What happens when AI is the first place children ask for help?
There is something that has been bothering me for a while now, because I know that I’ve been prone to this myself.
What does it mean when a child starts asking a chatbot for help before they ask a parent, a clinician, or any other trusted adult?
A new JAMA Pediatrics survey gives that question some hard numbers. In a nationally representative sample of US adolescents and young adults, 19.2% said they had used AI chatbots for mental health advice. Among those users, 63.3% said they had told nobody. Nearly half used them at least monthly, and 91.7% said the advice was somewhat or very helpful.
That last number is a serious concern.
Helpful output is one thing. Appropriate support is another. A system can feel reassuring, fast, and available at midnight while still being a poor place to lean on for emotional or psychological help. The JAMA authors make a similar point, noting that these tools are already part of young people's mental health information ecosystem and that adults around them need to understand that reality.
UNICEFs June 2026 brief, When AI becomes a friend, pushes the same conversation into rights and regulation. Its central argument is that the response needs to move from reactive to preventive, with clearer responsibilities across businesses, governments, parents, carers, and communities. That feels like the right framing to me. Once private reliance becomes normal, moderation alone is a very narrow answer.
AACAPs July 2025 guidance is useful here too. It flags overreliance, privacy, deepfakes, and unproven AI mental-health tools, but it also does something more practical. It treats AI as something that should support learning and life, while still needing boundaries, scepticism, and human relationships around it.
That is the design question I care about as a software founder.
For child-facing products, the safety bar is wider than content filtering. It includes whether the system clearly states what it is, whether it nudges a child away from dependence, whether it knows when to stop, and whether it can route a child towards real human support when the conversation needs that. Stanfords reporting on companion-style chatbots and young people shows why that matters, especially when systems are built to simulate intimacy and keep the interaction flowing rather than slow it down. Stanfords coverage of the Common Sense study is worth reading in full.
That concern reaches far beyond mental health apps. It touches games, chat features, open worlds, and any product where a child can ask privately for help.
If a child trusts a system enough to tell it something important, what should that system do next?
We'd love to hear your thoughts on this. Find us on the social channels linked at the top of the page.