A reasonable question follows anyone who solves the same difficulty of Sudoku every morning for a year: is this still doing anything? The honest answer is more interesting than either of the confident ones. It is not the flat “puzzles keep your mind sharp” of the advertising, and it is not the flat “brain training does not work” of the debunkings. It is that the thing which might matter is not the puzzle. It is whether the puzzle is still hard for you.
The companion article, does Sudoku help your brain, covers what the training and observational literatures actually establish, which is less than people hope. This one narrows to a single mechanism inside that question, because the mechanism is well evidenced even where the health claim is not: practice makes a task cheaper to perform, and a task that has become cheap is no longer the task you started doing. The same caveat applies as there — this is a summary of published research for general interest, not medical advice.
What repetition does to a task
The foundational work here is Schneider and Shiffrin's 1977 account of automatic and controlled processing. A controlled process is deliberate, attention-hungry, and slows down as you load it with more to track. An automatic process fires in response to its input without needing you to steer it, and it is remarkably indifferent to load.
The condition that converts one into the other is consistent mapping: the same input reliably calling for the same response, repeated. That is a precise description of what an easy Sudoku offers. A cell with eight of its peers filled always means the same thing and always wants the same response. Do that ten thousand times and you are no longer reasoning about it in any effortful sense — you are reading it, the way you read a word rather than sounding it out.
This is not a failure. It is the entire mechanism behind getting fast, which is why the guide on getting faster treats automaticity as the goal rather than a hazard. The point is only that automaticity is a one-way door. Once a pattern is automatic, running it again cannot make it more automatic, and it stops drawing on the effortful machinery it drew on while you were learning it.
The brain measurably spends less
There is a striking demonstration of this from an unlikely source. In 1992 Richard Haier and colleagues used PET imaging to measure regional cerebral glucose metabolic rate — roughly, how much energy a brain region is consuming — in participants playing Tetris, before and after several weeks of daily practice.
Performance improved more than sevenfold. Glucose consumption across cortical surface regions went down. And the relationship ran in the direction that makes the point sharpest: the participants who improved the most showed the largest metabolic decreases. Getting good at the task meant the brain doing conspicuously less to accomplish it, apparently by ceasing to recruit regions that were never well suited to it in the first place.
That result seeded what became the neural efficiency hypothesis, and it should be read carefully rather than triumphantly — eight participants, one task, a measure of energy use rather than of benefit. Lower metabolic cost is not straightforwardly bad, and nobody has shown that a metabolically cheap puzzle is a worthless one. But it does put a physical number on the intuition. The puzzle you have mastered is not asking your brain for what it asked a year ago, and you cannot recover the difference by doing more of them.
The study that separated demand from activity
If effortful novelty is what matters, the cleanest test would compare people doing something demanding and new against people doing something pleasant and undemanding for the same hours. That study exists. Denise Park and colleagues ran it as the Synapse Project, published in Psychological Science in 2014.
Older adults were randomised to spend roughly fifteen hours a week for three months either learning a genuinely new and demanding skill — quilting, digital photography, or both — or in a comparison condition: socialising around non-intellectual activities, or doing low-demand cognitive tasks alone. Episodic memory improved in the productive-engagement conditions relative to the receptive ones. The sustained, effortful learning of something unfamiliar was where the gain appeared. Pleasant social activity, by itself, largely was not.
Three honest qualifications, because this study gets over-quoted. Fifteen hours a week is an enormous dose — nothing like twenty minutes over coffee, and any extrapolation down to that has to be treated as speculation rather than as a finding. The outcome that moved was episodic memory, a specific measure, not general intelligence or everyday functioning. And nobody randomised anyone to Sudoku; a puzzle is not a quilt. What the study supports is narrower and still useful: that when hours and social contact are held constant, cognitive demand is the variable that carried the effect. Which is exactly the variable that automaticity quietly removes from a puzzle you have mastered.
Not all difficulty counts
The obvious inference — make it harder, then — is where this reasoning most often goes wrong, and the learning literature has an unusually clear correction for it. Robert Bjork's notion of desirable difficulties describes conditions that feel worse during practice while producing more durable learning: spacing sessions out, retrieving rather than reviewing, interleaving different problem types instead of blocking them.
Bjork has been insistent that difficulty is not desirable in itself. A difficulty earns the adjective only when the learner can actually overcome it with effort, and when it engages the processes the goal requires. Difficulty you cannot get through teaches nothing; it just produces failure and quitting.
For a solver, that is a genuine constraint. Jumping from comfortable singles to a grid requiring chains and colouring is not a desirable difficulty — it is a wall, and it is why the learning-order guide insists on a ladder. The productive zone is the puzzle that stops you, where the pattern that unsticks it is one you could reconstruct if you thought hard. Which is a demanding thing to arrange deliberately, and almost impossible to hit by picking a difficulty label at random.
What this does and does not license
Putting it together, without inflating it:
- Well supported: practice under consistent conditions makes a task automatic, and an automatic task makes fewer demands on controlled attention. This is about as settled as findings in cognitive psychology get.
- Well supported: mastering a task can reduce the metabolic cost of performing it, sometimes markedly.
- Reasonably supported: where hours are held constant, novel and cognitively demanding engagement outperforms undemanding engagement on at least some memory measures.
- Still not supported: that any of this makes Sudoku a treatment. Far transfer to general cognitive ability remains the thing controlled trials keep looking for and mostly not finding, and nothing here changes that verdict. If harder puzzles are better for you than easy ones, it is a difference between two things of unproven benefit.
So the defensible version of “easy Sudoku stops working your brain” is a claim about demand, not about health outcomes. A puzzle you have automated is demonstrably asking less of you than it used to. Whether the asking was ever buying you anything beyond the pleasure of solving is a separate question, and an unsettled one.
Staying at the edge on purpose
If the mechanism is right, the practical implication is not “solve harder puzzles” but “keep meeting patterns you have not automated yet” — which is a different instruction, and a more tractable one. Difficulty labels are a poor proxy for it, because a hard puzzle is often just a longer sequence of techniques you already own, as the guide on how difficulty ratings work explains.
The more direct approach is to select by technique rather than by rating. The practice pages give you puzzles chosen to require one specific technique, so you can work on the pattern you have not yet made automatic instead of waiting for it to appear. The technique combinations raise the demand by asking for two patterns in one grid, which is closer to interleaving than to simply grinding upward. And when a pattern genuinely will not come, the derivations in why subsets work and the fish patterns are there to make the difficulty an overcomable one rather than a wall.
There is a last point worth making, since the rest of this article has been about mechanisms. The reason to keep learning new techniques is not primarily that it might be prophylactic. It is that the moment of seeing a structure you could not see last month is the best thing the puzzle has to offer, and it is available only while there are still structures you cannot see. Automaticity spends that down. Learning renews it. That the arrangement might also be the cognitively demanding one is a convenient bonus rather than the argument.
Sources referred to
- Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human information processing: II. Psychological Review, 84(2), 127–190.
- Haier, R. J., et al. (1992). Regional glucose metabolic changes after learning a complex visuospatial/motor task. Brain Research, 570(1–2), 134–143.
- Park, D. C., et al. (2014). The impact of sustained engagement on cognitive function in older adults: the Synapse Project. Psychological Science, 25(1), 103–112.
- Bjork, R. A., & Bjork, E. L. (2011). Making things hard on yourself, but in a good way: creating desirable difficulties to enhance learning. In Psychology and the Real World.
- Simons, D. J., et al. (2016). Do “brain-training” programs work? Psychological Science in the Public Interest, 17(3), 103–186.
Summaries above are the author's reading of these papers and are simplified for a general audience; the originals carry caveats worth reading. Corrections are welcome via the about page.