Table of Contents
Your Brain on ChatGPT: What the MIT Study Found About Ownership
Your Brain on ChatGPT, read carefully: MIT found AI writers felt least ownership and couldn't quote their essays. The critique, and why teens are missing.
The most-cited finding from MIT’s “Your Brain on ChatGPT” study isn’t about brain waves. It’s this: adults who wrote essays with ChatGPT felt the least ownership of what they wrote, and many couldn’t accurately quote their own essays minutes later. The study, posted in June 2025 by Nataliya Kosmyna and colleagues at the MIT Media Lab, also reported that ChatGPT users showed the weakest brain connectivity of three groups. That brain result made headlines, and it’s the part most open to challenge. The study has 54 participants, is not peer-reviewed, and drew a published methodological critique within months. Then, in July 2026, two researchers pointed out the biggest gap of all: every participant was an adult. Teenagers, whose planning and reasoning skills are still being built, weren’t in it. Here’s what the study did, what holds up, what doesn’t, and what parents of 10-to-17-year-olds should take from it.
Key Takeaways
- 54 adults wrote essays over three sessions with ChatGPT, a search engine or no tools; 18 returned for a fourth session in which some switched conditions.
- “Self-reported ownership of essays was the lowest in the LLM group and the highest in the Brain-only group,” and “LLM users also struggled to accurately quote their own work.”
- EEG showed the strongest, most distributed brain networks in the no-tools group and the weakest in ChatGPT users, but a critique flags sample size, reproducibility and EEG-analysis concerns.
- The study is a preprint. Lower brain activity can also signal efficiency, so “less connectivity” does not automatically mean “worse thinking.”
- A July 2026 paper in Frontiers in Developmental Psychology calls adolescents the “missing” group and proposes that whether AI scaffolds or substitutes for a teen’s effort is the key question.
What the Your Brain on ChatGPT Study Did
“Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task” was posted to arXiv on June 10, 2025, and revised on December 31, 2025. The authors, led by Nataliya Kosmyna with Pattie Maes among the co-authors, are at the MIT Media Lab. The full paper runs 216 pages.
The design. Participants were split into three groups: one wrote essays using a large language model (LLM) chatbot, one used a search engine, and one used no tools, called “Brain-only.” Each wrote in three sessions under the same condition. In a fourth session, 18 participants switched: LLM users wrote without tools (“LLM-to-Brain”), and Brain-only users got the chatbot (“Brain-to-LLM”).
The measures. Researchers recorded brain activity with EEG, analyzed the essays with natural language processing, and had them scored by human teachers and an AI judge. They also interviewed participants about their essays.
How EEG connectivity works. EEG uses electrodes on the scalp to record the brain’s electrical activity. “Connectivity” measures how strongly signals at different sites rise and fall together, or in this study, using a method called dDTF, how activity at one site predicts activity at another. Higher connectivity is generally read as more coordinated engagement across brain regions. It is an indirect measure, and interpreting it is where much of the debate lies.
What It Found
From the paper’s abstract:
- Brain connectivity scaled with tool use. “Brain-only participants exhibited the strongest, most distributed networks; Search Engine users showed moderate engagement; and LLM users displayed the weakest connectivity.”
- Switching mattered. In session 4, “LLM-to-Brain participants showed reduced alpha and beta connectivity, indicating under-engagement.” Brain-to-LLM users “exhibited higher memory recall” and activation patterns similar to search-engine users.
- Ownership. “Self-reported ownership of essays was the lowest in the LLM group and the highest in the Brain-only group.”
- Memory. “LLM users also struggled to accurately quote their own work.”
- Homogeneity. Essays within each group tended to be similar to each other in named entities, phrasing and topics.
The authors’ summary is strong: “Over four months, LLM users consistently underperformed at neural, linguistic, and behavioral levels.” They coined “cognitive debt,” the idea that effort you skip now is paid for later. We look at that concept on its own in cognitive debt: what kids lose when AI does the thinking.
The Critique: What Doesn’t Hold Up as Well
In late December 2025, Milos Stanković, Ella Hirche, Sarah Kollatzsch and Julia Nadine Doetsch posted a formal comment on arXiv. It’s polite (it opens by congratulating the authors) and pointed. Their concerns:
- “study design considerations, including the limited sample size”
- “the reproducibility of the analyses”
- “methodological issues related to the EEG analysis”
- “inconsistencies in the reporting of results”
- “limited transparency in several aspects of the study’s procedures and findings”
They say some results “could be interpreted more conservatively.”
There’s a second, more basic interpretive problem. Lower brain activity doesn’t always mean worse thinking. Jorge Pereira Campos and Tatiana Koff, in a July 2026 paper in Frontiers in Developmental Psychology, note that “lower neural activity need not mean worse cognition; in studies of expertise it often marks more efficient processing.” An expert pianist’s brain works less hard on a scale than a beginner’s.
So, in fairness: the brain-connectivity finding is suggestive, not settled. The ownership and memory findings are on firmer ground, not because this study is stronger there, but because they match decades of earlier research.
Why the Ownership and Memory Findings Fit Older Research
The ownership result echoes the “AI Ghostwriter Effect” reported by Fiona Draxler and colleagues in ACM Transactions on Computer-Human Interaction: across two studies, people “do not consider themselves the owners and authors of AI-generated text,” and more influence over the text meant more ownership.
The memory result echoes the generation effect. In 1978, Norman Slamecka and Peter Graf showed across five experiments that people remember words they generate better than words they read. When the chatbot generates the essay, the writer is mostly reading. Not being able to quote it afterward is what that research would predict.
Claims vs. Evidence, at a Glance
| Claim you may have seen | What the study actually shows | How much weight to give it |
|---|---|---|
| ”ChatGPT damages your brain” | Lower EEG connectivity while writing with a chatbot | Low: indirect measure, contested analysis, may reflect efficiency |
| ”AI users don’t feel the work is theirs” | Lowest self-reported ownership in the LLM group | Moderate: small sample, but consistent with other studies |
| ”AI users can’t remember what they wrote” | LLM users struggled to quote their own essays | Moderate: fits the long-established generation effect |
| ”The effects last” | LLM-to-Brain switchers showed under-engagement in session 4 | Low: 18 people, one session |
| ”This applies to kids” | All participants were adults | Not tested; an open question |
The Missing Group: Teenagers
The Frontiers paper’s title asks the right question: “Your brain on ChatGPT, but whose brain?” Campos and Koff point out that the participants “were adults, whose executive functions, unlike those of adolescents, were already largely built.” Executive functions are the mental skills behind planning, holding information in mind, weighing evidence and controlling impulses. “Between ages 10 and 20, executive functions undergo rapid, experience-dependent maturation supported by the still-developing prefrontal cortex.”
Their central idea is a “cognitive offloading paradox”: generative AI is most useful for exactly the tasks, “such as planning, evaluating evidence, and constructing arguments,” whose effortful practice is thought to build those skills in adolescence. They offer it explicitly as a hypothesis, not a finding.
They also draw a distinction parents can use: AI as a scaffold (it asks questions, gives feedback, prompts revision, so the student still does the thinking) versus AI as a substitute (it produces the output). They argue that distinction “should be the central organizing question for the next phase of research,” and propose replicating the MIT study with 14-to-17-year-olds.
They’re also evenhanded. They cite meta-analyses finding AI assistance tends to improve academic performance and higher-order thinking, with benefits clearest “when the AI scaffolds a student’s thinking rather than doing the work for them.”
What Parents Can Do With This
Ask your teen to explain or quote their own work
The study’s most practical measure is one you can use at the kitchen table. After an essay is done, ask your teen to tell you their main argument and quote one line they’re proud of. If they can’t, the AI may have done more of the writing than they did.
Draft first, then bring in AI
The scaffold-versus-substitute line is easiest to hold if the first draft is theirs. AI can then give feedback, ask questions or check facts.
Treat ownership as the warning light
Kids who feel an assignment isn’t theirs disengage. If your child shrugs about work they “wrote,” that’s worth a conversation, whatever tool was involved.
Read headlines with the preprint label in mind
A small preprint with a published critique is a reason for attention, not alarm. Teaching your teen to ask “how many people, who were they, was it reviewed?” is a skill worth more than any single study. For a wider look at the research, see is ChatGPT making kids worse at thinking?
What not to do
Don’t tell your child “ChatGPT will rot your brain.” The evidence doesn’t support that, teens know it’s an exaggeration, and it costs you credibility for the parts that are well supported.
What to Watch For Over the Next 3 Months
- Week 4: Can your teen explain the argument of their last essay without looking at it? Is the first draft theirs?
- Month 2 red flags: Assignments finished unusually fast but explained poorly; your teen unable to quote or summarize their own work; indifference to grades on writing they supposedly did.
- Month 3 self-check: Watch for replication studies with teenagers, which the Frontiers authors called for. Until then, the scaffold-or-substitute question is the best guide you have.
Frequently Asked Questions
What did the “Your Brain on ChatGPT” study find?
In 54 adults writing essays, those using ChatGPT showed the weakest EEG brain connectivity, the lowest sense of ownership over their essays, and trouble quoting their own work, compared with search-engine users and people writing without tools. The study is a preprint that hasn’t been peer-reviewed.
Does ChatGPT damage the brain?
This study doesn’t show that. It measured lower brain connectivity while people wrote with a chatbot, which could reflect less engagement or more efficient processing. A published critique raised concerns about sample size and EEG analysis. The ownership and memory findings are better supported than the brain claims.
Did the MIT study include children or teens?
No. All participants were adults. A July 2026 paper in Frontiers in Developmental Psychology argues teens are the most important missing group, because the skills AI handles best are the ones still developing between about ages 10 and 20, and proposes testing 14-to-17-year-olds.
Is the study peer-reviewed?
Not as of its December 2025 revision; it’s posted on arXiv as a preprint. A formal comment by other researchers, also on arXiv, flagged design, reproducibility, EEG-analysis, reporting and transparency concerns.
Should my teen stop using ChatGPT for school?
Not necessarily. The more useful rule is scaffold, not substitute: your teen writes the first draft and does the reasoning, and AI asks questions, gives feedback or checks facts. If your teen can explain and quote their own work, the tool is probably supporting their thinking.
About the author
Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.
Sources
- Kosmyna, N., Hauptmann, E., Yuan, Y. T., Situ, J., Liao, X.-H., Beresnitzky, A. V., Braunstein, I., & Maes, P. (2025). “Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task.” arXiv (preprint, not peer-reviewed). https://arxiv.org/abs/2506.08872
- Stanković, M., Hirche, E., Kollatzsch, S., & Doetsch, J. N. (2025). “Comment on: Your Brain on ChatGPT: Accumulation of Cognitive Debt When Using an AI Assistant for Essay Writing Tasks.” arXiv (preprint). https://arxiv.org/abs/2601.00856
- Pereira Campos, J., & Koff, T. (2026). “Your brain on ChatGPT, but whose brain? The missing adolescent in AI-cognition research.” Frontiers in Developmental Psychology, 4. https://doi.org/10.3389/fdpys.2026.1885225
- Draxler, F., Werner, A., Lehmann, F., Hoppe, M., Schmidt, A., & Buschek, D. (2024). “The AI Ghostwriter Effect: When Users do not Perceive Ownership of AI-Generated Text but Self-Declare as Authors.” ACM Transactions on Computer-Human Interaction. https://doi.org/10.1145/3637875
- Slamecka, N. J., & Graf, P. (1978). “The generation effect: Delineation of a phenomenon.” Journal of Experimental Psychology: Human Learning and Memory, 4(6). https://doi.org/10.1037/0278-7393.4.6.592
- Common Sense Media. (2026). “Teens in the AI Era: Schoolwork and Skills That Matter.” https://www.commonsensemedia.org/research/teens-in-the-ai-era-schoolwork-and-skills-that-matter
