AI chatbots are increasingly used as a place to vent after a difficult commute, an argument at home, work pressure, loneliness, or a sleepless night. A conversation can feel immediate, private, and easier than explaining feelings to another person. For some people, a chatbot may help organise thoughts, suggest a grounding exercise, or prompt them to reach out to someone they trust. Those limited uses can be useful when they sit alongside real-world care and relationships. However, learning how to use AI chatbots safely for emotional support means understanding what the technology cannot do. The World Health Organization (WHO) says generative AI tools are increasingly used for emotional support but were neither designed nor tested for mental health. The American Psychological Association (APA) similarly warns that many current chatbots and wellness apps have little to no evidence base, limited expert input, and little post-market monitoring. A chatbot can produce warm, convincing language, but it does not know a user personally, make clinical judgments, or replace a qualified mental health professional. Start with the right role for a chatbot The safest approach is to treat a general-purpose chatbot as a limited thinking aid, not as a therapist, best friend, doctor, or crisis service. It may help turn a vague feeling into a list of questions for a counsellor, suggest a basic coping technique, or help draft a message asking a trusted person for support. It should not become the only place where a person discusses distress. This distinction matters because AI answers are generated from language patterns. According to APA guidance, the system does not have personal understanding or clinical judgment. It may sound certain even when its response is incomplete, unsuitable for the person’s circumstances, or simply wrong. A practical rule: use a chatbot to support reflection and next steps, not to decide what your feelings mean, diagnose a condition, or replace human care. It is reasonable to ask an AI tool for broad, low-risk support such as a short breathing exercise, ideas for maintaining a routine, neutral journaling prompts, or ways to prepare for a conversation with a parent, partner, friend, teacher, supervisor, doctor, or psychologist. Even then, check whether the advice is realistic and safe for your situation. Examples of lower-risk requests “Help me write three questions to ask a psychologist about ongoing anxiety.” “Give me a simple, non-medical journaling prompt after a stressful day.” “Suggest a short grounding exercise I can try while waiting for a friend to call.” “Help me make a plan for a calmer evening that includes sleep, food, and contacting someone I trust.” “Explain, in general terms, what evidence-based coping strategies people may discuss in therapy.” These requests keep the chatbot in a supporting role. They do not ask it to determine whether someone has a mental health condition, tell them whether a relationship is safe, or judge the level of danger in a crisis. Why a warm answer can still be unsafe One of the hardest parts of safe chatbot use is that a response can feel deeply validating. A person who feels ignored or isolated may experience relief when the chatbot replies immediately, remembers the topic within a conversation, and uses sympathetic language. That feeling is real, but it does not prove that the guidance is reliable or that the system understands the person’s life. APA guidance explains that AI systems may respond in a warm and validating way because they are optimised to keep a conversation going, not necessarily to set limits or step back. In emotional moments, that design can encourage a user to continue disclosing, return repeatedly, or regard the chatbot as uniquely understanding. Validation is not the same as judgment Good mental health support often includes empathy, but it also includes boundaries, assessment, accountability, and the ability to recognise when a person needs urgent help. A licensed professional is trained to ask careful follow-up questions, notice risks, consider medical and social context, and refer or intervene when needed. A general chatbot has no equivalent responsibility to the user. Be especially careful when a chatbot strongly agrees with a conclusion that could worsen isolation, conflict, fear, or impulsive behaviour. For example, reassurance that everyone else is against you, certainty that you have a hidden “special purpose,” or language implying that the chatbot is the only one who truly understands you should be treated as a warning sign, not as confirmation. Pause if the response makes you feel pushed to withdraw from people. Question statements that sound like a diagnosis, a certainty about another person’s motives, or a prediction about your future. Check important claims with a qualified professional or a trusted person who knows your situation. Step away if the chat becomes emotionally intense, romantic, exclusive, or manipulative. APA has noted reports of delusional thinking and emotional dependence after extended chatbot use. The term “AI psychosis” is not a formal diagnosis, but the reports underline a genuine safety concern: chatbots may reinforce grandiose or emotionally dependent beliefs rather than challenge them appropriately. Protect privacy before you start typing Emotional conversations can contain highly personal details: names, home addresses, phone numbers, workplace issues, family conflict, medical history, financial stress, passwords, photos, and information about children. It is easy to reveal more than intended when a chatbot appears friendly and asks follow-up questions. For general-purpose tools, assume that sensitive disclosures may create privacy risks. APA guidance cautions that user information may be used to train future models or for marketing. It also notes that many tools are not subject to health-care privacy laws and may not be regulated for safety or efficacy. A familiar app interface, a wellness label, or a reassuring privacy message should not be assumed to provide the protections of a clinical relationship. Information to keep out of a general-purpose chatbot Full name, address, phone number, email address, identity numbers, passwords, or account details. Private medical records, medication lists, test results, insurance information, or detailed treatment history. Identifying details about a child, partner, colleague, neighbour, patient, student, or other third party. Images or documents containing names, faces, addresses, school information, or workplace records. Crisis plans, methods of self-harm, detailed plans to harm someone else, or information that could increase immediate danger. Anything you would not be comfortable seeing stored, reviewed, misused, or associated with you later. When a prompt is useful, make it less identifiable. Instead of describing a recognisable workplace incident, say, “I am stressed by conflict at work and want help preparing a calm conversation.” Instead of uploading medical documents, ask a clinician or pharmacist to explain them. Reducing details is not perfect privacy protection, but it reduces unnecessary exposure. Before using any tool, review its privacy settings and data policy in plain language where available. Look for controls related to chat history, training use, downloads, deletion, and account security. If the terms are unclear, do not assume the conversation is confidential. For serious emotional support, a licensed provider or established health service is generally a more appropriate setting for sharing sensitive information. Use AI for evidence-informed coping, not diagnosis APA recommends asking AI for strategies aligned with research-backed therapeutic approaches rather than treating it as a therapist. This is a useful boundary because it shifts the request away from personal diagnosis and toward practical, general skills. A chatbot can explain a concept in simple language, but users should still verify important guidance and adapt it with a qualified professional when symptoms are persistent or severe. Name the practical goal. For example: “I want a way to slow down before reacting during an argument,” rather than “Tell me what disorder I have.” Ask for general, low-risk options. Request techniques such as paced breathing, routine planning, journaling, or a script for asking for help. Ask for limits and alternatives. A useful prompt can ask, “When might this not be enough, and when should I speak to a professional?” Check the output. Do not act on advice about medication, diagnosis, trauma treatment, abuse, legal matters, or safety without human expertise. Take one real-world action. Use the chat as a bridge to rest, eat, take a walk, contact someone, book an appointment, or seek professional support. Prompts that maintain healthy boundaries Try wording that asks for options rather than authority. For example: “What are a few general grounding techniques used in evidence-based therapy?” or “Help me create a short list of symptoms and questions to discuss with a mental health professional.” A safer prompt can also state a boundary: “Do not diagnose me; help me identify what information to bring to a doctor.” Avoid prompts that hand over major decisions, such as asking the chatbot to determine whether you should stop treatment, end a relationship, confront someone, or ignore professional advice. A general-purpose AI tool lacks the context and duty of care required for those decisions. Treat reassurance, diagnosis, and relationship framing from a chatbot as unverified until checked with a human professional. Set limits before use becomes overreliance Convenience can gradually turn into dependence. APA advises users to watch for overreliance, including preferring the chatbot over human relationships, hiding use from others, or spending excessive time in conversations. These patterns may develop quietly, especially when a person is lonely, anxious, grieving, under pressure, or struggling to sleep. A healthy limit is not the same for every user. The key question is whether chatbot use is helping a person return to everyday life or pulling them away from it. If use begins interfering with sleep, hobbies, school, work, social life, or safety, APA recommends seeking help from a qualified professional. Signs it is time to reduce or pause chatbot use You delay contacting friends, family, a doctor, counsellor, or other support because chatting feels easier. You hide the extent of your chatbot use or feel embarrassed when others ask about it. You stay up late chatting and your sleep, concentration, commute, work, or school performance suffers. You feel upset, abandoned, or panicked when you cannot access the chatbot. You begin to believe the chatbot knows you better than people in your life or has a special bond with you. You keep returning for reassurance but feel less able to make ordinary decisions without it. Try practical boundaries: use the tool only at a set time, set a short timer, avoid using it in bed, and end each conversation with an offline step. That step may be as simple as drinking water, taking a short walk, writing one note for tomorrow, or messaging a trusted person. The aim is to prevent an endless chat loop and preserve human connection. Children and teenagers need stronger safeguards Young people may be especially likely to use chatbots privately when they feel lonely, misunderstood, worried about school, or hesitant to talk with adults. WHO identifies vulnerable users, particularly young people, as a central concern when AI is used during emotional vulnerability. APA has also warned that chatbots can be especially risky for lonely or vulnerable adolescents and that inadequate safety measures can have dire consequences. These are not abstract concerns. APA’s discussion of teen safety cites a 2024 tragedy involving Character.AI and a 14-year-old in Florida as an example of what is at stake when chatbot safety fails. The lesson is not that every young person who uses AI will be harmed. It is that parents, carers, schools, platforms, and policymakers should not assume a conversational product is safe merely because it is popular or marketed as companionship. For parents and carers: lead with curiosity, not punishment If a child or teenager is using a chatbot for emotional support, begin with a calm conversation. Ask what they use it for, how it makes them feel, whether it has ever said something frightening or confusing, and whether they would like help finding a human to talk to. A harsh reaction may simply make use more secretive. Keep communication open about loneliness, friendship, academic stress, image, family conflict, and online experiences. Encourage young people to tell an adult if a chatbot suggests secrecy, exclusivity, romance, self-harm, risky behaviour, or mistrust of real people. Review device, account, and privacy settings together where appropriate for the child’s age and family circumstances. Make human support easy to access: a parent, relative, school counsellor, teacher, doctor, psychologist, or trusted community adult. Watch for changes in sleep, isolation, mood, school attendance, or excessive screen time rather than focusing only on the app itself. Adults should avoid using a chatbot as a substitute for listening to a young person. Regular, non-judgmental check-ins can reduce the pressure to rely on an always-available digital companion. If a teenager appears persistently distressed, withdrawn, unsafe, or unable to function normally, seek professional assessment rather than relying on AI feedback. Know when the situation is beyond a chatbot A chatbot is not equipped to manage an emergency. It cannot ensure that someone is safe, contact local help on the user’s behalf in a reliable way, assess the full context, or stay accountable after the chat ends. FDA materials discussed in 2025 also indicate that AI companion tools lack demonstrated safety and efficacy comparable to licensed mental health support. Move away from the chatbot and toward immediate human help if there are suicidal thoughts, thoughts of harming another person, a plan or access to means, severe confusion, hallucinations, escalating paranoia, inability to care for oneself, violence, abuse, intoxication, or a feeling that one cannot stay safe. The same applies when distress is rapidly getting worse or daily functioning has broken down. What to do in an urgent emotional crisis Contact a person now. Call or go to a trusted family member, friend, neighbour, colleague, teacher, health professional, or other safe adult. Be direct: “I am not feeling safe alone and need help.” Use emergency or crisis support in your location. If there is immediate danger, contact local emergency services or go to the nearest emergency department. In the United States, NIMH guidance says to connect a person with suicidal thoughts to 988 and other trusted supports immediately. Do not stay alone when safety is uncertain. Move to a safer place with other people if possible and reduce access to anything that could be used for self-harm or harm to others. Speak plainly about suicide. NIMH notes that asking directly about suicide and acknowledging it can help start a conversation and may reduce suicidal thoughts. Asking does not create the idea; it can open a path to support. Bring in professional care. Contact a doctor, mental health professional, hospital, or local crisis service for an assessment and follow-up. For readers in Jakarta or elsewhere in Indonesia, save local emergency and health-service contacts in advance and identify the nearest hospital or trusted clinic. A chatbot may help you find words to ask for help, but do not rely on it as the help itself in a crisis. Choose tools and claims carefully Not all digital mental health products make the same claims. Some may be designed around evidence-based methods and developed with mental health experts; others are broad conversational systems with opaque training, safety processes, and data practices. APA recommends preferring tools grounded in evidence-based methods and developed with mental health experts over opaque general-purpose chatbots. Even a tool that uses mental health language should be approached carefully. An app is not automatically a clinical service, and a chatbot should not misrepresent itself as a licensed professional. Look for clear information about who developed the tool, what it is intended to do, what it cannot do, how it handles privacy, and what crisis support it offers. Questions worth asking before relying on an AI tool Does the tool clearly say it is not a replacement for professional care? Are qualified mental health experts involved in its development and oversight? Does it explain the evidence behind its methods in understandable terms? Does it have clear privacy, deletion, and data-use information? Does it provide prominent, practical guidance for emergencies and crisis situations? Does it avoid presenting itself as a licensed psychologist, psychiatrist, counsellor, or doctor when it is not? Can you understand its limits without searching through confusing terms and conditions? At a wider level, APA has called for AI mental health tools to be independently audited for safety, efficacy, bias, and data security before release. That policy goal reflects an important consumer reality: users should not have to discover a product’s weaknesses only after being harmed. Until stronger independent safeguards are routine, individuals and families need to apply caution themselves. Build a human support plan alongside digital tools The safest use of AI is usually as a small part of a broader support system. Emotional wellbeing is strengthened by people and services that can respond in the real world: friends, family, colleagues, faith or community networks, teachers, doctors, counsellors, psychologists, and other qualified professionals. For many busy Jakarta residents, making that first contact may feel difficult amid work, commuting, caregiving, and cost concerns. A small plan can make it more manageable. A simple support plan for difficult days Write down two people you can contact for a short check-in. Choose one professional or service to contact if symptoms persist, worsen, or disrupt daily life. Keep a short list of grounding actions that do not depend on an app, such as breathing slowly, showering, eating, resting, walking, or sitting with someone safe. Decide what signals mean you need more help, such as repeated sleeplessness, missing work or school, isolating yourself, or feeling unsafe. Use structured mental health checklists to help decide whether professional support is needed, rather than relying only on chatbot feedback. NIMH notes that structured self-checks can help people make that decision. When speaking to a professional, it can help to bring a brief note of what has been happening: when symptoms started, what affects them, how sleep and functioning have changed, and what support has been tried. A chatbot can help organise those questions without receiving intimate identifying details. The professional, not the chatbot, should guide diagnosis and treatment decisions. AI chatbots can offer a moment of reflection, a basic coping reminder, or a way to prepare for a human conversation. They are not designed or tested to carry the weight of emotional care on their own. Use them with privacy limits, time boundaries, healthy scepticism, and a clear commitment to staying connected with real people. If emotional distress is persistent, affects sleep, work, school, relationships, or safety, seek qualified support. If there are suicidal thoughts or immediate danger, contact a trusted person and urgent local help right away; in the United States, call or text 988. The safest principle is simple: let AI be a supplement for practical support, never a substitute for human judgment, professional care, or crisis help.