She fell seventeen times in one hour yesterday. I counted, because that is the kind of thing you do when you are the parent of a two-year-old learning to run down a small hill. Seventeen falls, seventeen get-back-ups, zero requests for anyone to just carry her to the bottom. I stood there thinking about effort the way you think about oxygen — only because something made me notice it.
What made me notice it was the browser tab I’d had open since the night before: a Pew Research study reporting that 54% of American teenagers have now used AI chatbots for schoolwork help. I’d fallen asleep thinking about it. I woke up thinking about it. And watching my daughter pick gravel out of her knee with total concentration, I finally understood why it unsettles me in a way that goes past policy or pedagogy. It’s about what happens when the hill gets easy before you’ve learned anything from the falling.
She is two. ChatGPT is not her problem yet. But the question of when it becomes her problem — and what guardrails actually exist between now and then — is very much mine. So I did the research. Here is what I found, as of July 2026, with sources you can check yourself.
How fast this normalized: the numbers that stopped me cold
In 2023, Pew Research Center found that 13% of U.S. teens ages 13–17 had used ChatGPT for schoolwork. By fall 2024, that number had doubled to 26%. Then Pew fielded another survey in September–October 2025, and the results, published February 24, 2026, showed the number had doubled again: 54% of teens have now used AI chatbots for schoolwork help. About 30% use AI chatbots daily. Ten percent say they do all or most of their schoolwork with chatbot help.
That is not a gradual cultural shift. That is a threshold being crossed in real time, across a single cohort of kids, mid-childhood.
What I find genuinely interesting — and a little hopeful — is where teens themselves draw the line. In the January 2025 Pew data, 54% of teens said using ChatGPT to research new topics was acceptable. That dropped to 29% for solving math problems, and to 18% for writing essays. Teenagers are not ethically asleep here. They have their own lines. They just aren’t always sure whose job it is to enforce them.
What schools actually did (and keep doing)
New York City Public Schools blocked ChatGPT in January 2023. They reversed that decision within about four months. This whiplash turned out to be a preview of national policy: ban, reconsider, integrate, regulate, repeat.
As of January 2026, education departments in 31 states had issued K–12 AI guidance. In 2026 alone, 134 AI-in-education bills were introduced across 31 states. The dominant classroom model by 2025–26 has become disclosure-plus-teacher-discretion: students must say when they used AI, and teachers decide whether that’s acceptable for a given assignment, while schools often block public AI tools in favor of vetted classroom products.
A survey by the Center for Democracy and Technology found that 85% of teachers used AI in class during 2024–25, and 86% of students used AI for personal or school reasons. That near-symmetry is striking. The adults and the kids arrived at the same tool at roughly the same moment, which may be part of why the rule-setting has felt so improvised.
If you are reading this because your child’s school just changed its AI policy and you are trying to figure out what your own family policy should be — that is exactly what this post is for.
What the research says about learning — and what it doesn’t yet prove
In June 2025, MIT Media Lab researchers posted a preprint — not yet peer-reviewed — of an EEG study they called “Your Brain on ChatGPT.” Fifty-four participants wrote essays using one of three methods: an LLM, a search engine, or their own memory alone. The group using the LLM showed the weakest brain connectivity of the three. The researchers used the phrase “cognitive debt” to describe the pattern.
I want to be careful here, because I think this study is genuinely interesting and also genuinely preliminary. The sample was 54 people. It measured one task. It has not been peer-reviewed. “Cognitive debt” is the authors’ framing, not an established clinical term. It would be easy to read this study as proof of something it hasn’t proven yet. What it is, accurately, is early evidence worth watching — the kind of thing that should shape research priorities, not parenting verdicts.
The actual age guidance: what exists, what doesn’t
Here is the direct answer to the question in the title, assembled from what actually exists:
- Platform minimum: 13. OpenAI’s terms of service set 13 as the minimum age for ChatGPT. Parental controls became available in September 2025. This is a legal floor, not a developmental recommendation.
- Ages 6–12: supervision only. Common Sense Media’s guidance places children in this age range in a category where AI use — including homework-type use — should happen only with an adult present. They distinguish homework help (lower risk) from companionship or social use (higher risk), which is a useful distinction most coverage skips.
- The AAP has not set a minimum age. The American Academy of Pediatrics, as of July 2026, has no formal policy on AI use by children. What they have published on HealthyChildren.org is practical guidance: review chatbot conversations with your child, explain that only real people can offer loyalty, caring, or truthfulness, discourage sharing personal information, and keep devices in common areas when children are young. That is guidance. Call it what it is.
The honest summary is: there is no consensus answer from any authoritative body that says “age X, proceed.” What exists is a platform floor of 13, a supervision expectation for younger kids, and a set of AAP habits that apply across the childhood years. Anyone telling you there’s a precise, research-validated age is working from confidence they haven’t earned yet.
What we actually do in our house
My daughter is two, so this section is mostly aspirational — a framework I’m building now so I’m not improvising when she’s nine and her friends are using AI to write their book reports.
Our current position is no AI tools in her learning environment, full stop, for the foreseeable future. Not because I think AI is malevolent, but because I am specifically interested in what happens when a child has to sit with difficulty. The hill. The seventeen falls. I don’t think I can make that argument convincingly to a twelve-year-old if I haven’t lived it with her first.
What I plan to do, when it’s relevant: follow the AAP habits — conversations in common spaces, reviewing what she’s using and why, talking about what the tool can and cannot do. I will use Common Sense Media’s category distinction and treat homework help differently from any kind of social or companionship use. I will not pretend the tool doesn’t exist; I’ll just be in the room when she encounters it.
What’s hard about this: I write on a laptop. I use search constantly. The line between “research tool” and “thinking tool” is not crisp, and I am aware that my kid is watching me blur it daily. I don’t have a clean answer to that. I’m working on it.
Questions I keep getting
Is ChatGPT actually banned for kids under 13?
OpenAI’s terms of service set 13 as the minimum age. That’s a platform rule, enforced the way most platform age rules are enforced — which is to say, imperfectly, via self-reported birthdays. OpenAI added parental controls in September 2025, which gives families more practical tools, but the 13 floor is a legal and contractual threshold, not a technical lock. If your child is under 13 and accessing ChatGPT, they are outside the platform’s intended terms.
My kid’s school allows AI now. Does that mean it’s fine?
It means it’s permitted, which is different from fine. The school policy landscape has moved from blanket bans toward disclosure-plus-teacher-discretion models, and as of January 2026, 31 states have issued K–12 AI guidance. But school policy answers the question “is this allowed in this context” — it doesn’t answer the developmental question of what repeated AI-assisted work does to a child’s capacity to struggle productively, hold an argument together, or tolerate not knowing something. Those are separate questions, and the research on them is early.
What does the AAP actually recommend?
As of July 2026, the AAP has no formal policy setting a minimum age for AI use. Their published guidance on HealthyChildren.org recommends reviewing chatbot conversations together, explaining that AI cannot offer loyalty, caring, or truthfulness the way real people can, discouraging children from sharing personal information, and keeping devices in common areas when kids are young. These are habits, not rules — and they’re useful precisely because they apply at any age and travel with your kid as the technology changes.
Should I be worried about what AI is doing to my teenager’s brain?
Cautiously attentive is probably the right register, not worried. The June 2025 MIT Media Lab preprint — a small study of 54 people doing a single writing task — found weaker brain connectivity in the group using an LLM, and the authors described a pattern they called “cognitive debt.” That is genuinely worth knowing about. It is also a preprint, not peer-reviewed, with a small sample and one task type. The honest answer is: we don’t have enough evidence yet to make strong claims about long-term effects, which is a reason to stay engaged with the research and to make sure your teenager is still doing hard things the hard way — not a reason to panic.
