Cognitive Offloading: How Digital Tools Change Memory and Learning

Cognitive offloading is what you do every time you let something outside your head carry part of the thinking: a note instead of remembering, a search instead of recalling, a calculator instead of working it out, an AI instead of drafting. It is ordinary, it is ancient, and it usually works. The interesting question is not whether to do it. It is what changes inside you when you do.

The honest answer has two halves that are often mashed together. Offloading generally helps performance right now, because the tool is good at the bit you handed over. What it changes is what you end up remembering or practising yourself, which is a different thing and matters most when the point of the task was to learn. A shopping list you never memorise costs you nothing. A concept you never retrieved costs you the concept.

This guide separates three things that get confused: what happens when you actually use a device while learning, which is well established; what happens when a device is merely sitting there, which is not; and what happens when you think alongside an AI, which is genuinely open. It carries no new research. Every claim summarises something already sourced in one of the four articles it links to.

Using the device: divided attention while learning

If you split your attention while learning, you learn less. This is the sturdiest result here, and it is about active use, not proximity. Rosen, Carrier and Cheever (2013) observed 263 students studying at home and found that most switched away within minutes, and that the switching predicted worse academic performance. Uncapher and Wagner's 2018 review sets out the associated costs to working memory and sustained attention, while being clear that the literature is still sparse and that many studies find no group differences at all.

Sana, Weston and Cepeda (2013) added the detail people forget: laptop multitasking harmed comprehension for the multitasker and for students seated in view of the screen. It is worth knowing what that study was. It was a randomised experiment in a simulated lecture with a small sample and a single immediate comprehension test, not a measurement of an ordinary term's classroom learning.

Leroy's attention-residue work (2009) is often used to explain why. It is a reasonable explanation, but it should be labelled as an inference: Leroy tested adults switching between unfinished work tasks, not students revising, and a documented retrieval located no study applying attention residue directly to studying.

Why studying next to your phone wrecks your memory works through all of this, and is careful about where the evidence changes character.

Merely having it there: the claim that has not held up

The most shareable version of this story is that a phone face-down on the desk, untouched, drains cognitive capacity. That comes from Ward and colleagues (2017), the "brain drain" study, and it is worth tracking what happened to it, because the sequence is unusually clear.

A preregistered direct replication by Ruiz Pardo and Minda (2022) found no difference between smartphone-location conditions on either task. Then three meta-analyses arrived. Böttger and colleagues (2023) pooled 22 studies and found a small significant overall effect (g = -0.14), but driven by memory tasks, non-significant for attention, and significant in Asian samples while not significant in North American ones. Parry (2024) pooled 56 effects from 7,093 participants across six separate models and found significance only for working-memory capacity, with null summary effects for the other cognitive functions examined, alongside substantial methodological heterogeneity and generally poor statistical power. Hartanto and colleagues (2024), the largest by effect sizes, pooled 166 effects from 53 samples and 33 studies covering 4,368 participants and found no significant overall effect at all (d = -0.02, 95% CI -0.06 to 0.01). None of their moderators survived either, and the one task that did reach significance stopped doing so after correction for publication bias.

Put together: the clearest recent meta-analyses do not support a general claim that a silent phone's mere presence reliably reduces cognition. A small effect on working memory specifically remains possible, and it is the one signal that keeps reappearing, but it is fragile: it fell below the threshold its own author had set as meaningful in one analysis, and disappeared under bias correction in another. The literature is too heterogeneous and too underpowered to rule that small effect out. So neither "your phone drains your brain" nor "this has been debunked" is a fair summary. What is fair is that using the device while you learn has a demonstrated cost, while merely having one on the desk has not been shown to reliably reduce cognition.

Offloading the method: why hours of study leave nothing behind

There is a second kind of offloading that has nothing to do with devices. Rereading and highlighting hand the work of retrieval over to the page. The material feels familiar, and familiarity feels like knowing.

Dunlosky and colleagues' 2013 review graded ten common techniques and found rereading and highlighting, the two most popular, to have low utility, while practice testing and distributed practice rated high. Interleaving, often bundled with them, was rated only moderate in the same review.

The supporting work is consistent. Roediger and Karpicke (2006) showed the testing effect at delays of days and weeks, though restudying won at very short delays. Cepeda and colleagues' 2006 meta-analysis established spacing across 317 experiments of verbal recall, with the useful detail that the best gap scales with how long you need to remember. Rohrer and Taylor (2007) found that shuffling mathematics problems improved learning, and Kornell and Bjork (2008) found that spacing beat massing while about three-quarters of participants judged the opposite, which is why the methods that work often feel worse while you use them.

Karpicke and Blunt (2011) is the strongest single claim in this area and also the most contested: retrieval practice outperformed elaborative concept mapping in Science, but a published Comment argued the concept-mapping group had far too little training, and a later reanalysis reports the advantage disappearing for verbatim questions once time and instructions were matched. The broad principle stands; that specific head-to-head comparison does not.

You studied for hours and remember nothing covers the method problem. Study tools and apps actually worth it asks which software genuinely implements those principles. Anki's own documentation describes its spaced-repetition scheduling in detail, including the algorithms it uses. Brilliant's documentation describes interactive problem-solving and does not claim a spaced-repetition method, so the two should not be treated as carrying the same evidence.

Offloading the thinking itself

Generative AI can offload more of a task than earlier tools, but evidence about what that changes in unaided thinking remains limited.

Gerlich (2025) is a cross-sectional survey of 666 people with self-reported measures, and it carries a published Correction, so it can show an association and nothing about direction. Kosmyna and colleagues' 2025 MIT study is a preprint that has not been peer reviewed; 54 participants took part in the first three sessions and only 18 completed the fourth session from which the most-quoted claims come, and the authors have publicly asked readers not to describe it as showing brain damage.

Against those, Han, Peng and Liu's 2025 meta-analysis of 68 experiments found a moderate overall benefit for learning with generative AI. The moderators it actually reports are education level, discipline, intervention duration and sample size, with larger effects in shorter interventions and smaller samples, a pattern that is itself consistent with small-study bias. UNESCO's 2023 guidance recommends how institutions should govern these tools; it is policy, not evidence about learning outcomes.

How to use AI and keep thinking holds all of those against each other rather than picking the most dramatic one.

What the evidence does not settle

• Whether a phone's mere presence costs you anything. One original finding, one failed preregistered replication, and three meta-analyses pointing at somewhere between a small working-memory effect and nothing. Unresolved, and closer to nothing than the headlines suggest.
• Whether AI use harms thinking. The alarming studies are a cross-sectional survey and a small unreviewed preprint. The reassuring meta-analysis reports a moderate average benefit but does not test whether AI helps more when it supports thinking than when it supplies answers. Nothing here settles direction of causation.
• Whether offloading changes what you retain over years. Almost all of this literature runs on short tasks with student samples, and the spacing meta-analysis is restricted to verbal recall. Long-horizon effects on what a person can do unaided are essentially unmeasured.
• Whether retrieval practice beats concept mapping specifically. The principle is well supported. That particular comparison is disputed in the published literature.
• Whether any specific study app produces better retention. Spacing and retrieval practice are well supported as principles. Evidence that a particular product delivers them better than an index card is not.

Where to start

If you are trying to study and it is not sticking, start with You studied for hours and remember nothing, because the method is more often the problem than the device. If you are studying next to a phone you keep picking up, read Why studying next to your phone wrecks your memory, and note that the fix is about use rather than proximity. If you are choosing software, read Study tools and apps actually worth it. And if you are working alongside an AI, read How to use AI and keep thinking.

The rule that survives all of it is short. Offload the parts you never needed to hold, and keep doing the part you actually wanted to get better at.