The dementia question, and what the evidence actually shows

Will a daily Sudoku help prevent dementia? The hope underneath the question is reasonable. A puzzle is cheap, pleasant, and easy to fit into a morning. If a daily one bought even a bit of insurance against cognitive decline, every GP would have a routine of suggesting it. The reason no GP is suggesting it on a chart is that the evidence doesn't quite line up to where the puzzle apps would like it to. The route from what the trials actually show to prevents dementia is longer than the marketing version of this story makes it sound.

This article is the longer route, walked carefully. We've already written a broader piece on what the research says about puzzles and the brain that covers the wider transfer-of-training literature. This one is the dementia-specific zoom-in, with the trials that have tried hardest to answer the question.

Two kinds of evidence, telling different stories

The conversation about puzzles and dementia almost always slides between two kinds of research without flagging which one is being cited. The first is observational: scientists track large groups of people for years and ask what they do with their time, then look at who later develops cognitive decline. The second is intervention research: scientists assign people to do something, or not, and watch what happens.

The observational story is encouraging. People who report mentally engaging activities in midlife and later — reading, puzzles, social games, learning new things — tend to hold cognitive function longer than people who report fewer of these things. This pattern is well established and goes by the name cognitive reserve. It is also the finding most often cited when an app wants you to think a puzzle is doing the protecting.

The intervention story is the one the marketing leaves out. When you actually assign people to a cognitive activity and follow them, the effects you measure are smaller, narrower, and less impressive than the population-level association would predict — which is the central tension in this whole field, and the reason the answer to the dementia question keeps getting hedged.

What the observational data does and doesn't show

That midlife mental engagement correlates with later cognitive resilience is not in dispute. What's in dispute is why. People who do puzzles also tend to read more, sleep more regularly, exercise, eat differently, and maintain stronger social ties. Any of those is a stronger candidate explanation for the protective signal than a pencil and a grid. The cholesterol of the puzzle-doer probably matters more than the puzzle itself.

This is the difference between a finding being real and a specific causal claim being real. Cognitive reserve is real. Sudoku, specifically, builds cognitive reserve is a different claim, and it isn't supported by the data we have.

What the intervention trials actually found

The way medicine moves from association to causation is to randomise people into the thing or not, and watch. This has been tried in cognitive ageing more than once.

The biggest single trial of cognitive training in healthy older adults is the ACTIVE study, which assigned about 2,800 people over 65 to one of three single-domain training conditions — memory, reasoning, or processing speed — or to a no-training control. Ten years later, the trained groups still showed an advantage on the specific abilities they had practised, plus a modest spillover into self-reported everyday functioning⁴. What the study did not show, and what its authors did not claim, was protection against dementia diagnosis.

The FINGER trial, run in Finland on adults at elevated risk of cognitive decline, took a different shape. Instead of training one cognitive domain, it bundled four ingredients — diet, exercise, vascular-risk monitoring, and cognitive training — into a two-year multidomain intervention³. The bundled group did better than control on a composite cognitive measure. This is the closest thing in the field to an encouraging signal, and it is genuinely encouraging. But the cognitive-training piece was one of four ingredients, and the study was not designed to disentangle which ingredient did how much of the work. The honest read of FINGER is that complete lifestyle intervention helped; the reading of FINGER as a vindication of puzzle-a-day is a step the trial itself does not authorise.

Sitting alongside these, the BBC-backed Brain Training Britain trial — about eleven thousand participants over six weeks — found cognitive training improved performance on the specific games people trained on, with no measurable transfer to general cognitive ability¹. A 2014 meta-analysis pooling 52 studies found the same broad pattern: small effects, mostly within the trained tasks, weaker and less consistent the further you looked from the training itself².

Why this question stays unsettled

The honest reason there isn't a clean answer is that a clean answer is hard to design for. Dementia develops over decades; trial budgets run for two to ten years. People who volunteer for a cognitive trial are different from people who don't. Self-reported activity is unreliable. And the trial that would settle the question — randomising thousands of healthy adults to do or not do Sudoku for fifteen years and tracking diagnoses — is not a study any funder is currently running.

So the field sits in an honest fog. There's a strong observational signal on mental engagement broadly, a weak causal signal on any specific cognitive activity, and a multidomain trial (FINGER) that gestures at bundles of habits help without breaking out the contribution of the puzzle.

What to do with this if you're reading on someone else's behalf

The practical move is to suggest the puzzle as something the person might enjoy, the way you'd suggest a book or a walk. Recommending it as a brain-training treatment is the framing most likely to fail in the wild — some days the solver enjoys it, some days they don't, and a treatment frame turns ordinary off days into worry. A puzzle that the person likes is one of several small mentally engaging things in their week, and the wider weekly shape — reading, walks, friends, novelty — is what the cognitive reserve story is actually about.

If the changes you've been noticing are recent, real, and consistent, the conversation that matters is with their GP. We've written a practical piece for the family member buying for an older relative that goes further on the day-to-day side of this.

What to do with this if you're reading for yourself

The picture is the same shape, scaled down. A daily fifteen-minute Sudoku is a perfectly fine addition to a week. Whether it is buying you any insurance specifically against cognitive decline is a question the trials we have don't let us answer cleanly. The companion piece on what changes in puzzling after 60 is the more direct take if that's where you are; the narrow-transfer piece is the more direct take on why getting better at the puzzle doesn't generalise outward to anything you'd notice elsewhere.

The realistic honest framing is that puzzles sit inside a wider portfolio of mental and physical engagement that, together, correlates with better cognitive ageing. None of the ingredients in that portfolio is individually load-bearing; the puzzle is one ingredient among several. That's the version of the story the evidence supports.

The reason this article exists

There's a version of this site that would put prevents dementia on the home page and get a lot more clicks for it. That version is incompatible with the trials in our reference list, and we'd rather be a site that is still recommended in three years than one that read well in 2026. The puzzle is interesting in its own right — pattern recognition, the small arc of a finished grid, fifteen minutes of focused attention in a day that often doesn't have much of that. Those are real, and they don't need a clinical claim attached. The cognitive-reserve story may, eventually, give us cleaner answers. Until then, the honest answer to the dementia question is the one this article has tried to give.