Whizbee

Research · For ages 7–11

Does AI make kids dumber? What the research actually says for ages 7–11

By Whizbee · Updated June 2026 · Reviewed against primary research

Short answer: No — AI does not "make kids dumber" on its own. But the evidence does show a real risk. When a child uses AI to replace their own thinking — copying answers without effort — measurable signs of weaker engagement and recall appear, an effect researchers call cognitive debt. When AI is used to support thinking — prompting a child to reason, retrieve, and explain — that risk largely disappears. The danger isn’t AI. It’s mental delegation. The two are not the same thing, and the difference is everything.

This matters most for children aged 7–11, because this is precisely when the foundational habits of reading, reasoning, and memory are being built.

What the MIT "cognitive debt" study really found (and what it didn’t)

Almost every alarming headline about AI and children traces back to one study: Your Brain on ChatGPT, a 2025 preprint from the MIT Media Lab (Kosmyna and colleagues, with Wellesley and MassArt). It deserves to be understood accurately, because most coverage distorts it.

What the study did: 54 adults aged 18–39 wrote essays across several sessions, split into three groups — one using ChatGPT (GPT-4o), one using a search engine, and one using no tools at all ("brain-only"). The researchers measured brain activity with EEG.

What it found: brain connectivity scaled down with the amount of outside help. The brain-only group showed the strongest, most widely distributed neural networks; search-engine users showed moderate engagement; and the ChatGPT group showed the weakest connectivity. Reduced engagement and weaker recall persisted even after the AI was taken away — hence "cognitive debt."

The single most revealing detail: in a later crossover session, participants who had started without AI and only then switched to it kept higher memory recall and stronger brain engagement. Those who had leaned on AI first and then had it removed stayed under-engaged. The order matters: thinking first, then using AI as a check, looks healthier than reaching for AI first.

Now the part the headlines leave out. The study’s authors were so concerned about misrepresentation that they explicitly asked people not to describe their work with words like "dumb," "brain rot," or "brain damage." And they listed real limitations themselves:

  • It studied adults, not children. Nothing in it measured 7–11-year-olds.
  • The sample was small (54 people; only 18 finished the crossover session).
  • It tested only ChatGPT and only essay writing — not other tools or tasks.
  • EEG cannot see deep memory structures like the hippocampus; the authors call for fMRI follow-ups.
  • It is a preprint — early work, not yet peer-reviewed.

So the honest reading is narrower and more useful than "AI rots kids’ brains": outsourcing effortful thinking to AI has a measurable cognitive cost, the effect compounds with reliance, and using your own mind before reaching for the tool appears protective.

Mental delegation vs. reasoning support — the distinction that decides everything

The reason "does AI make kids dumber?" has no single yes/no answer is that "using AI" describes two opposite behaviours:

Mental delegation (the risk) Reasoning support (the benefit)
What the child doesAsks for the answer, copies itAttempts first, uses AI to check or unstick
Cognitive effortOffloaded to the machineStays with the child
What’s practisedPromptingThinking, retrieval, explaining
Likely effect over timeCognitive debt; weaker recallReinforced understanding
Example“Write my essay on volcanoes”“I think volcanoes erupt because of pressure — is that right, and what am I missing?”

This is the same lesson the MIT crossover finding points to. AI that hands over a finished answer encourages delegation. AI that responds with a question, a hint, or a "show me your reasoning" keeps the child doing the cognitive work.

Why ages 7–11 are a special case

The MIT study looked at adults, so applying it to children requires care — but the developmental stakes for 7–11-year-olds are arguably higher, not lower, for three reasons:

1. Foundational skills are forming now. Between roughly 7 and 11, children consolidate reading fluency, number sense, and the ability to hold and manipulate ideas in working memory. Skills built through effort at this age become automatic later. Skills skipped are harder to recover.

2. The "reading brain" is still under construction. Cognitive scientists like Maryanne Wolf (Reader, Come Home, 2018) argue that deep reading and sustained reasoning circuits develop through practice, and can be weakened by habits that reward shortcuts over effort.

3. Metacognition is immature. A 9-year-old is far less able than an adult to notice "I’m not really learning this, I’m just copying." Without that self-check, delegation can become the default without anyone choosing it.

Emerging 2026 research is starting to look directly at children — for example, studies on how kids form mental models of AI reasoning, and on how families build "AI literacy" at home — but a clear, child-specific consensus does not yet exist. The responsible position today is: take the adult evidence seriously as a warning, weight it more heavily for young children given the developmental stakes, and design for reasoning support rather than delegation.

What actually protects a child’s thinking

The flip side of cognitive debt is well-established learning science. The same things that build durable understanding are the things that keep a child’s mind active when AI is in the room:

Productive struggle (desirable difficulty). A little effort and even a little failure before help arrives strengthens learning. Cognitive scientist Robert Bjork’s work on "desirable difficulties" shows that making learning feel slightly harder in the moment improves long-term retention.

Retrieval practice. Pulling an answer from memory (rather than re-reading or being told) is one of the most reliably effective study methods in the research literature.

Spacing. Revisiting an idea after a gap beats cramming — a finding replicated for decades (the "spacing effect").

Explaining it back (teach-back). A child who has to explain what they learned, in their own words, reveals and repairs the gaps that copying hides.

Tools that ask before they answer. An AI that prompts a child to predict, attempt, and justify keeps the cognitive load where it belongs — with the learner.

Concern about cognitive offloading — letting a tool hold information or do the thinking so the brain doesn’t have to (Gerlich, 2025; and the older "Google effects on memory" line of research) — is real. But offloading is only harmful when it replaces a skill the child still needs to build. A calculator doesn’t harm a child who already understands what multiplication means; it harms one who never learned. The same logic applies to AI.

A practical test for parents: is this AI helping or replacing?

Before letting a child use any AI tool for learning, ask whether it passes this checklist:

  1. 1. Does it make the child attempt first, or hand over the answer immediately?
  2. 2. Does it ask the child to explain or predict, not just receive?
  3. 3. Does it let the child be a little bit stuck before helping?
  4. 4. Can you see what the child is learning — progress and understanding — rather than only a transcript of messages?
  5. 5. Is it built for a child’s age and a closed, safe context, not a general adult tool with the open internet attached?

A tool that fails 1–3 is training delegation. A tool that passes them is training thinking.

Frequently asked questions

Does using ChatGPT make children dumber?

No reliable study shows that AI lowers children’s intelligence. The MIT "cognitive debt" research — conducted on adults — found that outsourcing thinking to ChatGPT reduced brain engagement and recall. Used to support a child’s own reasoning rather than replace it, AI does not show that effect. The behaviour matters more than the tool.

What is cognitive debt?

Cognitive debt is the gradual weakening of thinking and memory that can follow from repeatedly letting a tool do mental work for you. In the MIT study, the effect persisted even after the AI was removed. It is a risk of over-reliance, not an automatic result of any AI use.

Is AI bad for a 7–11-year-old’s learning?

It depends entirely on how it’s used. At this age, children are building foundational reading, reasoning, and memory skills through effort, so AI that replaces that effort is riskier than for an adult. AI that prompts a child to attempt, retrieve, and explain can support learning.

How can I let my child use AI without harming their thinking?

Choose tools that make the child attempt first and explain their reasoning, that allow productive struggle before helping, and that show you proof of learning rather than just a chat log. Encourage your child to think before asking, and to teach back what they learned.

Is the MIT study proof that AI harms kids?

No. It is an early, not-yet-peer-reviewed preprint, with a small adult sample and a single task (essay writing). Its authors explicitly cautioned against framing it as "AI makes you dumb." It’s a serious warning about over-reliance, not a verdict on children.

Sources

Primary research and references behind this article. Where work is a preprint, it is described as such.

Explore related kid topics

Free explainers with a short quiz — useful after reading this research.

All topics →

Built for reasoning, not delegation

This is the principle Whizbee is built on: instead of handing a child the answer, it prompts them to reason toward it — attempt, get nudged, explain it back. More on that approach in Whizbee vs ChatGPT and what "safe AI for kids" should mean.

← Back to all research