Harold Robert Meyer
Reviewed: September 10, 2026 · Published: September 16, 2026
People have been asking whether machines are ruining their memory since well before ChatGPT. The honest answer to whether generative AI is different starts with what changed, and what didn’t.
Ask most people whether Googling something instead of remembering it “counts,” and they’ll shrug. Ask whether letting ChatGPT write their memo or draft their thinking counts the same way, and the shrug gets harder to hold. The two aren’t the same act, and the research on each is starting to show why.
Key Takeaway: Offloading information to a tool has been reshaping memory for over a decade; offloading the thinking itself to generative AI appears to do it faster and more deeply — but the effect depends on how the tool is used, not just whether it’s used.
Key Findings
- In a 2025 MIT Media Lab study, participants who wrote essays with ChatGPT showed the weakest EEG-measured neural connectivity of any group, and 83% could not accurately quote a sentence from the essay they had just written.
- A 2011 Science study found people recall information less well when they expect future access to it, remembering instead where to find it — the “Google effect,” a form of transactive memory.
- A 2026 study distinguishes “dependent” offloading (letting AI replace your thinking) from “autonomous” offloading (using AI as a scaffold while staying in control) — the two are not equally costly to cognition.
- Working memory deficits are well documented in ADHD research, independent of any AI exposure at all.
Cognitive offloading is not new. Notebooks, calculators, and search engines have always let people trade internal effort for external support, and the brain adapts by remembering where to find things instead of the things themselves — the pattern Betsy Sparrow, Jenny Liu, and Daniel Wegner documented in their well-known 2011 Google effect study. That’s a shift in what memory does, not a loss of the capacity to do it.
Generative AI is different in kind, because it doesn’t just store the product of thinking — it produces it. That’s the distinction a 2025 MIT Media Lab team, led by Nataliya Kosmyna, set out to measure. Fifty-four participants wrote essays across four months under three conditions: unassisted, using a search engine, or using ChatGPT. The EEG results were stark: brain connectivity in networks associated with memory and attention was strongest in the unassisted group, moderate in the search-engine group, and weakest in the ChatGPT group. Most strikingly, 83% of ChatGPT users could not accurately quote a line from the essay they’d written minutes earlier. When some participants were later asked to write unassisted after months of AI use, their brains showed under-engagement, as if the habit of not encoding had outlasted the tool itself.
That’s a small study measuring one writing task over four months, not a verdict on memory or intelligence in general, and it’s worth saying plainly since some coverage has run further than the data does. But it converges with a broader finding: offloading to AI isn’t uniform. A 2026 study distinguishes offloading that keeps you in the loop — using AI to check your own draft or push back on your reasoning — from offloading that hands over the whole cognitive job and accepts the output with little scrutiny. The first pattern looks a lot like using a calculator. The second looks more like never learning arithmetic.
A Note for Clinicians and Prescribers: The Rapport et al. study cited below found working memory deficits in ADHD across every subsystem tested — phonological, visuospatial, and central executive — even after controlling for reading speed, IQ, and age, resolving what had been a genuinely mixed prior literature. That history matters here, because it means the AI-and-memory question doesn’t land neutrally for people with ADHD. A tool that reliably handles the part of a task working memory struggles with could be a real accommodation. Watch for patients substituting AI use for medication or skills work they’re avoiding, versus using it as a scaffold alongside those supports.
“In thirty years of coaching adults with ADHD, I’ve watched compensatory tools go one of two ways. Some free up enough working memory that people finally get traction. Others become one more thing they can’t function without. The difference is never the tool. It’s whether the person is still doing any of the thinking themselves.”
No ADHD-specific study yet answers whether AI use nets out as accommodation or atrophy for this population. Until there is, the practical move is the same one that applies to anyone: use AI to check your thinking, not to skip it. Try to answer or draft something yourself first, even badly, before asking AI to do it — retrieval attempts build the memory trace, not exposure to a finished answer. If AI produces something for you, rewrite or explain it in your own words rather than passing it along unread; that step is what turns someone else’s output into your own encoding. And periodically do the task with no tool at all, the way the MIT study’s “brain-to-LLM” group did, just to see what’s still there.
The next step: if you use AI regularly for work, writing, or schoolwork, try one week of “attempt first” before you ask it anything, and notice whether your first drafts get easier or harder to produce.
Synopsis
A look at whether generative AI use erodes memory, distinguishing the well-established “Google effect” of offloading information from the newer and more concerning finding that AI use may reduce engagement in the neural processes that encode memory in the first place. Draws on MIT’s 2025 EEG study of ChatGPT-assisted writing and the emerging distinction between AI use that scaffolds thinking versus replaces it, with a note on why this question carries extra weight for people with ADHD given the population’s documented working memory deficits.
Topic: Cognitive Function / Technology & ADHD (pending Section 11 assignment — flagging for your review)
Audience: Adults; Clinicians & Prescribers; Educators
Keywords: cognitive offloading, working memory, cognitive debt, transactive memory, AI dependence
Bibliography
- 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.
- Rapport, M. D., Alderson, R. M., Kofler, M. J., Sarver, D. E., Bolden, J., & Sims, V. (2008). Working memory deficits in boys with attention-deficit/hyperactivity disorder (ADHD): The contribution of central executive and subsystem processes. Journal of Abnormal Child Psychology, 36(6), 825–837.
- 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.
- Zhu, Li, Dong, Chang, & Fan (2026). Not all cognitive offloading is equal: Distinguishing dependent and autonomous offloading to generative AI. Frontiers in Psychology, 17. ??? — author first initials unverified; last names and DOI confirmed via the journal’s own copyright line.
About the Author
Harold Robert Meyer founded The ADD Resource Center in 1993 and has spent more than thirty years translating the lived experience of ADHD into practical guidance for individuals, families, and the clinicians and educators who work with them. He co-founded CHADD of New York in 1989 and went on to serve on CHADD National’s board and executive committee, including as Vice President of Finance. He was later elected to lead the Institute for the Advancement of ADHD Coaching. An author and international speaker, he has presented at the American Psychiatric Association Annual Meeting, CHADD national and local conferences, NYU Langone, Mount Sinai Medical Center, and Weill Cornell Medical College. Reach him at [email protected].
📚 Come back often – we add new evidence-based ADHD resources daily.
The ADD Resource Center
[email protected] · +1 (646) 205-8080 · 127 West 83rd St., Unit 133, New York, NY 10024-0840 USA
X · LinkedIn · Facebook · Substack · Newsletter
Please consider the environment before printing.
Our content is educational and informational only – not medical, legal, financial, or other professional advice, nor a substitute for it, and should not be relied on as such. We strive for accuracy, but errors or omissions occur and information becomes outdated. Some content and images may be created with AI assistance and, though reviewed before publication, may contain inaccuracies. Our screening tools are not diagnostic instruments. Verify independently and consult a qualified professional before acting on anything here.
In the USA and Canada, call or text 988 anytime for free mental health and suicide prevention support, or call 911 if someone is in immediate danger. The ADD Resource Center is independent from these services and is not liable for any actions taken by you or by these services.
Under GDPR and CCPA, you may request access to, correction of, or deletion of your personal data at [email protected].
© 2026 The ADD Resource Center