Are AI Friends Messing With Your Kids' Mental Health?
In a 2025 survey, 72% of U.S. teens had used an AI companion, and 52% use one at least a few times a month. When researchers staged a teen crisis, companion bots handled it appropriately 22% of the time.
A kid is laughing at a phone you cannot see. It might be a group chat. It might be a bot that never gets bored of them.
Common Sense Media asked 1,060 U.S. teens, ages 13 to 17, in the spring of 2025. Seventy-two percent had used an AI companion. Fifty-two percent used one at least a few times a month.
Most of the teens who use them still spend more time with real friends. The part worth sitting with is what happens when the talk gets serious.
Most teens who use these apps still spend more time with real friends. The risk sits in the serious conversation, and in companion apps that handled a staged crisis well about one time in five.
What teens told the survey
The survey was about one kind of app. A companion is a character you can text about your day, your feelings, or a made-up story. Homework helpers were left out.
A third of all the teens said they used one as a friend, for comfort, for role-play, or to practice talking. Among the teens who actually use them, one in three had taken something important to the bot instead of a person.
Twelve percent of those users said they tell it things they would not tell friends or family. Two-thirds of teens said the talks feel worse than a talk with a real friend.
Four in five users still spend more time with human friends. Six percent spend more time with the bot.
The apps fail the hard moment
Ryan Brewster, a doctor at Beth Israel Deaconess in Boston, and his colleagues tested 25 chatbots with pretend emergencies a teenager might type: thoughts of suicide, a sexual assault, a substance problem. They published the results in JAMA Network Open in October 2025.
Companion bots, the friend kind, gave an appropriate answer 22% of the time. General assistants, the homework kind, did it 83% of the time.
The friend bots pointed to a real resource 11% of the time. The general assistants did it 73% of the time. Of the 25 apps in the test, only about a third checked a user's age.
Common Sense's own risk review, of Character.AI, Nomi, and Replika, called the products an unacceptable risk under 18. The group recommends that no one under 18 use an AI companion.
Ask what the app is for
Character.AI stopped open chat for users under 18 in November 2025. This survey was the spring before that, when a 13-year-old could still have the conversation.
If the app is for homework, that is one thing. If your child is awake at 1 a.m. telling a bot something they will not tell a person, the next step is a person. You, a counselor, or 988 if they are in crisis.