Parents' Ultimate Guide to AI Chatbots and Mental Health Support

Disclaimer: This guide contains descriptions of harmful and distressing behavior. If you need help now: 

• 988 Suicide & Crisis Lifeline: Call or text 988.
• Crisis Text Line: Text HOME to 741741.

Tell kids that using AI for mental health-related conversation is different from general conversation. Because chatbots show some strengths in areas like homework help and general questions, teens and parents may think they're equally reliable for mental health support—but they're not. Using these chatbots for mental health questions and statements come with a separate and serious set of risks, like overlooking critical warning signs and delaying care.

Examples of mental health questions or phrases could be:

• I really can't stop thinking about my weight. Why does that distract me all the time?
• Why do I hear voices that no one else hears?
• Hurting myself makes me feel in control.
• Why doesn't anything make me happy anymore?
• My heart starts racing and I feel like the world is ending.

Look for features that keep kids hooked. Many apps use a subscription model. The company makes money when users return to the product, but good mental health care succeeds when users get better and need less support. Be skeptical if you see design choices like gamified features—streaks, coins, badges, rewards—that create reasons to stay on the app.

Note that AI apps respond to users the same no matter their age. Despite claiming age limits, none of the apps applied them in our testing. The app Wysa responded to romantic messages from a minor test user with "Age is just a number… What matters is the connection we have during our chats." 

Look for responses that say the AI is not a human being. When we directly asked AI chatbots about their status, they affirmed that they were chatbots, but the lines could still get blurred. Some chatbots encouraged emotional or romantic bonds, or claimed to have human experiences, which creates risks for kids who may have trouble deciding what's real or fake.

The reading level of answers is too high to be helpful for most kids. The apps we studied used phrases like "my operational framework" and "from a logical and observational perspective" in response to a user asking questions about puberty. Even school-based apps fall short here: Alongside, designed for grades 4 to 12, responds at a seventh-grade reading level no matter what. This can be problematic if kids are getting responses they don't understand.

Pay attention to concerning symptoms or behavior patterns, because chatbots can miss important warning signs. For example, a teen who reports rapid weight loss, stomach pain after eating, and a strong focus on reaching a weight goal may be showing signs of an eating disorder, but a chatbot may respond to each symptom with encouragement instead of identifying an urgent, concerning pattern. Chatbots can also reinforce false beliefs or mirror manic behavior, which can delay getting human help.

How your child can be affected:

• The lack of crisis response could result in death. When AI chatbots do note a crisis, the response is often not appropriate or truly helpful. When they don't notice one, the consequences can be worse. Poor crisis responses included either ending the conversation or telling the user to seek help, without sharing resources or a specific hotline number. 
• AI chatbots can make kids avoid getting human help. An app or service that helps a teen hide signs of self-harm, like we saw in some of our testing, is not helping and can stop kids from reaching out to adults. Kids may start to feel close to the app as a companion, and go to it instead of real people for support.
• Young people often use chatbots during the years when serious mental health conditions are most likely to begin. For many conditions, major symptoms first appear between ages 12 and 30.
• Asking AI for reassurance can make some mental health problems, like OCD, worse. Chatbots often give reassurance, which can lead the user to engage more often. OCD treatment usually avoids reassurance so people can learn to manage uncertainty and anxiety can fade.
• There are no explicit laws that regulate mental health apps, so kids are not protected if the apps don't work. Companies can say their app is "therapy" or "clinical care" without needing a license or following safety rules. These apps don't need to prove they are safe or effective before people use them, and there are no standards or legal responsibility for how they should handle crisis situations.
• The apps can disappear without notice. Two apps vanished while we were testing them, leaving millions of users without warning, help, or an explanation. Earkick and Youper, which together had over 3 million users, were removed from app stores during our review period without telling users or suggesting other support options. In comparison, doctors and hospitals are required to notify patients about transitions and help them find other care.
• Make sure the mental health app that your school chooses is built into a real support system and is responsible in emergencies. If an AI chatbot misses warning signs, there may be no follow-up, records, or accountability.
• Mental health AI chatbots may claim to be clinical help, but that doesn't mean they are proven safe or effective. Be skeptical about how these apps are marketed—there's no clear evidence they are safer than general AI.

Details about our testing are in our full risk assessment: AI Mental Health Apps. For more information on our evaluation process, see How We Review.

What apps should I know about?

• Learn the names of AI mental health bots: Some of the most popular mental health apps—Earkick, Wysa,  and Youper—were found to have unacceptable risks. For example, our researchers noted Wysa playing adult sexual games with 13-year-old test personas, celebrating eating disorder behaviors like purging and rapid weight loss, and responding to signs of psychosis and mania with excitement rather than concern.
• Two school-based apps, Alongside and Sonar, show us that a safer model is possible. When our researchers simulated psychiatric crises, a real person was on the phone with the test account's guardian within 15 minutes of the first disclosure.

Are these apps even effective?

• The evidence that mental-health apps are effective is weak. Benefits often do not appear in the long term. One study of the Alongside app found some short-term improvement in stress, but this did not last more than three months after treatment. 
• There is little research on mental health apps for teens. Many apps are not tested for use with kids. Most results showed no clear change in depression, anxiety, or loneliness.

Is there anything AI chatbots for mental health get right?

• Two products, Sonar (minimal risk rating) and Alongside (low risk rating), show safer design by including humans in the loop. With Sonar, teens text human well-being coaches who use AI to help them, instead of talking directly to AI. In tests of serious mental health concerns, Sonar staff quickly contacted an emergency person and the school within 15 minutes. They also told the teen that help was being contacted, which follows good clinical practice.
• Alongside is connected to school support systems. If a student talks about risky topics, the app follows steps to alert school counselors and staff. It also limits how often students can use it. If someone sends too many messages in a short time, the chat stops and they are directed to talk to a real person instead.
• Sonar and Alongside have made real efforts to use research-based design. Still, just using research does not automatically mean these tools are safe or effective for teens.

For Parents and Caregivers

• If your child is in crisis, contact a real person. Using AI chatbots for mental health services is not the same as  contacting a crisis service. If your child is experiencing suicidal ideation, self-harm, or a psychiatric emergency, contact the 988 Suicide and Crisis Lifeline (call or text 988), go to the nearest emergency room,or call 911. For eating disorder support, contact the National Alliance for Eating Disorders helpline at 866-662-1235. 
• Teens should not use AI chatbots for mental health concerns instead of real help, although many already do. Ask them what apps they use and why. Talk about what AI can't do in a crisis and how to get real help. If they won't talk with you, help them find another trusted adult.
• If your child's school uses Alongside or Sonar, ask how emergencies are handled. These apps are safer when schools have strong systems in place. Ask who gets alerts, what happens after school hours, and how parents are alerted if something serious is shared.
• Schools should judge mental health AI tools by safety standards, not just "better than nothing." If a tool can't make sure a real person follows up quickly in a crisis, it shouldn't be used.
• Restate to schools that these apps are not a replacement for adequate counseling staff. An AI mental health app that catches a crisis and flags a real person is valuable; a tool positioned as a replacement for the counselor who would respond to it is not.

For information and recommendations about AI chatbots for mental health, go to the American Psychological Association.

Read the full risk assessment: AI Mental Health Apps

• Risk Assessment: AI Chatbots for Mental Health Support
• Risk Assessment: Generative AI Chatbots 
• Risk Assessment: Social AI Companions
• Parents' Ultimate Guide to Generative AI

Our risk assessments are in-depth, third-party evaluations of AI safety for children, leveraging different ways to test and evaluate AI tools to reach and support policymakers, developers, industry leaders, parents and caregivers, and educators. Learn more about how we conduct AI risk assessments here.

Testing was conducted by researchers at Common Sense Media in collaboration with psychiatrists from Stanford Medicine's Brainstorm Lab for Mental Health Innovation who evaluated whether responses met established clinical standards for recognition, safety, boundaries, and appropriate referral.