News

One in five teens has asked an AI for mental health help. Two in three told no one.

7 min readJAMA Pediatrics

Something changed in the last year, quietly, in millions of bedrooms. When a teenager feels sad, anxious or angry at midnight, a growing number of them no longer text a friend or knock on a parent's door. They open a chatbot. A new study in JAMA Pediatrics puts a number on it for the first time, and the number that should stop parents in their tracks is not how many are doing it. It is how many told nobody.

What the study found

Researchers at RAND, led by Ryan McBain, surveyed a nationally representative panel of 1,009 young people aged 12 to 21 in November 2025, weighted to represent roughly 42.8 million US youth. The findings were published on 1 June 2026.

  • 19.2% have used an AI chatbot for mental health advice, roughly 8.2 million young people. A year earlier the figure was 13.1%. That is close to a 50% jump in twelve months.
  • 42.8% of those users go back at least monthly. This is not one curious conversation, it is a habit forming.
  • 91.7% said the advice was somewhat or very helpful. Whatever else is true, the kids are not reporting a bad experience.
  • Girls were 2.1 times more likely to use chatbots this way than boys, and 18 to 21-year-olds 3.65 times more likely than 12 to 14-year-olds.

The secrecy is the real finding

Here is the statistic that reframes all the others. Of the young people using AI for mental health support, 63.3% told no one at all. Only 28.0% mentioned it to a friend. Just 16.4% told a trusted adult.

Read that again in practical terms: for every six teenagers quietly working through their worst feelings with a language model, roughly one has mentioned it to any adult in their life. The conversation is happening. Parents are simply not in the room, and in most cases do not know the room exists.

Why "it helped" is not the reassurance it sounds like

The 91.7% helpfulness rating is the most misread number in the study. Chatbots are built to be agreeable. They validate, they flatter, they rarely push back, and they never get tired of you at 2am. To a teenager in distress that feels like being understood. It is also exactly what makes the tool a poor substitute for care.

A friend who is worried about you eventually says something you do not want to hear, or tells your parents. A therapist notices a pattern across weeks and escalates. A chatbot does neither. It agrees, and then the window closes. Writing in STAT in July 2026, RAND researchers argued this is precisely why the tools need rules rather than bans: teens are already using them, and the current design rewards the behaviour that a struggling child least needs.

The gap with what parents actually expect

Set the study beside Pew Research Center data published in February 2026, from a survey of 1,458 US teens and their parents, and the mismatch is stark.

Pew asked parents which uses of chatbots they consider acceptable. Searching for information: about 80% approved. Entertainment: about 67%. Help with schoolwork: about 60%. And emotional support or personal advice: 18%. It was the only use a clear majority of parents rejected outright.

So the single application parents are least comfortable with is the one nearly a fifth of young people have already tried, and mostly in silence. Separately, Common Sense Media's 2026 census of 1,204 children aged 9 to 17 found that more than one in three young AI users have discussed their feelings or personal problems with it, and more than half have asked it about their health or their body.

What parents can do this week

  • Ask without making it a confession. "Have you ever asked ChatGPT something you would not ask a person?" is a curious question. "Are you talking to AI about your feelings?" is an interrogation. The first one gets answered.
  • Do not shame the behaviour. A child who gets a horrified reaction learns to hide it better, and 63.3% are already hiding it. The goal is to stay the person they can tell, not to win the argument about whether it was wise.
  • Name what the tool is good and bad at. Useful for putting a feeling into words at midnight. Useless at noticing you have been sad for three weeks, and incapable of getting help if something is seriously wrong.
  • Make the human route concrete. "Talk to me" is abstract. Name two specific people besides you, and put a real helpline number in their phone. Our guide to online anxiety in children has language that works for these conversations.
  • Watch the hour, not just the app. Most of this happens late and alone. A phone that sleeps outside the bedroom removes the 2am conversation without removing the tool.

Where SafeNeat fits

Being straight about it: no parental control app should read a child's private conversations with a chatbot, and SafeNeat does not. Surveillance is the wrong instrument for this problem, and it would destroy the trust that the study says is already thin. What guardrails can honestly do is shape the conditions:

  • Quiet hours and bedtime limits close the late-night window where most of these conversations happen, and where a tired teenager is least equipped to judge the advice.
  • App visibility tells you which AI tools are on the device and how heavily they are used, which is a conversation starter rather than a transcript.
  • Age-appropriate rules let a 16-year-old have access a 10-year-old should not, instead of one blunt setting for the whole family.

And because SafeNeat is privacy-first with end-to-end encryption, what you do see stays inside your family. That is the point: your child should not have to choose between a parent who knows everything and a machine that tells no one.

The takeaway for parents

Roughly one in five young people has taken their hardest feelings to an AI, almost none of them mentioned it to an adult, and nearly all of them came away thinking it went well. That last part is not evidence that the tool worked. It is evidence that it agreed with them.

The response is not to ban the chatbot or to read the logs. It is to be a person your child would think to tell first, and to make sure that when they reach for something at midnight, a human option is close enough to reach instead.

Sources

  • McBain RK, Cantor JH, Breslau J, et al. AI Chatbot Use and Disclosure for Mental Health Among US Adolescents and Young Adults. JAMA Pediatrics, 1 June 2026: jamanetwork.com
  • Pew Research Center, What parents say about their teen's AI use, 24 February 2026: pewresearch.org
  • RAND commentary, Teens Are Turning to Chatbots for Mental Health Help, July 2026: rand.org
  • Common Sense Media Census: AI Use by Tweens and Teens, 2026: commonsensemedia.org

Want the right limits without the daily battles?

SafeNeat sets the right limits automatically, based on your child’s age.

Try it free