Is Sudoku Good for Your Brain? What the Studies Actually Found

Sudoku makes you better at Sudoku. That part is settled, and it is the only part that is.

The biggest piece of evidence under the headlines is the PROTECT study, run by the University of Exeter and King's College London and published in 2019. More than 19,000 adults aged 50 to 93 said how often they did puzzles, then sat tests of attention, reasoning and memory. The regular number-puzzlers scored better, by a margin the authors described as equivalent to several years of age-related decline. Big sample, real result.

Now read the design. Everyone in that study picked their own hobbies, so what it shows is that puzzling and sharper scores turn up in the same people. It cannot tell you which one caused the other, and the authors said so themselves. There is no good evidence that Sudoku protects anybody against dementia. A page telling you otherwise has gone past what was measured.

Worth doing. Not medicine.

Want a grid that stretches instead of soothes? Sudoku Master carries 4,000 classic puzzles across four grades, works with no signal, and shows no ads while you solve.

The four claims, sorted

Almost everything written about Sudoku and the brain is one of these four sentences. They are not equally true.

The gap between row two and row three is where most of the nonsense lives. Row two is a measurement. Row three is a hope attached to it.

What PROTECT measured, and what it could not

The size is the strength. Nineteen thousand people is a rare sample in a field where a trial with 60 undergraduates passes for normal, and the pattern held across every age in it. The question put to people was ordinary and the tests were standard.

The design is the limit, and it is not a technicality. Take two people at 62. One picks up a hard grid, finds it absorbing, and finishes it on the bus. The other loses the thread somewhere in the third box, feels stupid, and puts it down. Twenty years later the first has solved a thousand puzzles and the second has solved none. A study that asks both of them how often they puzzle, then tests their attention, sees a clear pattern. It has no way of knowing whether the puzzles built the attention or the attention made the puzzles enjoyable enough to keep doing.

There is a second wrinkle worth naming. The puzzle habit was self-reported, so the answer to "how often do you do puzzles" came out of the memory of a person whose memory was the thing being measured. That does not sink the result. It does mean the result is softer than a headline number makes it sound.

None of this makes PROTECT a bad study. It makes it an observational one, which is what its authors called it.

The transfer problem, which keeps showing up

Practise Sudoku and you get faster at Sudoku. That is near transfer, and it is reliable. You start seeing a locked pair without looking for it. You stop re-scanning boxes you already cleared.

Far transfer is the claim that the improvement leaks into things you never practised. Remembering where you left your keys. Following a legal argument. Doing your job. Evidence for that is thin across brain-training research generally, which is why the field has a name for the problem. The ACTIVE trial and the brain-training reviews that followed it are the usual references, and their pattern is the same: trained tasks improve, untrained ones mostly do not.

Two bodies of evidence sit side by side and point different ways. That is a real disagreement, not a settled question. PROTECT reports a strong association between number puzzles and cognitive scores; the transfer literature reports that practised skills stay put. Both can hold at once if sharper people gravitate to puzzles and puzzles mainly sharpen the puzzle. Any writer who quotes one of those and not the other is selling something.

One minute of Sudoku, written out

Descriptions of "mental exercise" are vague. The actual work is not. Here is an easy grid, 36 givens.

..9.6.5.8
.6..917.2
....5....
...4.6.21
.8.9.3...
.2..15..6
5....7683
..75..19.
418.39..7

Take the digit 2 into box 5, the middle box, rows 4 to 6 and columns 4 to 6. Three cells there are empty: r4c5, r5c5 and r6c4.

• Row 4 already holds a 2 at r4c8. That kills r4c5.
• Row 6 holds a 2 at r6c2. That kills r6c4.
• One cell survives. Write 2 at r5c5.

One digit placed. To get it you ran a visual search across nine cells, held two eliminations in your head at the same time, and applied a rule on purpose. Then you did it again for the next digit, with the board slightly different.

The same cell falls the other way too, which is the part that makes the game interesting. Write the candidates out: r4c5 is {7,8}, r6c4 is {7,8}, and r5c5 is {2,7}. The 7 and the 8 are locked into those two cells between them, so no third cell in the box can take a 7. r5c5 is a 2 again, reached from a different direction.

Underneath both routes is one fact about the grid. Twenty other cells share a row, a column or a box with the cell you are staring at, and every move you make is a claim about those twenty. Hold that and the game is a chain of small proofs. Lose it and it is guessing with extra steps.

Is that good for you? It is unquestionably more than a feed asks of you, and nobody has shown it carries over to anything else. Both sentences are true and they belong together.

If your reason for solving is your head, solve harder

Here the honest evidence runs out and reasoning takes over, so treat the rest of this section as argument rather than science.

If practice only sharpens the thing practised, the useful question is what you are practising. An easy grid you finish on autopilot is recall. You already know the moves, you are just executing them faster. The part that is actual work is the technique you do not have yet.

Which suggests a plan. Sit one rung above comfortable on the ladder, and go up when the rung goes quiet.

1. Cross-hatching and scanning
2. Naked singles and hidden singles
3. Naked pairs, triples and quads
4. Hidden pairs and triples
5. Pointing pairs and box-line reduction
6. X-Wing
7. Swordfish and XY-Wing
8. Colouring, chains and beyond

Difficulty in Sudoku is not clue count, and that trips up more people than anything else in the game. A 30-clue puzzle can be harder than a 24-clue one. What sets the grade is the hardest technique the grid forces you to use, and how many times it forces you. So a grid that is "hard" for you is one that needs a rung you have not learned. Learning that rung is the whole benefit. Grinding out your fortieth easy puzzle is not.

Three habits follow from that, and each one costs you something if you skip it.

Learn a technique instead of doing another puzzle. An hour spent understanding pointing pairs changes every hard grid you meet afterwards. An hour spent solving three easy grids changes nothing. The full ladder is written out here.

Never guess. A properly made puzzle has exactly one solution and can be finished by logic alone. If you feel you need to guess, either the puzzle is broken or you are one technique short of it. Guessing replaces the reasoning with a coin flip and then hides the damage: a wrong digit stays legal for a dozen more placements before the contradiction surfaces.

Watch when a grade goes quiet. The signal you want is not a streak. It is the moment your times at one grade stop improving, because that is the moment the grade has stopped teaching you anything.

I built Sudoku Master, and the part relevant to this page is the ladder it is built on: 1,000 puzzles at each of easy, medium, hard and extreme, at 36, 31, 26 to 28 and 21 to 27 givens. It records best time, average time and mistakes per grade, which is useful for exactly one thing, telling you when a grade has gone quiet. It will not teach you a technique, and it does not pretend to. Its solver returns an answer, not a lesson.

What is true and needs no study at all

The claims that survive scrutiny are the modest ones, and they are stated here as experience rather than as findings.

• It absorbs you. Twenty minutes go missing. For a lot of people that is the entire point, and it needs no research behind it.
• It is cheap practice at paying attention. A grid is a low-stakes place to hold one thread for fifteen minutes. Nothing is riding on it.
• It has an ending. A finished grid is finished. A feed is never finished, and that difference is the reason many people reach for a puzzle instead.
• People report it as calming. Plenty do. Plenty also find a stalled extreme grid maddening, so pick your grade accordingly.
• It is not arithmetic. Nothing in a classic grid is added up or weighed for size, so dreading numbers is not a reason to skip this one.

Those are enough of a case on their own. They do not need a study bolted to them, which is fortunate, because the study does not say what people claim.

Claims to walk away from

Any sentence pairing "proven" with "your brain". The strongest honest verb is "associated with". If a page reaches for something stronger, check whether it names a study at all.

"Cuts your dementia risk by 40 percent." No number like that comes out of the evidence discussed here. Numbers like it circulate anyway, usually with no study attached and sometimes attached to a study that measured something else.

"Makes your brain ten years younger." PROTECT's authors described a margin on cognitive tests as equivalent to several years of age-related decline, in people who chose their own hobbies. That is a way of expressing an effect size. It is not a claim that anyone got younger.

"Ten minutes a day is all it takes." No study sets a dose, because the observational work asked how often people puzzled rather than assigning them an amount. A daily ten minutes is a fine habit. It is not a prescription.

Anything that mentions "studies" without naming one. The naming is the test. There is one large study people are usually gesturing at, and once you can name it you can also read what its authors said about its limits.

Questions people ask

Does Sudoku improve memory?

It improves the memory you use inside a grid, which is holding several eliminations at once while you scan. In PROTECT, people who did number puzzles regularly scored better on memory tests, but they chose their own hobbies, so that is an association. No good evidence shows it carrying into remembering names, appointments or where you left your keys.

How much Sudoku should I do?

No study answers this, because none of them assigned a dose. One puzzle a day at a grade you can finish is a habit that lasts, which matters more than any number you could pick. Two hard grids a week that teach you something beat fourteen easy ones that do not.

Is Sudoku better for your brain than a crossword?

Nobody has run the head-to-head that would settle it. The two puzzles load different things, since a crossword pulls on vocabulary and general knowledge while Sudoku is pure constraint logic with no language in it at all. Pick the one you will actually keep doing, and there is a fuller comparison here.

Can Sudoku raise your IQ?

No evidence says so. That claim needs far transfer, and far transfer is the part of brain-training research that keeps failing to replicate. Expect to get better at Sudoku.

Does Sudoku count as maths practice?

Not in the way people mean it. The reasoning is elimination rather than calculation, so a grid gives you practice at ruling things out and none at all at sums. If arithmetic is the thing you want to exercise, the cage totals in Killer Sudoku are the version that supplies it.

Is doing Sudoku on a phone worse than doing it on paper?

No evidence separates them. Screens make some things easier, like undo, notes and a check button, and easier is not always better if the difficulty was the point. Paper makes you keep more in your head. Both sides are laid out in Sudoku on paper or on a phone.

Does Sudoku help with stress?

Many people find it settling, and that is a report rather than a finding. A grid gives you one problem with a definite answer, which is a rare shape in a normal day. The caveat is that a puzzle above your level does the opposite, so drop a grade if you want the calm rather than the challenge.

Is 70 too late to start?

No. PROTECT ran to age 93, and the association held across that range. Start on easy grids with a large enough print size, learn cross-hatching properly, and move up when easy stops being interesting. The on-ramp for older solvers is here.

Do hints ruin the benefit?

A hint that shows the answer skips the thinking. A hint that names the technique teaches you something. If your reason for solving is your head, use hints to learn a rung of the ladder rather than to finish a grid, and there is more on that distinction here.

Keep reading

• Sudoku or crosswords: which one should you do?, for the comparison people ask about next
• Every Sudoku technique, ordered from easiest to hardest, because the next rung is where the actual work is
• What actually makes a Sudoku easy, medium, hard or extreme, for why clue count is not difficulty
• A daily Sudoku habit that actually sticks, for matching the grade to the time you really have
• Sudoku for seniors: large print and an easier on-ramp, if you are starting later or helping somebody who is
• The full Sudoku guide, if you would rather see the whole ladder at once

Get it: Sudoku Master, free on iPhone and Android. Four grades, 4,000 classic puzzles that work offline, per-grade times so you can see when a grade has gone quiet, and no ads while you solve.