The MIT "Your Brain on ChatGPT" study: what it did and did not find
It is a small preprint from the MIT Media Lab: 54 people, essay writing, brain activity recorded with EEG. People who wrote with ChatGPT showed the weakest brain connectivity of three groups, and most of them could not quote their own essay minutes later. It did not test IQ, it has not been peer reviewed as of 13 November 2026, and its authors ask the media not to call the result "brain rot".
The study in one table
| Title | Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task |
|---|---|
| Who | Nataliya Kosmyna and colleagues, MIT Media Lab |
| Status | Preprint, under review. First posted June 2025, revised December 2025. Not yet peer reviewed |
| People | 54 adults aged 18 to 39 from five universities near Boston. 18 took part in the last session |
| Task | Timed essays on set topics, over three sessions, plus a fourth where groups swapped |
| Groups | ChatGPT, a search engine, or no tools at all |
| Measured | Brain activity (EEG), the essays themselves, grades from teachers, and interviews |
| Source | arxiv.org/abs/2506.08872 |
What it found
- Less brain engagement with more help. Connectivity was strongest in the group with no tools, in the middle for the search group, and weakest for the ChatGPT group.
- Weaker recall of their own words. In the first session, 15 of the 18 people in the ChatGPT group could not correctly quote a sentence from the essay they had just written. In each of the other two groups, 2 of 18 could not.
- Less sense of ownership. The ChatGPT group reported the lowest feeling that the essay was theirs.
- Order mattered. In the fourth session, people who had written alone and then used ChatGPT showed higher connectivity than the ChatGPT group had shown in its first three sessions. People who had used ChatGPT and then wrote alone showed weaker connectivity.
What it did not find
- Nothing about IQ. The paper contains no intelligence test.
- No damage. EEG during a task shows how hard the brain was working on that task. It does not show harm, and it cannot see deep brain structures.
- No long-term effect. Four sessions. The swap session had 18 people, about 9 in each direction.
- Nothing beyond essays and ChatGPT. One task, one tool, one group of students from one area.
The authors list these limits themselves. Their FAQ asks journalists not to use "stupid", "dumb", "brain rot", "harm", "damage", "passivity", "trimming", "brain scans" or "brain damage" about the findings.
The number that gets misquoted
You will see "83% of ChatGPT users cannot remember what they wrote". The real figure is 15 of 18 people, in one session, who could not quote one sentence. Fifteen out of eighteen is 83.3%, but it is fifteen people, not 83% of ChatGPT's users.
A fair one-line summary. "A small MIT preprint found lower brain engagement and weaker recall when people wrote essays with ChatGPT, and better engagement when they wrote alone first."
Why the fourth session is the useful part
Most coverage stopped at "ChatGPT group, weakest brain activity". The more practical result is the swap. Writing on your own first, then bringing in the AI, went with more engagement than starting with the AI. It is a small sample and it needs repeating. But it points the same way as larger trials: a field trial with about 1,000 students found that AI which gave hints avoided the exam drop seen with AI which gave answers.
What to do with this
For anything you want to remember or own, draft first. A rough paragraph is enough. Then ask the AI for feedback on your draft instead of a draft of its own.
Questions
Is the MIT ChatGPT study peer reviewed?
Not as of 13 November 2026. It is a preprint on arXiv. The authors' FAQ says peer review has started and is at the very beginning.
How many people were in the MIT ChatGPT study?
54 in total, 18 in each of three groups. 18 came back for the fourth session.
Does the study say ChatGPT makes you dumber?
No. The authors reject that wording. The study shows lower engagement during one kind of task, not a change in ability.
What does "cognitive debt" mean in the title?
It is the authors' phrase for the idea that effort you skip now may cost you later, when you need to do the task alone. The study suggests it. It does not prove it.