Last Wednesday night, my daughter was stuck on a homework question — explain, in her own words, why a character in her reading book made the choice she did. I watched her open a chatbot app on the family tablet, type the question almost verbatim, and copy the answer back into her worksheet in under ten seconds. She never once looked like she was thinking. She looked like she was retrieving.

I’m Ethan, and BloomPath is the parenting app I’ve been building — I’m a software engineer by trade, more than a decade into the job, and for the last few years I’ve been teaching myself how to fold AI into tools for families. So when I saw my own kid outsource a “what do you think” question to a language model, I didn’t have the luxury of dismissing it as some abstract tech-anxiety headline. I use these tools every day. I know exactly what they’re good at, and exactly what they let you stop doing.

This article is part of our Screen Time in 2026 Guide.

TL;DR

  • A widely covered MIT Media Lab study found that people who used an AI chatbot to write essays showed weaker brain activity in regions tied to executive thinking and memory than people who wrote unaided — a pattern researchers called “cognitive debt.” The study is a preprint (not yet peer-reviewed) and tested adults, not children.
  • Kids anthropomorphize AI more than adults do, which means a chatbot’s tone and praise can shape their behavior more than we’d expect.
  • The risk isn’t AI itself — it’s using AI as a shortcut past the thinking, the same way copy-pasting code without reading it creates technical debt that bites you later.
  • The fix isn’t a ban. It’s a small set of household rules that treat AI like a calculator: useful after you’ve tried the problem, not instead of trying it.

Is AI Actually Changing How Kids’ Brains Work?

The honest answer is: we don’t fully know yet, but the early signal is worth taking seriously. The most-cited piece of research here is the MIT Media Lab’s “Your Brain on ChatGPT” study, which tracked participants’ brain activity while they wrote essays using either an AI assistant, a search engine, or no tool at all. The group using the AI assistant showed measurably weaker activity in the frontal and parietal regions — the areas associated with executive function and contextual memory — compared to the unaided group. When those same AI-assisted writers were later asked to write without help, their brain engagement stayed low, as if the shortcut had become the default setting.

Two caveats matter here, and I’d be lying to you if I skipped them. First, as of this writing the study hasn’t completed peer review, so treat the findings as a strong early signal, not settled science. Second, the participants were adults, not seven-year-olds doing reading homework. Nobody has run the equivalent brain-scan study on kids yet. What we do have is a related, more established finding: researchers studying AI use in classrooms and among teens have consistently found that younger, less experienced users show a greater drop-off in independent critical thinking when they lean on AI tools than older, more practiced users do. That part tracks with what child-development researchers already knew about kids before AI existed — their reasoning skills are still under construction, so a tool that skips the “figuring it out” step removes exactly the repetitions that build the skill.

What “Cognitive Debt” Means (And Why This Sounds Familiar to Me)

The MIT researchers borrowed the word “debt” on purpose, and it landed for me immediately, because I deal with a version of it every week at work. In software, “technical debt” is what happens when you ship the fast, copy-pasted fix instead of the one you actually understand — it works today, and it costs you double next month when nobody, including you, remembers why the code does what it does. Cognitive debt is the same trade, just happening in a kid’s head instead of a codebase. The AI answer works today. The muscle that would have built the answer from scratch doesn’t get used, so it doesn’t get stronger.

I’ve watched this exact pattern on my own team — engineers, adults, professionals — reach for an AI-generated function before they’ve spent thirty seconds thinking about the problem themselves. It’s not a character flaw. It’s what happens when a fast answer is one click away and a slow answer requires effort. Kids have far less built-up reasoning capacity to fall back on than adults do, which is exactly why researchers flag them as the more vulnerable group.

So How Worried Should I Actually Be?

Less worried than the “AI is rotting your kid’s brain” headlines want you to be, and more intentional than “it’s just a tool, don’t overthink it” wants you to be. Here’s where I’ve landed after reading the research and watching my own daughter use these tools:

  • Age matters enormously. A toddler or preschooler isn’t typing questions into a chatbot. The relevant risk window is school-age kids and teens doing independent work — homework, projects, anything that’s supposed to build reasoning.
  • The task matters more than the tool. Using AI to check spelling or find a synonym isn’t the same as using it to generate the entire argument of an essay. One replaces a mechanical step; the other replaces the thinking step.
  • Whether the child explains the answer afterward is the single best signal. If she can’t tell you why the answer is right in her own words, the thinking didn’t happen — the retrieval did.

The FTC’s updated child-privacy rules (COPPA), which tighten what AI products can collect from and target at kids, take effect in 2026 — a sign that regulators are also treating this as a live concern, not a hypothetical one.

The Framework We Use at Home: AI as Calculator, Not Crutch

Mei and I didn’t ban AI tools in our house — that ship has sailed for most families, and pretending it hasn’t just pushes the behavior underground. Instead we built three rules that treat AI the way a good math teacher treats a calculator: fine after you’ve tried it yourself, not instead of trying.

  1. Brain first, AI second. For homework and open-ended questions, she has to write or say her own attempt before she’s allowed to open any AI tool. It doesn’t have to be right. It has to be hers.
  2. “Explain it back to me” is non-negotiable. If she used AI for help, she has to walk me through the answer in her own words afterward. If she can’t, we go back and rebuild it together — slowly, out loud, at the kitchen table.
  3. AI answers questions of fact, not questions of judgment. “What year did this happen” is a fine AI question. “What do you think the character should have done” is not — that one’s hers to wrestle with, badly, for as long as it takes.

None of this is about being anti-technology. I build AI features for a living. The goal isn’t to keep her away from these tools — she’ll be using far more capable ones as an adult than the ones we have today. The goal is to make sure the thinking muscle is still there when the tool isn’t.

The Real Skill We’re Trying to Protect

What actually unsettled me that Wednesday night was how unbothered she was by skipping the hard part — the ten-second answer itself barely registered next to that. Kids have always looked for the shortcut, and that’s neither new nor a sign of bad character. What’s changed is how much faster, more convincing, and more available that shortcut has gotten. My real job has less to do with out-competing it and more to do with making sure she still knows how to do the slow thing when the fast thing isn’t the point.

BloomPath uses illustrated AI characters (Mei and Ethan) to protect our daughter’s privacy. The content is real; the avatars are illustrated. Learn more →

Frequently Asked Questions

Is AI actually bad for kids’ brains? The research so far is early and mostly done on adults, not children. The strongest study (MIT Media Lab) found weaker brain activity tied to executive function and memory in AI-assisted writers compared to unaided writers, and reduced engagement persisted even after the AI was taken away. That falls short of proof of permanent harm, but the signal is strong enough to take seriously, especially for kids whose reasoning skills are still developing.

What is “cognitive debt”? It’s a term researchers use to describe what happens when repeatedly outsourcing effortful thinking to AI leaves the underlying skill weaker over time — similar to how technical debt in software builds up when you ship a quick fix instead of a real one.

At what age should I worry about my child using AI chatbots? The relevant risk window starts around school age, once kids are doing independent homework and reasoning tasks. Toddlers and preschoolers aren’t typically the ones typing into chatbots directly — the bigger influence at that age is what they observe from parents’ own AI and screen habits.

Should I just ban AI tools for my kids? Most researchers and parents who’ve tried outright bans find they push the behavior underground rather than eliminating it. A more workable approach is teaching kids when AI is appropriate (checking facts, spelling) versus when it replaces the thinking they need to practice (arguments, opinions, working through a problem).

How can I tell if my child actually understood something or just copied an AI answer? Ask her to explain it back to you in her own words, without looking at the screen. If she can walk you through the reasoning, the thinking happened. If she can only repeat the answer, the retrieval happened instead — and it’s worth going back through the problem together.

Is using AI for homework considered cheating? That depends entirely on the assignment and the teacher’s policy, not on some universal rule. The more useful home question isn’t “is this allowed” — it’s “did my child actually build the skill this assignment was meant to build.”

Products We Recommend

  • The Anxious Generation by Jonathan Haidt — Not about AI specifically, but the clearest breakdown I’ve read of why kids’ developing brains need more friction, not less, from the tools we hand them.
  • The Tech-Wise Family by Andy Crouch — Practical, non-preachy framework for deciding what technology earns a place in your house. Helped Mei and me write our actual AI rules, not just feel vaguely uneasy about them.
  • ThinkFun Gravity Maze — A logic puzzle with zero shortcuts — you either work through the spatial reasoning or you don’t solve it. Our go-to for rebuilding the “sit with a hard problem” muscle.

Want a simple way to track the small, real moments of your child’s development — without another feed to scroll? The BloomPath app is built for that. Download BloomPath →