When Children Tell Machines What They Cannot Tell Us

A child can ask a chatbot anything without blushing, being interrupted or hearing, “Can we talk about this after dinner?”

From Homework to Heartache

In August 2026, Australia’s eSafety Commissioner released the results of a survey of 1,950 Australian children aged 10 to 17. The numbers should make every parent, educator and children’s yoga teacher sit up a little straighter — no posture correction required.

The study found that 78 per cent of children had used an AI assistant and 8 per cent had used an AI companion. Among those who had used either kind of tool, 20 per cent were using one daily or more often.

Most children still used AI for ordinary practical things: finding information, learning, translating, writing and editing. But 54 per cent had also used it for personal or social reasons. One-third had asked what to do in a situation. One-third had sought advice about their physical health. Twenty-two per cent had discussed feelings or challenges, and 20 per cent had asked for mental-health or wellbeing advice.

Among children with a diagnosed mental-health condition, the figures were more striking: 39 per cent had sought mental-health or wellbeing advice from AI, while 35 per cent had used it to talk about their feelings or challenges.

The machine had moved quietly from the homework desk to the child’s inner world.

There were benefits. Eighty-five per cent of young users said an AI interaction had made them feel something positive at least sometimes — informed, reassured, clever, happy or accepted. We should not pretend that comfort is meaningless simply because it arrived through a screen.

But comfort is not the same as safety.

Almost half — 47 per cent — said an AI tool had also made them feel something negative, such as confusion, frustration, anxiety, shame, hopelessness or feeling unsafe. One in five had experienced a potentially inappropriate or harmful interaction, including controlling behaviour, hateful material, bullying, harmful advice or sexual content. Thirty-seven per cent had shared personal, private or sensitive information; even after names and birth dates that might have been used for registration were excluded, the figure was still 32 per cent.

And although 85 per cent of parents knew when their child had used an AI assistant, only 43 per cent knew when their child had used an AI companion.

“AI is not a trusted adult.”

Julie Inman Grant, eSafety Commissioner

That may be the most important five-word technology lesson we teach this decade.

Please Do Not Blame the Children

It would be easy to respond with a grand adult panic: ban the apps, confiscate the phones and announce that in our day we spoke to actual people — conveniently forgetting all the things we absolutely did not tell actual people.

Children are not turning to machines because they have suddenly lost all interest in humanity. They are turning to machines because machines offer several things humans sometimes fail to offer:

The chatbot never gasps, laughs, tells Aunt Margaret or looks disappointed. It does not need to pick up another child, answer an email or put the lentils on.

Of course a child may choose it.

UNICEF researchers report that some teenagers say talking to AI feels easier than talking to peers or adults. That does not prove that AI is replacing relationships, but it does hold up an uncomfortable mirror. Perhaps the urgent question is not, How do we stop children talking to algorithms?

Perhaps it is:

How Do We Become the Adults Children Would Rather Approach Than an Algorithm?

This question asks more of us than installing parental controls.

It asks whether we are emotionally available, not merely physically nearby.

It asks whether a child can tell us something messy without immediately receiving a lecture, an interrogation or twelve solutions.

It asks whether our face says, I can handle your truth, even when our nervous system is shouting, WHAT ON EARTH?

Children do not only assess whether an adult is kind. They assess whether the adult is safe to unsettle.

That distinction matters.

A trusted adult is not someone who always knows what to say. A trusted adult is someone who can stay present long enough for the child to keep speaking.

A Machine Reads Words. A Human Can Read a Child.

After more than thirty years of teaching yoga, I have learned that children often tell us the truth with their bodies before they can tell us in words.

• A child who says “I’m fine” may be holding their breath.
• The class comedian may be working very hard to keep sadness from catching up.
• The child who refuses a partner pose may not be “difficult”; they may be telling us that closeness does not feel safe today.
• The child who does every pose perfectly may still feel completely outside the circle.

A chatbot receives the sentence a child types. A present human can notice the pause before the sentence, the shoulders curling in, the sudden silence, the eyes searching the room and the joke that arrived half a second too quickly.

A machine does not know the family, the playground, the culture, the diagnosis, the history or what happened just before the child opened the app. It cannot walk into the next room, bring in another safe adult, contact a qualified professional or sit beside the child while the worst wave passes.

And crucially, it is not accountable in the way a human teacher, parent, counsellor or clinician is accountable.

AI can imitate empathy with astonishing fluency. But simulated warmth is not the same as a relationship that bears responsibility.

Real Relationships Are Gloriously Inconvenient

AI companions are often designed to be endlessly agreeable. That feels lovely — for approximately the same reason it feels lovely when everyone lets you win at a board game.

But real friendship is not a customised mirror.

Real people misunderstand us. They have needs of their own. They disagree. They require us to listen, apologise, compromise, wait, repair and try again. UNICEF warns that if agreeable AI interactions displace human ones, children may lose opportunities to practise communication, empathy and conflict resolution.

That does not make AI evil. It makes it incomplete.

AI is not the villain. But it is not the village either.

The World Health Organization estimates that one in seven young people aged 10 to 19 experiences a mental-health condition. It identifies social and emotional skills, regular movement, supportive families, safe schools and healthy peer relationships as important protective factors.

In other words, the answer to artificial companionship is not simply less technology.

It is more belonging.

What Yoga Can Offer That a Chatbot Cannot

Yoga is sometimes marketed to children as a collection of techniques for calming down: breathe like this, stretch like that, now please return to class as a more convenient small person.

But its deeper potential is relational.

In a good children’s yoga class, we do not merely teach a child to manage themselves alone. We create a living laboratory of connection.

We breathe together. Balance together. Make shapes that only work when everyone has a place. We ask before touching. We learn to say no and hear no without collapsing. We wobble, laugh, negotiate, support someone else’s weight and discover that needing help is not a character defect.

In Rainbow Yoga, circles, partner poses and group adventures are not decorations around the “real” yoga. They are the real work of becoming human with other humans.

This is co-regulation: nervous systems helping nervous systems. It is much harder to package than an app, but wonderfully, no software update is required.

Sometimes the most healing sentence a teacher can say is not profound at all:

Thank you for telling me. You did the right thing. Let’s find the right person to help.

A New Skill for the AI Age: Pause, Name, Check, Tell

Children will use AI. Telling them simply not to is about as effective as telling a group of children not to look inside a mysterious box.

Instead, we can give them a body-based decision process that works in a classroom, at home or at 11:47 at night.

We should practise these four steps before a crisis, just as we practise a fire drill before the building is smoky.

Give Every Child a Human Help Map

This is embodied digital literacy: not only knowing what the technology can do, but noticing what it is doing to us and knowing when to step out of the chat and into human care.

The Adult’s Side of the Practice

We cannot teach children to tell without teaching adults how to receive.

When a child approaches:

And please, do not make the child’s AI use the main scandal when the real issue is what they were trying to survive, understand or say.

The prompt history is not the child. It is a trail leading toward a need.

Not an Anti-AI Campaign — a Pro-Human One

AI can help children explore ideas, practise a language, invent twelve names for a dragon, understand a difficult paragraph and ask questions they are still learning how to form. Used thoughtfully, it can be useful, creative and genuinely empowering.

This is not a campaign to drag childhood backwards or pretend the technology will disappear if adults frown hard enough.

It is a call to keep the categories clear.

And an always-available response is not the same as a responsible relationship.

Our task is not to win a competition against machines. It is to offer what machines cannot: bodies in the room, eyes that notice, arms that can help, adults who can act, communities that remember and relationships strong enough to hold an inconvenient truth.

The most hopeful finding in the eSafety research may be this: when an AI interaction made children feel bad, they were still more likely to tell a real person than to tell the machine.

The bridge to us is still there.

Let us make it easier to cross.

Let us become less shocked and more available. Less interrogating and more curious. Less eager to correct and more capable of listening.

Let us create homes, schools and yoga circles in which a child does not need perfect words, perfect timing or a crisis large enough to justify our attention.

Because the future of children’s wellbeing will not be secured by teaching them never to speak to machines.

It will be secured by making sure that, when something truly matters, they know which humans will listen — and trust that we can bear what they need to say.