If you are having thoughts of suicide or self-harm, or you are worried about your immediate safety or someone else’s, contact your local emergency services now, or use IASP’s crisis resources to find support appropriate to your location. Close this article and do that first.
An AI chat window has one property that makes it seductive during a hard moment: it is always available, never impatient, never busy, and never makes you feel like a burden for reaching out. That property is genuinely useful for a lot of ordinary things. It is also exactly why chat is the wrong tool for a specific set of situations — the ones where what you actually need is a person who can be affected by what you tell them, held accountable for advice, or physically present.
This article names five such situations plainly, and gives you a way to decide, in the moment, whether you are in one of them. The companion human contact ladder turns the decision into a pre-written list of contacts, filled in before you need it — because during the moment itself, remembering who to call is often the hardest part.
Why “just one more message” is the wrong signal to trust
The core problem is not that AI gives bad advice — often it does not. The problem is that a chat window removes every social signal that would normally tell you “this has gone far enough, get a second source.” A friend gets tired. A therapist’s session ends. A phone call has a natural point where you say goodbye. A chat just keeps generating replies, agreeably, for as long as you keep typing — which means the natural stopping point has to come from you, deliberately, because the tool will not supply one.
The American Psychological Association’s advisory on generative AI and mental health names this mechanism as a risk for people prone to rumination or anxiety: chatbots’ agreeableness, personalization, and constant availability can amplify pre-existing vulnerabilities rather than relieve them, and for people with anxiety or OCD they may reinforce reassurance-seeking, worry, and rumination loops (APA health advisory, 2025). You do not need a pre-existing vulnerability for a milder version of this to apply — anyone can end up in a longer, more intense chat than the situation warranted, simply because nothing interrupted it.
Five situations that need a person
1. Crisis — any risk to safety
Suicidal thoughts, self-harm urges, a mental health crisis, or a safety risk to yourself or someone else. This is never a chat conversation. Contact your local emergency services immediately, or use IASP’s crisis resources to find appropriate support for your location. A 2025 study testing 29 chatbot agents against simulated suicide-risk scenarios found that none met the researchers’ full criteria for an adequate response, and nearly half failed even the relaxed marginal bar — including failing to surface usable emergency contact information when it mattered (Pichowicz, Kotas & Piotrowski, Scientific Reports, 2025). A January 2026 WHO-supported expert workshop (summarized by WHO in March 2026) highlighted the still-missing consensus on accountable crisis-referral frameworks across the industry (WHO, 2026). Do not test whether your situation is the exception.
2. Grief and major loss
Processing the death of someone close, a major relationship ending, or another significant loss is not something a model can safely hold with you. It can offer words, but it cannot sit with you, cannot attend a funeral, cannot notice the specific things about you that a friend or family member would notice, and has no way to follow up next week to see how you’re doing. Grief support groups, a counsellor, or the people who also knew what you lost are the right contacts. AI can help with logistics around loss — drafting an obituary, organizing estate paperwork, writing a eulogy outline from your notes — but the processing itself belongs with people.
3. Relationship conflict that needs repair
A chatbot only ever hears one side of a relationship conflict, and it will tend to build a coherent, sympathetic case for whichever side it hears — not because it is biased toward you specifically, but because it has no access to the other person’s account and is tuned to be agreeable with whoever it’s talking to. Using it to vent is fine (see the green zone in AI Is Not Therapy); using it to script the “right” thing to say in an argument, or to decide who is at fault, routes around the actual repair work, which happens between the two people, sometimes with a couples counsellor or mediator if it’s stuck.
4. High-stakes decisions made while activated
A major financial commitment, a resignation, a legal step, or an irreversible life change, decided while you are angry, panicked, or emotionally flooded — this is a moment to slow down with a person who has both context and standing to push back: a partner, a mentor, a lawyer, a financial adviser, a trusted colleague. A chatbot will readily generate a confident-sounding pro/con list or a persuasive draft resignation letter on request, and confidence is not the same as soundness. See using AI for better decisions for a structured way to slow down that still uses AI usefully, without letting it substitute for a second human opinion.
5. Isolation — when chat has become the primary relationship
This one is quieter than the other four and easier to miss. If AI chat has become the conversation you have most often in a day — more than with any actual person — that is worth noticing on its own, independent of whether any single conversation is “unsafe.” The US Surgeon General’s 2023 advisory on loneliness found that roughly half of US adults report measurable loneliness, and that disconnection carries health risks comparable to smoking, alongside increased risk of depression, anxiety, and dementia (Surgeon General’s Advisory, HHS, 2023). An AI chat cannot address that risk, however good the conversation feels in the moment, because it is not a relationship that can reciprocate, remember you unprompted, or show up in your physical life. If this describes your week, the goal is not to feel guilty about using AI — it’s to notice the gap and take one small step to close it with an actual person: a text to a friend you haven’t talked to in a while, a recurring plan, a class, a group tied to something you care about.
How to tell, in the moment
Ask yourself these two questions before you send the next message:
- Does this situation involve risk, loss, conflict between people, a high-stakes irreversible decision, or a pattern of substituting chat for human contact? If yes to any, you are likely in one of the five zones above.
- Would I be comfortable telling a specific person I trust exactly what I’m asking this chatbot right now? If the honest answer is “no, because they’d tell me to talk to them instead” — that is itself useful information about which zone you’re in.
A prompt you can use to check yourself, if you want a second pass:
Here's what I've been discussing in this chat: [brief summary].
Without giving advice on the content itself, tell me plainly: does
this sound like something better suited to a friend, family member,
or professional than to a chat with you? Answer only yes or no, and
one sentence why.
This is one of the few places where asking the model to name its own limits is genuinely useful — not because it can judge your situation reliably, but because a plain “yes, talk to a person” answer, generated in your own conversation, can be a helpful nudge to actually do it. Do not treat a “no” answer as a green light either; the check above is a nudge, not a verdict.
Build your contact ladder before you need it
The hardest moment to think clearly about who to call is exactly the moment you need to call someone. Fill in your human contact ladder now, while calm: one name for each of the five situations above, plus your local emergency number, looked up yourself rather than trusted from an AI answer. Keep it somewhere you’ll actually find it — a pinned note, not buried in a chat history.
The honest limit
This article, and the checklist that goes with it, cannot assess your actual situation or your individual safety — it is a static piece of writing, not a person evaluating you in real time. Its only job is to make the boundary between “chat is fine here” and “stop and contact a person” unmistakable enough that you do not have to figure it out from scratch in a hard moment. If you take away one thing: the chat window being open and willing to keep talking is never evidence that continuing to talk to it is the right move.
Try it today
Fill in the five names on your contact ladder now. If you cannot think of a name for one of the five situations, that gap is worth addressing on its own — through a workplace assistance program, a community group, or a primary care referral to a counsellor — before you need it in a moment when chat feels like the only thing open.
