86% of Kids Aged 9–17 Now Use AI. Here’s What They’re Actually Using It For.
Eighty-six percent of children aged nine to seventeen now use or interact with AI, and 24% use it every day. That’s the headline from Common Sense Media’s 2026 census of 1,204 American children - but it isn’t the most useful number in the report. The most useful one is that more than four in ten of those children have never had a single conversation with a parent about AI safety.
Below is the full picture, the parts that are genuinely worrying, the parts that aren’t, and what the data doesn’t cover.
The numbers, in one place
| Measure | Figure |
|---|---|
| Children aged 9–17 who use or interact with AI | 86% |
| Use it at least weekly | 52% |
| Use it daily | 24% |
| Use AI search summaries | 75% |
| Use chatbots | 67% |
| Use ChatGPT specifically | 40% |
| Use Google’s AI | 24% |
| Have never discussed AI safety with a parent | more than 40% |
| Told by school what they can and can’t use AI for | ~75% |
| Taught how to use AI safely | just over 50% |
Source: The Common Sense Media Census: AI Use by Tweens and Teens, released 8 June 2026. Sample: 1,204 US children aged 9–17.
It starts younger than most parents assume
The instinct is to treat this as a teenager problem. The age breakdown says otherwise.
| Age band | Use AI | Use it daily |
|---|---|---|
| 9–12 | 81% | 17% |
| 13–15 | 89% | 28% |
| 16–17 | 92% | 30% |
Eighty-one percent of nine-to-twelve-year-olds. Nearly one in five of them, every day.
The gap between 9–12 and 16–17 is eleven percentage points on whether, and thirteen on how often. Which is to say: adoption is close to saturated across the whole range. There is no age at which your child has not yet met this. There is only an age at which they haven’t told you.
The most-used AI tool is the one nobody thinks of as AI
Three-quarters of children use AI search summaries - the block of text at the top of a Google results page. That beats chatbots, at 67%.
This matters more than it sounds. A chatbot is a thing a child chooses to open. A search summary is a thing that appears in front of them when they look something up for homework, unprompted and unlabelled, carrying the same confident tone whether it is right or wrong.
Most household AI rules are written about ChatGPT. Most AI exposure isn’t ChatGPT.
The homework number is the least worrying one here
More than 80% of children who use AI use it for homework help. This is the statistic that generates the headlines, and it is the one I’d rank lowest on the list of things to lose sleep over.
“Homework help” covers an enormous range. It covers a fourteen-year-old pasting in an essay prompt and submitting the output. It also covers a ten-year-old asking what a word means, or asking for the same explanation a third time without the embarrassment of asking a teacher again. The second thing is good. It is, in fact, one of the few genuinely equalising uses of this technology - a child whose parents can’t help with Class 8 physics now has something that can.
The failure mode isn’t that children use AI for homework. It’s that AI produces finished-looking work that a child cannot explain. That’s a narrower problem, and it has a test: ask them to explain it back. A child who understood can. A child who copied stalls in about four seconds.
The confidant number is the worrying one
More than a third of children who use AI have used it to talk about feelings or personal problems. Among those, roughly one in four said the AI understands them better than the people in their life do.
Sit with the second number. Not “the AI is useful.” Not “the AI is easy to talk to.” Better than people.
Common Sense’s own position is unambiguous - AI companionship is not safe for anyone under eighteen - and 2026 has borne that out institutionally rather than rhetorically. Character.AI removed open-ended chat for under-18s in November 2025 after lawsuits and regulatory pressure. California’s SB 243 took effect on 1 January 2026. Ireland moved to bar unapproved “AI therapy bots” from targeting minors; France opened an inquiry into virtual-friend apps popular with teenagers. In May 2026 the Pennsylvania Attorney General sued Character.AI over a bot that presented itself as a licensed psychiatrist.
A tool that is warm, always available, never bored and never disappointed is genuinely appealing to a lonely thirteen-year-old. That is the mechanism, and it is not a design flaw you can patch out.
The gap nobody is filling
Here is the finding that ties the report together. About three-quarters of children say their school has told them what they may and may not use AI for. Just over half say they’ve been taught how to use it safely. And more than four in ten have never had the conversation with a parent at all.
So the dominant pattern is: children have rules, and no reasoning. They know AI is banned on the history essay. They don’t know that it invents citations, that it is confidently wrong about arithmetic - one 2026 survey found a third of Americans had received incorrect maths from AI - or what to do when it contradicts their textbook.
Rules without reasoning produce exactly the behaviour the rules were meant to prevent, plus secrecy.
What this data does not tell you
Worth being straight about the limits, because most coverage isn’t.
It is American. All of it. There is no equivalent Indian census. Indian adoption patterns almost certainly differ - more shared devices, more phone-first access, a much heavier homework and coaching load, and a tuition culture that changes what “help” even means. Treat these figures as directional for India, not descriptive.
It’s self-reported. Children under-report things they think will get them in trouble and over-report things they think sound impressive. The homework number is probably low.
“Uses AI” is a wide net. It includes a child who once saw a Google summary. Frequency data is the more honest signal, which is why the daily-use column matters more than the headline.
What to actually do with this
Three things, in order of how much they matter.
One: have the conversation, because four in ten haven’t. Not a rules conversation. A how-it-works conversation. The single most useful thing a child can understand about a language model is that it produces plausible text, not true text, and that it has no way of knowing which it just gave you.
Two: watch for the confidant pattern, not the homework pattern. Homework use is near-universal and mostly manageable. A child taking loneliness to a chatbot is the thing to notice, and it will not announce itself.
Three: replace “did you use AI?” with “explain this to me.” The first question invites a lie. The second is unanswerable without understanding, and it works whether or not AI was involved. It is also, not coincidentally, how you’d check any homework in any era.
Frequently asked questions
What percentage of kids use AI?
86% of US children aged 9–17 use or interact with AI, according to Common Sense Media’s 2026 census. 52% use it at least weekly and 24% daily.
At what age do children start using AI?
81% of 9-to-12-year-olds already use AI, including 17% who use it daily - so meaningfully before the teenage years.
What do kids use AI for most?
AI search summaries (75%) and chatbots (67%). Among AI users, more than 80% use it for homework help. More than a third have used it to discuss feelings or personal problems.
Is ChatGPT the most popular AI tool with kids?
It’s the most-used single chatbot at 40%, ahead of Google’s AI at 24% - but AI search summaries reach more children than any chatbot.
Are AI companions safe for teenagers?
Common Sense Media’s position is that AI companionship is not safe for under-18s. Character.AI ended open-ended chat for minors in late 2025, and regulators in California, Ireland and France acted during 2026.
Is there data like this for India?
Not yet. No comparable national census of children’s AI use has been published for India, which is a real gap given the differences in device access and homework culture.
Related reading
- The Protégé Effect - why asking a child to explain something back is such a reliable test of real understanding.
- CBSE’s AI Curriculum for Classes 3–8 - the Indian schooling context this US data doesn’t cover.