Sudoku is routinely marketed as exercise for the brain, and the claim is repeated so often that it has stopped sounding like a claim at all. It deserves better than that. The research on puzzles and cognition is genuinely interesting, genuinely mixed, and a good deal more careful than the headlines it generates. This article summarises what is reasonably well established, what is contested, and what is simply unknown.
A note on what follows: this is a summary of published research for general interest, not medical advice, and nothing here should inform a decision about diagnosis or treatment. If you have concerns about your cognitive health, that is a conversation for a clinician.
The crux of the whole question: transfer
Almost every dispute in this area reduces to one concept. When you practise a mental task, you reliably get better at that task. Nobody argues about this. The question is whether the improvement transfers to anything else, and researchers distinguish two kinds:
- Near transfer — improvement on tasks closely resembling the trained one. Practise Sudoku, get better at other number-placement puzzles. This is consistently observed and not very surprising.
- Far transfer — improvement on cognitive abilities in general: working memory, reasoning, everyday functioning, resistance to age-related decline. This is what “brain training” advertising implies, and it is where the evidence becomes thin and contested.
Keeping the two apart is most of what it takes to read this literature sensibly. A study showing that puzzle practice improves puzzle performance is not evidence for the advertised claim, however impressive the effect size.
What the training studies found
The most widely cited single result is a 2010 study by Adrian Owen and colleagues, published in Nature, which recruited over eleven thousand participants through a BBC television programme and had them train on cognitive tasks online over six weeks. The participants improved substantially on the tasks they practised. On untrained tests of general cognitive ability, the improvement was not meaningfully different from that of a control group. Near transfer, yes. Far transfer, no.
In 2016, a large review by Daniel Simons and colleagues in Psychological Science in the Public Interest examined the brain-training literature as a whole. Its conclusion was similar and blunter: the evidence that these programs improve performance on the trained tasks is strong, the evidence for improvement on closely related tasks is weaker, and the evidence that they improve everyday cognitive performance is weak to absent. The review also noted recurring methodological problems — small samples, inadequate control conditions, and expectation effects, since participants who know they are being trained tend to try harder on the outcome tests.
The ACTIVE trial deserves separate mention, being unusually large and long. It randomised roughly 2,800 older adults to training in memory, reasoning, or speed of processing, and followed them for years. Training produced durable gains — the improvements were still detectable a decade later — but they were largely specific to the trained ability, with modest and inconsistent effects on self-reported everyday functioning. That is a real result, and it is also a narrower one than the popular summary suggests.
What the observational studies found
A different line of research does not train anyone. It observes large numbers of people over time, records what they do, and looks for associations with later cognitive outcomes. These studies consistently find that people who engage in cognitively demanding leisure activities — reading, board games, crosswords, number puzzles — tend to show better cognitive function and lower rates of dementia diagnosis than people who do not.
A frequently cited example is a 2003 New England Journal of Medicine paper by Joe Verghese and colleagues, which followed older adults over several years and found participation in cognitively engaging leisure activities associated with reduced dementia risk. More recently, analyses from the PROTECT cohort in the UK, involving roughly nineteen thousand adults over fifty, reported that self-reported frequency of word and number puzzle use was associated with better performance on tests of attention, reasoning and memory — with the association appearing to strengthen the more frequently people reported puzzling.
These are large, carefully conducted studies and the associations are real. What they cannot establish is direction.
Why the observational findings are hard to interpret
Three problems recur, and none of them is a criticism of the researchers, who generally flag them explicitly:
- Reverse causation. Sudoku might preserve cognitive function. Or good cognitive function might cause Sudoku. People whose reasoning is sharp find puzzles rewarding; people experiencing early, undiagnosed decline find them frustrating and quietly stop. Since preclinical changes can precede diagnosis by many years, an observed association between puzzling and later cognition may partly reflect the disease suppressing the hobby rather than the hobby resisting the disease.
- Confounding. Puzzle-doers differ from non-puzzle-doers in many ways that independently predict cognitive outcomes: education, income, occupational complexity, physical activity, social connection, health-care access. Statistical adjustment helps but cannot fully remove what it cannot measure.
- Cognitive reserve. A leading interpretation is that mentally engaging activity across a lifetime builds reserve — the capacity to tolerate a given amount of neuropathology before symptoms appear. If that is right, the protection comes from decades of general mental engagement, not from a specific puzzle, and starting Sudoku at seventy is a different proposition from having been intellectually active since twenty.
So what can honestly be said?
Putting the two literatures together, a defensible summary looks roughly like this:
- Well supported: practising Sudoku makes you better at Sudoku, and at closely related puzzles. The improvement is real, substantial, and comes from pattern recognition becoming automatic — the mechanism described in the guide on getting faster.
- Not well supported: that Sudoku raises general intelligence, working memory, or everyday cognitive performance. Controlled trials have looked for this repeatedly and mostly not found it.
- Genuinely uncertain: whether regular puzzling meaningfully slows age-related cognitive decline. The observational association is robust; the causal interpretation is not settled, and the studies that could settle it — long, randomised, decades-long — are extremely difficult to run.
- Reasonably safe to say: as leisure activities go, one that is absorbing, mentally demanding, free, and carries no known downside is a defensible way to spend twenty minutes, whatever the transfer literature eventually concludes.
A better reason to solve puzzles
There is something slightly self-defeating about framing a hobby as medicine. The health-benefit case, whatever its eventual status, is not why anybody actually keeps solving. People solve because a well-constructed puzzle offers something increasingly rare: a problem that is completely specified, completely fair, and completely solvable by reasoning, with an unambiguous answer and no dependence on anyone else's opinion. Every stuck moment has a resolution that is genuinely available to you, and finding it is the entire experience.
That is a real and unusual pleasure, and it does not require a clinical justification. If it also turns out to build reserve, so much the better — but the honest recommendation is to solve because the reasoning is satisfying, and to treat any cognitive dividend as an unproven bonus rather than the point.
If you want the reasoning to be as rich as possible, the useful move is to keep meeting patterns you have not automated yet, since that is where the actual thinking happens. The learning-order guide lays out a sequence, and the practice pages let you work on a specific technique rather than hoping one turns up.
Sources referred to
- Owen, A. M., et al. (2010). Putting brain training to the test. Nature, 465, 775–778.
- Simons, D. J., et al. (2016). Do “brain-training” programs work? Psychological Science in the Public Interest, 17(3), 103–186.
- Ball, K., et al. (2002). Effects of cognitive training interventions with older adults: the ACTIVE randomized controlled trial. JAMA, 288(18), 2271–2281.
- Verghese, J., et al. (2003). Leisure activities and the risk of dementia in the elderly. New England Journal of Medicine, 348, 2508–2516.
- Brooker, H., et al. (2019). An online investigation of the relationship between the frequency of word puzzle use and cognitive function in a large sample of older adults. International Journal of Geriatric Psychiatry, 34(7), 921–931.
Summaries above are the author's reading of these papers and are simplified for a general audience; the originals are worth consulting for the caveats their authors attach. Corrections are welcome via the about page.