5 min read
Working memory is the mental scratchpad you use to hold a phone number for the six seconds between hearing it and dialling it, to keep a running total while you add up a bill in your head, or to remember the first half of a sentence long enough to understand the second half. It is one of the most-studied ideas in cognitive psychology, and also one of the most oversold by apps that promise to “boost your brainpower.” Here is what training working memory actually does, and what it does not.
Long-term memory is storage — years-old facts, faces, skills, the kind of thing you are not actively holding onto right now but can retrieve on demand. Working memory is different: it is a small, temporary holding space for information you are using this second, and it empties out fast once your attention moves elsewhere. The classic estimate is that it holds only a handful of items at once — commonly cited as somewhere around four, give or take, once you strip away memory tricks — which is a strikingly small number for something so central to thinking. Almost everything that feels like “concentration” or “juggling ideas” is really working memory doing the juggling.
Working memory is the bottleneck behind a huge range of everyday tasks, not because those tasks are inherently hard, but because they all demand holding several things in mind at once. Following multi-step instructions, doing arithmetic without paper, holding a plan while you’re distracted mid-task, understanding a long sentence — all of it leans on the same small scratchpad. When working memory is overloaded, the symptom rarely looks like “forgetting” in the usual sense; it looks like losing your train of thought, re-reading the same paragraph, or asking someone to repeat themselves because the first half of what they said fell out of the buffer before the second half arrived.
The claim that made “brain training” into a billion-dollar industry — that practising working-memory tasks makes you generally smarter, sharper, or better at unrelated mental work — has not held up well under scrutiny. Large, well-controlled studies consistently find that people get better at the specific task they practised and see little to no improvement on genuinely different tasks, even ones that also lean on working memory. If an app promises that five minutes a day of one narrow exercise will meaningfully raise your general intelligence or fix unrelated problems with attention, that is a much bigger claim than the evidence supports.
None of that means practice is worthless — it just works in a narrower, more specific way than the marketing suggests. You get reliably better at tasks that are structurally similar to what you practised, which is exactly why athletes drill specific plays and musicians drill specific passages rather than doing generic “reaction exercises.” Chunking is one of the few tricks with genuinely strong evidence behind it: working memory holds roughly the same number of chunks whether each chunk is one digit or one whole word, so recoding “1-9-4-5” as “the year the war ended” trades four slots for one. Reducing simultaneous load helps too — writing things down, doing one thing at a time, and cutting background noise and notifications all effectively free up scratchpad space that would otherwise be spent holding irrelevant information instead of the task at hand.
If you want to train working memory specifically, deliberately hold information without external aids: repeat a shopping list back before you write it down, do a calculation across several steps entirely in your head, or replay a short sequence of moves or directions before acting on them — the same idea our mental math guide leans on for arithmetic specifically. Span tasks — where a sequence grows by one item every time you succeed — are a genuinely good format, because the difficulty tracks your own ability automatically rather than staying fixed at a level that is either trivially easy or permanently too hard. The honest expectation is task-specific improvement: you will get better at holding sequences, or better at holding spatial layouts, or better at holding numbers, largely in proportion to what you actually practised.
A few of the games on this site are, structurally, working-memory span tasks with better production values than a lab experiment. Memory Path is a spatial span task: a path lights up across a grid and you retrace it from memory, getting longer as the grid grows on harder difficulties, which is exactly the “sequence grows until you fail” format research uses to measure span. Color Match leans on a shorter, more perceptual kind of working memory — holding a specific colour in mind for a few seconds with nothing to rehearse it against but your own sense of hue and shade, which our colour perception guide covers in more depth. Balloon Match asks for something similar with size instead of colour: a visual quantity held briefly, then reproduced with no reference to check against. None of the three will make you generally sharper, but every round is genuine, deliberate practice at the exact skill it is testing — which, per the research above, is precisely the kind of practice that actually pays off.