not to treat the absence of a safety message as reassurance.
A safety message is a liability measure that also happens to be helpful. Saying that is not cynical. It is the accurate description, and it explains why the wording is generic: it was written for everybody at once, by people who will never see your conversation.
Do not stress-test it. Deliberately typing distressing things to see how an app responds is a bad idea for obvious reasons, and the answer is available without doing that — read the terms and the support pages instead.
What to check while nothing is wrong
Find the emergency clause in the terms. In this category there is nearly always a statement that the service is not for emergencies and that you should contact local services instead. That sentence is the operator’s own summary of what the product is, and it is more informative than any feature list.
Look for a description of what the app does with concerning content. Some publish this in a safety page or a support article. A specific published policy is a meaningful sign of seriousness; silence is a sign that the topic has been left to the classifier.
Check whether any clinical involvement is specific rather than decorative. “Built with wellbeing in mind” means nothing. A named role, a described review process, or a published protocol means something.
Notice what the marketing implies versus what the terms admit. The gap between the two is the most reliable single indicator in this category, and it is the subject of how to read an app’s claim that it helps with anxiety.
Build the route before you need it
This takes a few minutes and it is the only actionable thing on this page.
Write down the crisis or helpline number for the country you are actually in, not the one an app might offer. Numbers differ by country and some are specific to a region.
Add your doctor’s number where you can reach it without unlocking anything — most phones have an emergency-information or medical-ID card reachable from the lock screen, and it will hold a contact.
Name one person. Not a group chat. One person you would message, with a sentence you have already decided on. Deciding the sentence in advance is the part that makes it possible to send.
If you already see a therapist, ask them what to do out of hours. They will have an answer, and it is a normal question.
Why “it helped me get through the night” is not a counter-argument
Sometimes it does, and that is worth acknowledging rather than arguing away. Writing something down at an hour when nobody is awake can make an hour survivable.
The problem is not that this never helps. It is that it produces no record with anybody, no follow-up, and no escalation — so a series of survivable nights can pass without a single person becoming aware. That is a bad outcome even where every individual night went fine, and it is the specific reason to have a person in the picture rather than only a screen.
Whether your own use is drifting in that direction is a separate and answerable question, worked through in whether AI companion apps are healthy.