Does ChatGPT make kids dumb? What the MIT study says and what it does not

It lands in the class group chat in the same format every time: a screenshot, a headline in capitals and the letters MIT. And the study exists, it is called Your Brain on ChatGPT and it comes from a team at the MIT Media Lab. What almost nobody has opened is its 216 pages, because there is not one child in them: there are 54 adults from five Boston universities writing 20 minute essays with an electrode cap on.

What was actually done

• 54 participants aged 18 to 39 (average 22.9), students from MIT, Wellesley, Harvard, Tufts and Northeastern.
• Three groups: one wrote with ChatGPT, one with a search engine, the third with nothing but their own head.
• Three sessions, each with a 20 minute essay on a topic taken from the SAT, the US college admissions test.
• A fourth session with the roles swapped: those who had used ChatGPT wrote with no tools, and those who had written with no tools used ChatGPT. 18 people came back for that fourth session.
• All of it over four months, wearing a 32 electrode cap that records how strongly the regions of the brain talk to each other while you write.

What came out

Three things, and all three are about essays, not intelligence. Brain connectivity was strongest in the group writing unaided, moderate with the search engine and weakest with ChatGPT. The ChatGPT group struggled most to quote a line from their own text minutes after handing it in. And they felt the least that the essay was theirs. The part nobody forwards is the fourth session: those who had written three times on their own and used ChatGPT only at the end remembered better and showed activation similar to the search engine group. Order matters, and that is the finding a family can use.

Please do not use the words like 'stupid', 'dumb', 'brain rot', 'harm', 'damage'

That is the team's literal answer to the question of whether it is fair to say AI makes us dumb. The paper is a preprint: it has been on arXiv since June 2025, the current version is from December 2025, and according to its authors peer review is under way and will take months. They list the limits themselves: few participants, a single region, only ChatGPT and only essays. There was no outside funding. So the headline does not hold, but the finding does, and it is more uncomfortable than the headline: what you let the machine do first does not stay with you.

The one rule that follows from the study

You first, then the AI. It is not a ban, it is an order of events. Essay, problem, summary, whatever it is: ten minutes of your own attempt on paper or in a blank document, and only then does the chat open, to compare, correct or ask for criticism. At home it fits in one sentence that applies to the adults too: the AI gets opened with something already written. The email to the school, the shopping list, the difficult message: your own draft first. When the parents keep the rule, it gets argued about a lot less.

Three sentences for the conversation with your teenager

• That study does not say you are dumb or that you will be: it says 54 university students remembered an essay less well when they wrote it with ChatGPT. Read it with me if you like.
• What it does show is that order matters: the ones who wrote on their own first and used the AI afterwards remembered better. That is all I am asking for, the order.
• Try it yourself: write one paragraph with AI and one without, and tomorrow try to quote a line from each. Then tell me how it went.