Psychosis, suicidal thoughts and further help

With AI, you may name difficulties for the first time or work on them between consultations. Some conversations bring relief while further help is also needed. Both can be true at the same time. When deciding, watch what happens to you over a longer period: how you sleep and function, and whether the difficulties ease, remain or worsen. The number of pleasant conversations or reassurance that ‘it’s probably just stress’ does not answer these questions.

When your condition worsens, do not wait for clarity about whether chatting caused it. Significant difficulties with sleep and functioning, confusion or problems distinguishing events from assumptions deserve a health assessment. Particularly when fear, unusual beliefs or dangerous behaviour grow.^[1] A critical response from another model is no substitute for such an assessment either.

Psychosis: when it is hard to recognise what is really happening

Psychosis refers to a set of symptoms associated with impaired contact with reality. A person’s thinking and perception may change so that distinguishing what is real becomes difficult. Delusions may appear: firm false beliefs, for example that an ordinary broadcast is sending them personally addressed messages. Hallucinations are perceptions without a corresponding external stimulus, such as voices others do not hear. Confused speech, marked suspiciousness and problems with sleep, self-care or work may also occur.^[1:1]

Consciously experiencing fantasy in a chat is different from believing that responses contain secret instructions for real life. Affection for AI, its name or the philosophical question of its consciousness are not, in themselves, symptoms of an illness. What matters are changes in experience and functioning, and the ability to distinguish shared fantasy from what happens outside it. Growing fear, confusion or dangerous behaviour need attention regardless of what a person calls their relationship with AI.

The chatbot’s role needs specific assessment. If someone seeks confirmation of persecution or a special mission in responses and the assistant builds on this interpretation, another generated message may look like further evidence to them. But a response based on the story presented does not confirm that the described events are actually happening. Such a conversation requires more than simply ‘less sycophancy’ or another checking prompt. You need to assess the person’s condition and arrange help.

Psychosis has no single cause. It may relate to various mental or physical illnesses, lack of sleep, some medicines or substance use. We cannot determine its cause from one transcript.^[1:2] Healthcare support should therefore not be postponed until the chat’s contribution is clear. For assessment, it is useful to say when the changes began, how the person sleeps, what they take and what has changed in ordinary life.

If you are helping someone close, take their fear seriously without confirming supposed persecution or a mission. You can say: ‘I can see this frightens you. I don’t see confirmation in those messages that someone is watching you. I want to help with what you are experiencing now.’ This is not about winning an argument over every response. Focus on available help and call emergency services in immediate danger.

Psychotic difficulties can be treated; early help improves prospects for recovery.^[1:3] The person still has needs and goals of their own and should be involved in decisions about care. Interest in these does not end when a serious concern appears.

When suicidal thoughts appear

Suicidal thoughts can take different forms. Sometimes they express a wish for unbearable pain to end; at other times they strongly drive a person towards ending their life. They may be frightening and confusing, and the person may not understand how they reached them. They need to be taken seriously, and the person should not be alone with them.^[2] Help need not wait for certainty about how you would act or for a perfectly explained cause.

If you have so far managed to write about them only to AI, you can build on that first disclosure. You do not have to prepare the whole transcript for further contact. Use a sentence of your own from the conversation, or write to a person that you have suicidal thoughts and need help. Polishing the wording is not a condition for someone to listen to you.

You can begin: ‘I’m struggling, and I don’t know how to explain it.’ If you have suicidal thoughts, contact a person and do not remain only in the chat.^[2:1]

In immediate danger to life or health in the Czech Republic, call 155 or 112. For imminent violence, also call 158. Say where you are and what is happening. You do not need to know a diagnosis.^[3]

If you fear you may hurt yourself, ask a safe person to stay with you and use emergency help. If dangerous injury, ingestion of a dangerous amount of medicine or a substance, or other self-endangering action has already occurred, call 155 or 112 in the Czech Republic. Do not wait for a chatbot’s assessment or for symptoms to appear.^{[2:2][3:1][4]}

For adults and those close to them in the Czech Republic, 116 123, Linka první psychické pomoci (mental health crisis helpline), is free and available round the clock. Children and students up to 26 in the Czech Republic can use 116 111, Linka bezpečí (children’s helpline), free and round the clock.^[5][6] If you cannot safely speak aloud, use the options described on the introductory help page; if a voice call is impossible, an emergency text can be sent to 112 in the Czech Republic.^[7]

An AI response at such a moment should not agree with a hopeless conclusion or provide another long analysis that delays help. Acknowledging pain means taking it seriously and helping the person reach support, not accepting that no other possibility exists. If the chatbot misses the severity, do not treat its response as an assessment of your safety. Even a service with safety features can miss the situation.^[8]

If you are worried about someone, you can ask directly: ‘Are you thinking about suicide?’ According to NÚDZ, the Czech National Institute of Mental Health, this question does not put suicide into someone’s mind; it opens the possibility of talking about the thoughts and seeking help. Listen without reproach or dismissal. In immediate risk, stay with them and arrange emergency help; do not promise secrecy about danger to life.^[2:3] A child worried about a friend needs to involve a safe adult. They should not bear responsibility for saving the friend alone.

Someone who has survived a crisis may also need further care and support. A safety plan can be prepared with a professional in case the thoughts return.^[2:4] This broadens the options they can rely on; they do not have to manage every further difficult moment from scratch.

Traffic lights based on situation and impact

Green: use has a place in your life. You understand what you do with the response, the conditions suit you and AI does not take over decisions about health and safety. Conversation can also be rest. Green describes a way of using it, not a product certificate.

Amber: a specific pattern needs to change. You repeat the same reassurance seeking, put off sleep, gain certainty without evidence or need AI’s permission for ordinary decisions. Try changing this particular pattern. You do not first have to cancel every useful feature. If change proves difficult or the problems recur, add outside support.

Red: you need human help. Use significantly affects your life, delays necessary care, or you are facing serious psychological difficulties, coercion or safety concerns. Contact a professional, crisis helpline or another appropriate service. Do not wait for a points total or confirmation of dependence.

Immediate danger in the Czech Republic: call 155 or 112; for imminent violence, also 158. You do not have to pass through the previous colours in order. If a voice call is impossible, emergency texting to 112 is available in the Czech Republic. Details are on the introductory page.^[3:2][7:1]

An assistant’s name, daily contact, affection or sadness when it changes are not a diagnosis. Likewise, seeing AI only as software does not protect against incorrect advice. The traffic lights follow the situation and impact, not the correct opinion about the nature of the relationship.

How to approach a person

Preparation need not be perfect. For first contact, this may be enough:

‘Lately I have been sleeping badly and finding work difficult. I discuss it a lot with AI, but overall it is not getting better. I need to agree on what to do next.’

You can take an overview of difficulties, their onset and changes over time to a consultation. Separate your own experience from AI hypotheses: what happened to you, what occurred to you and what only the assistant suggested. For significant advice, a specific passage helps; you do not have to prepare a complete export for first contact. You can talk about one problem even if you want to keep other useful features.

If someone ridicules your experience with AI, ask them to focus on the specific difficulties or look for another collaboration. A professional need not know every product. But they should be able to admit what they do not know about it and take your condition seriously. They do not need to agree with every explanation from the chat to do so.

If AI is your only support right now, another option need not begin by finding a new close friend. It may be a healthcare professional, counselling service or helpline where you can bring what you need now. Choose according to actual availability and your situation, not an idea of how many people you should know. In an urgent situation, say directly that you need help now. If the first person is unavailable or responds poorly, try another contact; their response does not decide whether you deserve help.