You know the message because you've probably typed it. It's 1am, the room is dark except for your phone, and you open a chatbot and write the sentence you haven't said to a single human being: "I don't know why I feel like this."
Turns out you weren't alone in that. Not even close.
In early December, EdSource reported on a new study finding that roughly one in eight American adolescents and young adults now use AI chatbots for mental health advice. One in eight. That's not a fringe behavior anymore. That's a couple of people in every group chat you're in. Millions of quiet 1am conversations that nobody mentions at brunch.
My first reaction to the number wasn't shock. It was recognition. Of course it's one in eight. It might be higher by the time you read this.
why this was always going to happen
The backdrop matters. Researchers writing in Health Affairs Scholar have been tracking mental health crises and help-seeking among US adults through 2024 and 2025: demand for support is high, and people are actively looking for it. And anyone who's tried to book a first therapy appointment knows the friction on the human side. Waitlists. Cost. The sheer effort of asking a person for anything. The JED Foundation flagged youth mental health as one of the defining concerns entering 2025, before anyone was counting chatbot confessions.
So you have a generation that grew up online, feels worse than it wants to, and can't easily get an appointment. Then someone hands them a free, instant, judgment-free text box that answers at any hour.
One in eight is what happens next. It was never going to be zero.
The adults in the room are, predictably, alarmed. And the alarm has real institutional history behind it. The Surgeon General already issued a formal advisory on social media and youth mental health, which means public health authorities officially treat young people's relationship with technology as a safety issue, not a lifestyle choice. Congress is circling too: the Kids Off Social Media Act is sitting in the Senate right now. The regulatory mood around young minds and screens is the most serious it's been in a decade, and a study saying an eighth of young people take mental health advice from software lands directly on that nerve.
I get the worry. I even share a version of it. But I think most of the discourse is aimed at the wrong target.
the question nobody's asking the one in eight
Here's what I'd want to know about those millions of late-night conversations, and what almost no headline asks: what were people actually hoping to get back?
The framing, in the coverage and in nearly every panicked take on it, is about advice. AI dispensing guidance. A language model playing therapist, handing out coping strategies it can't be accountable for. That's a legitimate worry. A chatbot will confidently produce a five-step plan for your grief because producing confident answers is what it's built to do.
But think back to your own 1am message. Were you asking for a plan?
I don't think most people were. I think most people typed "I don't know why I feel like this" for the same reason you'd say it to a friend on a long drive: not to be solved, but to be heard. To have the thing exist outside your own head for once. The advice that came back was, in a lot of cases, an answer to a question nobody asked.
We all know the human version of this. The friend who, four seconds into your rough week, says "have you tried going to bed earlier?" They mean well. You leave the conversation lonelier than you entered it. What we've done, at national scale, is build that friend into every phone and then act surprised that young people are confiding in it.
So my honest take on the study is this: the problem isn't that one in eight young people brought AI into their emotional lives. That ship has sailed, and shaming people for boarding it helps nobody. The problem is that the default AI they reached for is stuck in fix-it mode. It advises because advising is its whole personality. The distinction that decides whether this trend is help or harm isn't AI versus no AI. It's being advised versus being listened to, and almost everything on the market only knows how to do the first one.
This is, I'll admit, exactly the hill my colleagues and I chose to build on. It's why we built Joice the way we did: you talk out loud, and the AI's only job is to listen like a good friend and ask the next gentle question. No unsolicited five-step plans. It's a bet that what most people need at 1am isn't a strategy but a witness.
There's decent evidence that how we use our devices matters more than the raw fact of using them. A randomized controlled trial published in BMC Medicine this year found that reducing smartphone screen time improved participants' mental health. That sounds like an argument against phones until you look at what it's really measuring: the compulsive, ambient, default way we use these things. Deliberate use is a different animal. Ten minutes of intentionally saying what's on your mind is not the same act as two hours of reflexive scrolling, even though both happen on the same slab of glass.
try this before you ask anything for advice
So what do you actually do with this study, besides forward the "1 in 8" stat to your group chat?
A small experiment. When the 1am pull hits and you're about to ask something for advice, whether it's a chatbot, a friend, or the void of a search bar, do one thing first: name the feeling out loud, once, with no fix attached. Not "how do I stop feeling anxious" but just "I'm anxious and I don't know why." Say it to your ceiling if you have to.
Then notice what happens. Sometimes the urge to be solved deflates the moment the thing is spoken, because being solved was never the actual craving. Being heard was, and it turns out you can do a surprising amount of the hearing yourself.
And if you're one of the one in eight, no judgment from me. You saw a door that was open at 1am and you walked through it, which is more help-seeking than a lot of people ever manage. Just check, now and then, whether what's on the other side of that door is listening to you or merely answering you.
There's a difference. Your 1am self already knows what it feels like.