Your Brain on ChatGPT — Kosmyna 2025 Deep Dive

Summary

Kosmyna et al. (2025), 'Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task,' arXiv:2506.08872.

EEG study; n=54 across Sessions 1–3 and n=18 for Session 4.

Brain-only participants showed the strongest, most distributed neural networks. Search-engine users showed moderate engagement. LLM users showed the weakest connectivity.

Self-reported essay ownership was lowest in the LLM group and highest in the Brain-only group.

LLM users struggled to accurately quote their own work.

The Stanković et al. (2026) comment (arXiv:2601.00856) flags small-n, EEG methodology, and reproducibility concerns. We cite it alongside Kosmyna for intellectual honesty.

What did the MIT cognitive-debt study actually find?

The 2025 Kosmyna paper measured EEG connectivity, self-reported ownership, and quotation accuracy in essay writers under three conditions — brain-only, search-engine, and LLM. The brain-only group showed the strongest, most distributed neural networks during writing. LLM users showed the weakest connectivity, the lowest self-reported essay ownership, and the most difficulty quoting their own work. The paired Stanković et al. (2026) comment flags the small sample size and EEG methodology questions. The honest read: it's a directionally important finding that needs replication at scale, not a settled clinical result.

What the source says

The paper's headline numerical results are concrete: brain-only participants showed the strongest, most distributed EEG networks during essay writing; search-engine users showed moderate engagement; LLM users showed the weakest connectivity. On self-report, the LLM group reported the lowest ownership of their essays and the Brain-only group reported the highest. On quotation, LLM users had the most trouble quoting from the essays they had just written.

Armitage (Br J Gen Pract 2025;75(758):410, PMC12723506) reported secondary numbers that were widely cycled in press — up to 55% reduced connectivity in the LLM group and 83% being unable to quote from the essays they had just written. These commentary numbers travelled faster than the primary paper; cite the primary numbers from Kosmyna directly.

The MIT cognitive-debt study (with its critique).

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:2506.08872. arxiv.org/abs/2506.08872. n=54 (Sessions 1–3), n=18 (Session 4); EEG result: “Brain-only participants exhibited the strongest, most distributed networks; Search Engine users showed moderate engagement; and LLM users displayed the weakest connectivity.” Self-reported essay ownership was lowest in the LLM group and highest in the Brain-only group; LLM users also struggled to accurately quote their own work.

Secondary commentary: Armitage R., Br J Gen Pract 2025; 75(758):410, PMC12723506 — reports up to 55% reduced connectivity in the LLM group and 83% being unable to quote from the essays they had just written.

Status: arXiv preprint, not yet peer-reviewed as of May 2026.

Paired critique: Stanković, M., Hirche, E., Kollatzsch, S., & Doetsch, J.N. (2026). Comment on: Your Brain on ChatGPT. arXiv:2601.00856. arxiv.org/abs/2601.00856. Flags the small sample size, EEG methodology questions, and reproducibility limits — we cite it alongside Kosmyna for intellectual honesty.

What the source does not say

The paper does not show that ChatGPT use causes lasting cognitive harm. It documents real-time differences during an essay-writing task under three conditions. It does not measure durable cognitive change, intelligence, or clinical outcome.

The sample size is small — n=54 across Sessions 1–3 and n=18 for Session 4. The Stanković et al. comment flags this directly: EEG analysis at this scale has well-known reproducibility limits, and the headline numbers should not be read as settled.

The headline framing of “ChatGPT makes you dumber” is not what the paper says. The paper says LLM-assisted essay writing engages weaker neural networks than unassisted essay writing, that essay ownership is felt less when an LLM generates the draft, and that quoting one's own work is harder when the work was generated rather than written.

What this means for daily practice

For daily practice, the honest take: keep some cognitive work unmediated. Not because AI is bad — because the published evidence we have so far suggests that the daily volume of unmediated practice is what maintains the underlying skills. The Senwitt daily Set is the seven-minute counterweight to a ChatGPT-heavy workday.

The result also has to be read in context. The cognitive-offloading tradition has documented similar patterns long before LLMs existed — Sparrow et al.'s 2011 Science paper on the Google effect (Sparrow, 2011), the UCL 2020 paper on GPS and spatial memory (Dahmani & Bohbot, 2020). The Kosmyna result is a specific 2025 data point in a longer line, not a stand-alone discovery. The right reading uses it as one input among several.

For citations, both Kosmyna 2025 and Stanković 2026 must appear together. See our full AI brain fry blog for the working-life version of the same point, and the cognitive debt research page for the cross-domain version.

From Senwitt · advertisement

The text above is editorial. What follows is a promotional message from Senwitt, the maker of this site. Senwitt is a brain-exercise app and is not a medical product. Read the full disclaimer in the footer.

References — canonical order.

1.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:2506.08872. arxiv.org/abs/2506.08872. Primary anchor.

2.Stanković, M., Hirche, E., Kollatzsch, S., & Doetsch, J.N. (2026). Comment on: Your Brain on ChatGPT. arXiv:2601.00856. arxiv.org/abs/2601.00856. The methodological critique — paired with Kosmyna.

3.Risko, E.F. & Gilbert, S.J. (2016). “Cognitive Offloading.” Trends in Cognitive Sciences 20(9):676–688. DOI: 10.1016/j.tics.2016.07.002.

4.Sparrow, B., Liu, J., & Wegner, D.M. (2011). “Google Effects on Memory: Cognitive Consequences of Having Information at Our Fingertips.” Science 333(6043):776–778. DOI: 10.1126/science.1207745.

5.Simons, D.J., Boot, W.R., Charness, N., Gathercole, S.E., Chabris, C.F., Hambrick, D.Z., et al. (2016). “Do ‘Brain-Training’ Programs Work?” Psychological Science in the Public Interest 17(3):103–186. DOI: 10.1177/1529100616661983.

6.FTC v. Lumos Labs, Inc. (2016). “Lumosity to Pay $2 Million to Settle FTC Deceptive Advertising Charges for Its ‘Brain Training’ Program.” Stipulated $50M judgment, suspended on payment of $2M. ftc.gov press release (Jan 5 2016).

7.Max Planck Institute for Human Development & Stanford Center on Longevity (2014). A Consensus on the Brain Training Industry from the Scientific Community. Signed by 70 neuroscientists/psychologists. longevity.stanford.edu.

Not brain training. Brain exercise.

Senwitt is a daily brain exercise app, not a brain training program. We do not claim to improve general cognition, prevent cognitive decline, or treat any condition. Independent scientific consensus — the 2014 Stanford Center on Longevity / Max Planck Institute statement signed by 70 neuroscientists, the 2016 Simons et al. review in Psychological Science in the Public Interest, and the FTC's 2016 settlement with Lumos Labs — has concluded that “brain training” claims are not supported by the evidence. Senwitt is built on a different premise: skills you actively practice get sharper; skills you stop practicing fade.

Why we avoid old brain-training claims

Your Brain on ChatGPT — what the MIT cognitive-debt paper actually found.