Parents underestimate how much their kids use AI

Ask a parent whether their child used AI this week and roughly a quarter will say yes. Ask the children themselves and closer to four in ten will. Same households, same seven days, different story.

That gap sits at the centre of new research from the Family Online Safety Institute (FOSI), a Washington-based online-safety group, published on 9 July 2026. Its fourth Online Safety Survey, fielded by Ipsos, put the same questions to more than 4,000 parents and children aged 10 to 17 across the United States and Australia. On almost everything it measured, kids reported doing more online than their parents thought. The widest split was around generative AI.

The number worth sitting with

When asked whether their child had used generative AI in the past week, 27% of parents said yes. Among the children, 38% said they had. An 11-point gap, on a single direct question, about the same week.

It would be easy to wave this away as kids exaggerating or parents forgetting. But the pattern repeats across the survey in a way that is hard to dismiss. More children than parents reported scrolling social media (54% against 46%). More reported posting to it (38% against 30%). Every time the two groups disagreed, they disagreed in the same direction: the child’s online life was a little larger than the parent pictured.

Not panic, but not nothing either

Two things are worth holding at once. First, the sky is not falling. Generative AI use among American children in the survey actually levelled off, from 74% in the previous wave to 72% now. This is not a runaway. Second, parents are quietly cooling on it. FOSI found parental optimism about AI has slid from 52% in its first wave to 42% now. So the mood is souring at the same moment the visibility is thinning. That is an awkward pairing: adults feel warier about a thing they can see less and less of.

There is a reason the blind spot is specific to AI. A parent can walk past a child watching YouTube and know roughly what is happening. A child typing to a chatbot looks identical whether they are working through algebra, drafting a message to a friend, or asking something they would never say out loud to an adult. The activity is legible. The content is not.

Why the gap is the actual problem

The instinct, faced with a number like this, is to reach for monitoring software. The research points somewhere less dramatic and more useful.

Common Sense Media, the media-and-technology non-profit, found something similar in its 2024 study of teenagers and AI at home, reported by KQED: about a quarter of parents did not believe their child used AI at all, even as those same teens said they did. But the more telling finding was about guidance. Teens who had discussed AI’s benefits and risks at school were markedly more likely to check what a chatbot told them, 55% against 43% for those who had not. As Common Sense researcher Amanda Lenhart put it, many kids “don’t have the guidance to know what it is they’re supposed to be doing”.

That reframes the gap. The worry is not simply that a child uses AI a parent cannot see. It is that a child may be using it without anyone having taught them the basic reflexes: that a confident answer can be wrong, that a chatbot is not a friend, and that some questions are better taken to a person.

What closes the gap

FOSI’s own numbers hint at the way through. Ninety per cent of the children surveyed said they feel safe talking to their parents about something that worries them online. The channel is open. Most kids are not hiding. Their parents just have not asked the right question lately.

So ask a better one. Not “are you using AI”, which earns a shrug, but “show me the last thing you asked it”. Sit through one real exchange. You will learn more from watching your child reason with a chatbot for two minutes than from any dashboard, and you will be there for the moment that actually matters: when the tool gets something wrong and your kid has to decide whether to believe it.

(A note on our stake: Mentus AI builds AI mentoring tools for children, so a parent knowing what their child does with AI sits close to our business. We think that is the point, not a caveat.)

The survey caught families at an unusually charged moment. FOSI gathered some of its data around the time Australia’s ban on social media for under-16s took effect, with children’s digital lives under national debate. Rules like that will keep coming. But a law can only decide what a child is allowed to do. Whether a child can tell a good answer from a confident wrong one is the harder part, and that still gets settled at home, by a parent who knows what is on the screen.