Is it safe to talk to an AI about depression?
An honest answer, including the limits — what it is good for, what it is not, and what to check first.
We build one of these, so treat this page accordingly — but a guide that dodged the question would be worth nothing, so here is a straight answer with the uncomfortable parts left in.
“Safe” is really three separate questions: is it safe for your privacy, is it safe when things are bad, and is it safe in the sense of being good for you. They have different answers.
Privacy: it depends entirely, and you can check
The words you type are data, and where they go varies enormously between products. Things worth establishing before you write anything you would mind seeing again:
Is it tied to your identity? An account with an email is connected to a named person. One without is not. That difference matters more than any assurance in the marketing copy — there is more on this in mental health apps that do not need an email.
Is your conversation used to train models? Many general-purpose assistants may use conversations for improvement; dedicated mental health products often contractually prevent it. Check rather than assume, in either direction.
Who else sees it? Advertising and analytics are the usual routes by which something leaves.
Can you delete it, concretely?
A general chatbot is not the same product as a purpose-built mental health app, and the privacy answers are usually quite different. The questions are the same though, and all four are answerable in about five minutes.
In a crisis: no, and this is the important part
An AI is not a crisis service and should not be treated as one. It cannot send help, cannot call anyone, cannot assess risk properly, and cannot be relied upon to respond appropriately when the stakes are highest.
If you are thinking about harming yourself, the right move is a human — a crisis line, your doctor, emergency services, or someone in the room. The numbers at the bottom of this page are free, staffed around the clock, and answered by people trained for exactly that conversation.
A responsible product should be direct about this, should recognise crisis language, and should put those routes in front of you rather than continuing the conversation. That is a reasonable thing to test before you rely on something.
Is it good for you? Honestly, it depends what you use it for
What it is genuinely useful for:
Getting things out of your head at 3am, when the alternative is circling the same thought until morning. Articulation helps, and something that responds is easier to talk to than a blank page.
Finding the words before you need them. A lot of people find it easier to say a hard thing to a person after having written it once. That is a real use, and it is the one nearest to what describing this to a doctor asks of you.
Not being alone with it at hours when nobody else is available. Availability is not a small thing at 4am on a Sunday.
What it is not:
It is not therapy, and not a therapist. There is no treatment being delivered, no clinical judgement, no accountability, and no relationship in the sense that word means clinically. Any product implying otherwise is overselling.
It cannot diagnose you, and should not try.
It will not push back the way a person would. This is the subtle one. An AI is agreeable by construction. A good friend, or a good clinician, will sometimes say the thing you do not want to hear. That is often exactly the thing you needed, and it is the hardest quality to get from software.
It is not a reason not to see someone. If talking to an app is making it easier to postpone a GP appointment, that is the app working against you.
How to use one sensibly
Use it as a place to put things, not as the plan. Keep the human routes open in parallel — the GP, the referral, the friend who knows. Notice if it is becoming a way of avoiding people rather than a way of getting through to them. And check the privacy answers once, at the start, rather than wondering later.
The short version
Privacy varies and is checkable in five minutes. In a crisis, use a human — always. As somewhere to think out loud at hours when nothing else is open, it can genuinely help, provided it stays alongside real support rather than quietly replacing it.
— Richard MacLean
Related: Woebot shut down — what now? and anonymous mental health apps: an honest guide.
Nagi is a mental health companion, not a medical device, and nothing here is medical advice. If you are struggling, please reach out to a professional or someone you trust.