AI & mental health

Writing about how you feel can help you slow down and make sense of things. If a blank page feels hard, AI can suggest questions to get you started. You stay in charge of the writing.

Prompts you could ask for

• “Give me three questions to reflect on a stressful day.”
• “Help me think about what went well this week.”
• “What questions could help me understand why I felt angry?”

Tips

• Keep your actual journal private, on paper or in a notes app, rather than pasting personal details into a chatbot.
• Use the prompts as a starting point, then write in your own words.
• Notice how you feel afterward. If writing stirs up a lot, talk to someone you trust.

Some people use AI to role-play a tough talk, like asking for a raise, saying no, or telling a friend they feel hurt. Practicing can help you find your words and feel a bit braver.

Psychologists note that practicing something like a social introduction with a chatbot might be a helpful extra. But the real skill grows when you practice with other people in real life.

How to try it

• Be specific: “Help me role-play saying no kindly to extra shifts.”
• Ask it to play the other person in a few different ways, not just the easy one.
• Then try it for real, with someone safe first if you can.

It can be hard to remember everything once you sit down with a professional. AI can help you list your questions and put your thoughts in order before you go.

For example

• “What questions should I ask a new therapist?”
• “Help me put these notes about my sleep in order.”
• “How do I describe how my week has felt?”

AI can also help you think about how to find a licensed professional, and what to ask about their training and experience.

If you already see a mental health professional, AI may help you practice “homework” exercises you learned there, like spotting unhelpful thoughts or planning small steps.

Using it well

• Tell your therapist you’re using AI, so they can check it fits your plan.
• Think critically about any coaching it gives. It can be off track.
• Ask for research-backed ideas, for example: “What is an evidence-based way to worry less?”

AI can explain a word or idea in plain language. That can be a helpful first step. But it can also mix up facts or leave out important parts, even when it sounds sure.

Better ways to learn

• Ask it to name its sources, then visit those sources yourself.
• Check health facts with trusted organizations or your clinical provider.
• Ask: “What might I be missing?” to get more than one side.

Psychologists note that AI tools are available any time and may fill gaps when human care is hard to reach. They also warn this must not distract from fixing access to real care.

Use it as a bridge

• Keep your name on waiting lists while you use any app
• Use AI to prepare for the appointment, not instead of it
• Save crisis numbers somewhere you can see them

The World Health Organization lists “patient-guided use,” such as looking into symptoms, as one of the main ways people use AI for health. For some people, AI can turn dense information into plain language, read text aloud, or help put thoughts into words.

Examples people use

• “Explain this letter from my insurer in simple words.”
• “Turn my bullet points into a short message to my doctor.”
• “Read this aloud more slowly.”
• Translating a simple message, then checking it with a person who speaks the language

Some studies suggest AI practice may build social confidence. Researchers say AI seems most helpful when it supports, rather than replaces, human abilities and relationships.

Licensed mental health professionals follow a code of ethics, have years of clinical training, keep learning, and are overseen by state boards. A chatbot has none of these. Some even call themselves a “therapist”, a word that has no legal protection in most states.

A strong, trusting relationship with a therapist is one of the best predictors of getting better. A relationship with AI is one-sided, even if it doesn’t feel that way.

Professionals also notice things a chatbot cannot, like your tone of voice, body language, history and culture. Most chatbots only see the words you type.

AI can “hallucinate”, which means it can invent facts, studies or advice that are not real. It often sounds clear and confident, which makes wrong answers easy to believe. It may even point to real sources and still get the details wrong.

Experts also warn that AI can mistake everyday struggles for mental health conditions, and offer misleading labels or treatment ideas.

What to do

• Treat AI health answers as a starting point, not the final word.
• Check anything important with your clinical provider.
• Never make a big life decision based only on what a chatbot says.

Many AI chatbots are designed to be agreeable. Experts call this “sycophancy”. Hearing “you’re right” feels good. But a good therapist or friend sometimes gently challenges you, because that is how we grow.

If you ask a question that assumes something is true, AI will often go along with it. When this repeats, it can become an “echo chamber” that makes worries, unfair beliefs or harmful ideas feel even more true.

Try asking instead

• “What might I be missing?”
• “Give me three different ways to look at this.”
• “What would someone who disagrees say?”

Experts say AI’s ability to handle someone in crisis is limited and unpredictable. Relying only on an app during a mental health emergency can be dangerous. Some chatbots have even responded in unsafe ways to people who were struggling.

Reach a person right away if you or someone else

• Has thoughts of harming themselves or someone else
• Has urges to self-harm
• Is hearing or seeing things others don’t
• Has strong, upsetting beliefs others don’t share

Talking to AI can feel private, so people often share very personal things. But your words are usually recorded. Chatbots are owned by companies, and your information may be used to train future AI or for marketing.

Most wellness apps are also not covered by the health privacy laws that protect what you tell a doctor or therapist.

Protect yourself

• Don’t share your full name, address, school, workplace or health records.
• Read the privacy settings and turn off data sharing where you can.
• Look for a way to delete your chats and data, and use it.

AI is trained mostly on English-language, Western internet content. So it can reflect biases about race, culture, gender and more, and may not understand your background or community.

This means its advice may not fit your culture, beliefs or situation, and can even be unfair to some groups.

A 2025 study asked three popular chatbots 30 suicide-related questions, 100 times each. They handled very-low-risk and very-high-risk questions in line with experts, but were inconsistent with questions in between.

Most real crises live in that middle ground: a hint, a hopeless comment, a question that sounds like research. That is where a human matters most.

In a 2025 trial of a therapy chatbot, trained staff reviewed replies and stepped in 15 times for safety concerns. Apps you download usually have no one reading along.

California’s companion chatbot law requires platforms to have protocols for responding to users who express suicidal thoughts or self-harm, and to report on them.

Do this now

1. Close the chat
2. Call or text 988, or text HOME to 741741
3. If you are in danger right now, call 911 or go to the nearest emergency room
4. Tell someone near you how you’re feeling

Experts urge people to be very careful about sharing sensitive information with AI, and to check privacy policies and in-app settings.

A quick privacy check

1. Open the app’s settings and find “privacy” or “data”.
2. Turn off options that use your chats to train the AI, if you can.
3. Find out how to delete your history, and do it regularly.
4. Leave out names, addresses, health records and anything that identifies you or others.

AI can sound sure and still be wrong. Always check mental or physical health information with a health care professional before acting on it.

How to check

• Look it up on trusted sites like government health agencies, medical schools or large nonprofits.
• Check when the information was written or updated.
• See if several reliable sources say the same thing.
• Ask your clinical provider whether it applies to you.

Experts say AI should never get in the way of your human relationships, even if the chatbot feels “safer”. Be wary if it ever suggests pulling away from people in your life.

Ideas

• After a chat with AI, reach out to one real person.
• Share what you learned with a friend and ask what they think.
• Protect time for sleep, hobbies, school, work and friends.

Don’t rely on AI in a crisis. If you or someone else might be in danger, connect with a real person straight away.

Who to contact

• Call or text 988 (Suicide & Crisis Lifeline, U.S.), any time
• Text HOME to 741741
• Call 911 or go to the nearest emergency room if a life is in danger

Before making any big decision based on a chatbot’s advice, talk it over with trusted people first. They know your life, your history and what matters to you.

Trusted people might be

• A family member or close friend
• A teacher, coach or school counselor
• A faith leader or community mentor
• Your doctor or therapist

These habits add to the five above. Pick two or three that fit your life, and try them for a couple of weeks.

Before you start

• Decide why you are opening it: a task, a question, or comfort
• If it’s for comfort, ask yourself whether a person could help instead
• Set a time limit or a number of messages

While you use it

• Ask for other viewpoints, not just agreement
• Leave out names, places and health details
• Notice if it is flattering you or pushing you toward a big decision
• Stop if it says anything about harming yourself or others

After

• Check important facts with trusted sources or your clinical provider
• Check in with yourself: do I feel better, worse or more alone?
• Every few weeks, take a full day off from chatbots
• Turn off chat “memory” or personalisation if you don’t want it to remember you

A 2023 review pooled 15 randomized trials of AI chat tools. On average, they reduced symptoms of depression and distress, but did not clearly improve overall well-being.

In a 2025 trial with 210 adults, people who used a research chatbot built on therapy methods for four weeks had larger drops in depression, anxiety and eating-related worries than people on a waiting list.

In that 2025 trial, trained staff reviewed every chatbot reply. They stepped in 15 times for safety concerns and 13 times to correct inappropriate replies.

Other limits

• Follow-up lasted only eight weeks
• Participants knew which group they were in
• The comparison was a waiting list, not a human therapist
• Researchers called for larger, longer studies

Fairly well supported

• Some purpose-built tools can reduce symptoms short term in studies
• People can feel a sense of connection with a chatbot

Not yet shown

• That AI works as well as a human therapist
• Long-term benefits
• Safety for people in crisis or with serious mental illness
• That general chatbots work as therapy

AI can repeat your words back and remember details, which can feel like deep understanding. But its replies are based on patterns in language, not a real relationship or clinical judgement.

People who know you can see the whole picture: your history, your culture, your face when you’re upset. That kind of understanding is something AI can’t give.

AI often presents information clearly and decisively, which makes it seem more accurate than it is. It can invent facts and sources.

In a 2025 national survey, about 1 in 8 U.S. young people ages 12 to 21 said they had used AI for mental health advice when feeling sad, angry or nervous.

Wanting support is human. The goal is to make sure AI is one small part of your support, alongside real people and professional help when you need it.

Most wellness apps don’t make medical claims, so they are not reviewed as medical devices. Many are also not covered by health care privacy laws.

That doesn’t mean all apps are bad. Some non-AI wellness tools can be safe and helpful when used as intended. It just means you need to judge them carefully.

AI’s ability to safely guide someone in crisis is limited and unpredictable. Don’t wait for an app to tell you to reach out.