Cognitive Offloading: What to Hand to AI When You Study

You put deadlines in a calendar so you don't have to hold them in your head. You save the lecture slides so you don't have to remember where they are. That's cognitive offloading, and you do it all day. AI lets you offload far more: the summary, the problem set, the outline, the explanation. Some of that helps you learn. Some of it quietly takes the learning with it.

What cognitive offloading is

Psychologists define it as any physical action that changes a task so it needs less mental effort (Risko & Gilbert, 2016). Writing a shopping list counts. So does setting a reminder, counting on your fingers, or searching for a date instead of recalling it.

None of this is new, and most of it is sensible. Your working memory is small. Anything that holds information for you leaves more of it free for the task in front of you. The question is never whether to offload. It's what you're handing over.

When offloading helps you learn

In one set of experiments, students studied a file of words, then either saved it or closed it without saving, and then studied a second file. They remembered the second file better when the first one had been saved. When saving couldn't be trusted, the benefit disappeared (Storm & Stone, 2015).

Your memory lets go of what it trusts is kept somewhere safe, and that makes room for the next thing. For a student, that means getting the slides, the recordings, the readings, and the deadlines out of your head and into one place you trust isn't laziness. It's what lets you listen to the lecture instead of worrying about losing it.

When offloading replaces the learning

Learning comes from producing, not receiving. In a classic experiment, people who came up with a word themselves from a cue (rapid: f___) remembered it better than people who simply read the pair (rapid: fast). It's called the generation effect (Slamecka & Graf, 1978). When a tool produces the answer, the explanation, or the essay, you get the result without doing the part that builds memory.

A large field experiment shows what that looks like with AI. Nearly 1,000 high school students in Turkey had four 90-minute math practice sessions with either their notes and textbook, a standard ChatGPT-style assistant, or a version designed as a tutor that gave hints instead of answers. During practice, the standard assistant group scored 48% higher than the notes group, and the tutor group 127% higher. Then the AI was taken away for the exam. The standard assistant group scored 17% lower than students who never had it. The tutor group scored about the same as the notes group. Students in the standard group didn't feel they had learned less (Bastani et al., 2025).

The AI didn't make them worse at math. It did the part of the practice that was supposed to teach them, and they couldn't tell.

A small MIT Media Lab study points the same way. Of 54 people writing essays, the group that wrote with ChatGPT fell behind at quoting the essays they had written minutes earlier (Kosmyna et al., 2025). It's an unreviewed preprint on a single task, so treat it as a hint, not a finding.

Offloading feeds itself

Handing something off makes it easier to hand off the next thing. In one study, people answered hard trivia questions either from memory or with Google, then got a set of easy questions they could look up or answer themselves. The people who had used Google were more likely to reach for it again, and about 30% of them didn't try a single easy question from memory first. The pattern held even when searching meant using an older, slower device (Storm, Stone & Benjamin, 2017).

A survey of 666 people found that heavier use of AI tools went with lower critical thinking scores, and that offloading was the link between them (Gerlich, 2025). That's a correlation. It doesn't show that AI caused the drop, and people who think less critically may also lean on AI more. It's still worth noticing which way the habit pulls.

How to tell the difference

One question sorts most cases: if the tool disappeared tomorrow, would you still be able to do this?

For storage and logistics, it doesn't matter. Nobody needs to memorize where the slides are. For anything your exam will ask you to do, it matters a lot.

Fine to hand off:

• Keeping your slides, recordings, readings, and links in one place
• Finding things, like where the professor covered the Coase theorem
• Transcribing a lecture so you can listen instead of typing
• Keeping track of what to review and when
• Writing the practice questions (as long as you answer them)

Keep for yourself:

• Working the problem before you see the solution
• Explaining the idea in your own words
• Recalling the answer before you check it
• Writing the essay, the proof, or the argument
• Judging whether an answer is right

How to use AI without losing the learning

• Try first, then ask. Attempt the problem on your own. Then ask the AI to check your work or give you a hint, not the answer.
• Ask it to quiz you. "Quiz me on the causes of the 1848 revolutions" does more for you than "Explain the causes of the 1848 revolutions."
• Ask where the answer came from. An answer that points to the page or slide is one you can check, and checking is thinking.
• Close it and say it back. When an explanation clicks, close the chat and explain it in your own words. If you can't, you've found what to study. The Feynman technique works the same way.

How Repeatica handles it

Repeatica is built around that line. It takes the bookkeeping. Your lectures, slides, readings, and saved pages live in one Library. Repeatica transcribes your recordings, and Ask Repeatica finds what you need and answers with the page, slide, or moment it came from, so you can check it.

The thinking stays with you. Lessons in a course end with practice you answer yourself, and Repeatica will quiz you on anything when you ask. With Remember, Repeatica keeps track of what to bring back and when; when a question comes back, you answer it, then choose Missed it or Got it. Repeatica won't write your assignments.

The repeating is Repeatica's job. Learning it once is yours.

What we don't claim

• That offloading to AI is bad. Most of it is fine, and some of it helps you learn.
• That AI damages your brain. None of these studies shows lasting harm.
• That the high school results transfer exactly to university. It was one school, one subject, and a few sessions. The pattern, that doing the work for you feels like learning and isn't, fits a lot of other research.
• That Repeatica improves grades or memory by any amount. We haven't measured that yet, so we don't put a number on it.

FAQ

What is cognitive offloading?

Using something outside your head, like a note, a calendar, a search engine, or an AI, to reduce the mental effort a task takes. The term comes from cognitive psychology, where Evan Risko and Sam Gilbert reviewed the research in 2016.

Is cognitive offloading bad?

No. Offloading storage and logistics frees your mind for the work that matters, and saving information can even help you learn what comes next. It becomes a problem when you offload the thinking the task was meant to build, like solving, explaining, or recalling.

Does using ChatGPT make you worse at learning?

It can if it does the work for you. In a field experiment with high school math students, those who practiced with a standard ChatGPT-style assistant scored 17% lower on the exam than students without it. A version that gave hints instead of answers avoided that drop.

How do I study with AI without it doing the work for me?

Try the problem before you ask, ask for hints or a quiz instead of answers, check where each answer came from, and explain what you learned in your own words once the chat is closed.

Keep reading

• Desirable difficulties: why good studying feels hard
• Active recall: the study technique most students skip
• The Feynman technique: how to learn by explaining
• Why you forget what you studied after the exam

---

Repeatica is the agentic study space for students. Put everything from your classes in one place. Repeatica gets you through the exam, and makes sure you keep what you learned. Learn it once.

Sources

• Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), e2422633122. doi.org/10.1073/pnas.2422633122
• Gerlich, M. (2025). AI tools in society: Impacts on cognitive offloading and the future of critical thinking. Societies, 15(1), 6. doi.org/10.3390/soc15010006
• Kosmyna, N., et al. (2025). Your brain on ChatGPT: Accumulation of cognitive debt when using an AI assistant for essay writing task. arXiv preprint. arxiv.org/abs/2506.08872
• Risko, E. F., & Gilbert, S. J. (2016). Cognitive offloading. Trends in Cognitive Sciences, 20(9), 676-688. doi.org/10.1016/j.tics.2016.07.002
• Slamecka, N. J., & Graf, P. (1978). The generation effect: Delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4(6), 592-604. doi.org/10.1037/0278-7393.4.6.592
• Storm, B. C., & Stone, S. M. (2015). Saving-enhanced memory: The benefits of saving on the learning and remembering of new information. Psychological Science, 26(2), 182-188. doi.org/10.1177/0956797614559285
• Storm, B. C., Stone, S. M., & Benjamin, A. S. (2017). Using the Internet to access information inflates future use of the Internet to access other information. Memory, 25(6), 717-723. doi.org/10.1080/09658211.2016.1210171