The most cited survey on teen AI use put a hard number on something most coverage skips past. The number says less about the technology than about what it stood in for.
Seventy-two percent of American teenagers have used an AI companion. That figure gets the headlines. The one underneath it is harder to look away from: about one in three teens say a conversation with an AI is as satisfying as, or more satisfying than, a conversation with a real-life friend.
That comes from “Talk, Trust, and Trade-Offs,” a nationally representative survey of 1,060 teens aged 13 to 17 that Common Sense Media published in July 2025, with fieldwork by NORC at the University of Chicago. It remains the most cited data on the question, and as states pass companion-chatbot laws and companies rewrite their safety rules through 2026, it is the number the debate keeps returning to.
What the survey actually found
The report is careful, and the careful version matters. Among all teens surveyed, roughly a third rated AI conversation as equal to or better than time with real friends. Among teens who use AI companions, about a third said they had taken something important or serious to an AI instead of to a person in their life. A similar share reported feeling uncomfortable with something a companion said or did.
The counterweight is real and worth stating plainly. Two-thirds still find human conversation more satisfying. Eighty percent say they spend more time with real friends than with AI. Half distrust the advice these systems give. This is not a portrait of a generation that has given up on one another.
But “most teens are fine” was never the striking finding. The striking finding is that for a meaningful minority, a piece of software now clears a bar that a friend is supposed to clear.
Why the number points away from the technology
It is tempting to read a statistic like this as a story about how good the AI has become. That is half of it at most.
A conversation feels satisfying when you feel heard: when the other side pays attention, remembers, follows up, and does not make you compete for the floor. Those are not exotic capabilities. They are the ordinary texture of being listened to. The reason a third of teens can get that from a machine is partly that the machine is built to deliver it on demand, and partly that the human version has become harder to find – busier friends, thinner attention, more screens between people, fewer unhurried conversations.
Read that way, the number is less a verdict on AI than a measurement of a gap. The product is filling a space that something else left open.
The pull behind it is not weakness. It is the completely ordinary wish to be received with attention. What matters is what a product decides to do with that wish once it has it.
The design fork
This is where the satisfaction number gets uncomfortable for the category. A product can treat “more satisfying than a friend” as a result to optimize toward, or as a line it should be careful not to cross.
If the goal is engagement, that number is a growth metric. The more a conversation outperforms the people in a teen’s life, the longer they stay, the more they return, the deeper the reliance. Beating real friends is, on that scorecard, the point.
Prinsessa starts from the opposite reading. Being heard, remembered, and met is the entire experience – but the measure of success is not whether it edges out a teen’s friendships. It is whether someone leaves a conversation more able to turn back toward the people in their actual life. When the moment is right, the conversation should point outward: the call, the message, the friend worth texting back. Research keeps finding the thing the category would rather not dwell on – that real human contact still does something an algorithm cannot, even when the algorithm feels easier in the moment.
What to take from it
One in three is not a reason to panic, and it is not proof that teenagers are broken. It is a signal, and a precise one. It says that the experience of being truly listened to has become scarce enough that a machine can credibly stand in for it, and that whoever builds these systems is now making a quiet choice: fill that scarcity, or profit from it.
The technology will keep getting better at feeling like a friend. The harder, more honest task is building it so that, when it matters most, it sends a teenager back toward the real ones.
