In 2025, a nationally representative survey of 1,060 US teens found that 72% had tried an AI companion at least once and 52% used one regularly. Buried inside those headline numbers is the detail that unsettled the researchers: a third of the teens already using these tools had turned to a bot, not a person, for a serious conversation.
The survey was published by Common Sense Media in a report called “Talk, Trust, and Trade-offs: How and Why Teens Use AI Companions,” based on responses from 1,060 US teens aged 13 to 17. By “AI companion” the report means apps built to act as a digital friend, like Character.AI or Replika, though teens often use general tools the same way. The lead author, Michael Robb, was candid that the scale caught the team off guard. He told CBS News that before doing the research, the team “had no understanding of how many kids are actually using AI companions.”
A note before going further. I am not a psychologist, a therapist, or a researcher of any kind. This is one curious adult reading a survey and thinking out loud. The numbers come from a single nationally representative study and a few experts reflecting on it, which is to say they describe patterns across groups of people, not rules about your kid or mine.
What I find steadying, before we get to the part that worries people, is how skeptical these same teens turned out to be. Half of them said they distrust the advice these systems give, and among the teens who do use AI companions, 80% still spend more time with real friends than with a bot. This does not seem to be a generation that has confused a machine for a person.
The number that should give us pause
Among the teens already using these tools, a third had discussed something serious or important with an AI companion instead of with a real person. Not asked it to draft an email. Actually confided in it. And 31% found those conversations as satisfying, or more so, than talking with actual friends.
The simplest explanation is perhaps the most uncomfortable one. A bot does not judge you and never tires of you, and it does not make your problem about itself. As Megan Ice, a psychologist at the Child Mind Institute, told the Institute’s own publication that an AI “friend” who won’t judge a teenager is “uniquely enticing for this population.”
Her colleague Dave Anderson framed the other half of it, which is less about the bot and more about what teenagers say is missing. Research, he noted in the same piece, keeps finding that teens cannot point to a single nonjudgmental adult in their lives. Not a teacher, a counselor, a coach, or a parent who they feel will simply listen. Put that next to the “serious conversation” stat and it stops looking like a story about clever apps. It starts looking like a story about a shortage of patient, unhurried listeners.
Where this leaves the rest of us
Robb’s own preference is modest and clear. He told CNN he doesn’t want kids turning to AI for serious conversations “in lieu of a friend, a parent or a qualified professional.” Common Sense CEO James Steyer put the surrounding context in starker terms, calling these tools something arriving when, in his words, “kids and teens have never felt more alone.” That is an advocate’s framing rather than a measured finding, and I would not treat teen loneliness being at an all-time high as settled. But the direction he is pointing in is hard to dismiss.
So here is the question I cannot shake. If a third of the teens already using these tools find a machine easier to confide in than the people around them, what does that say about the people around them? Not about the cleverness of the apps, not about the design of the chatbots, but about the rooms these kids walk into every day and the listening they do not find there.
A bot that never judges and never makes it about itself is a very low bar for a generation to prefer over us. The uncomfortable possibility is not that the machines have grown too good. It is that we have grown too distracted, too hurried, too busy with our own noise, to be the patient listener a teenager would otherwise choose first.
