Word Games and Brain Health: What Research Shows

Word games have real but modest evidence behind them, and the strongest finding is more interesting than the usual claims. In a randomised trial, crossword puzzles outperformed purpose-built computerised brain training games for memory in older adults with mild cognitive impairment. Not matched. Beat. And separately, a survey of more than three thousand older adults found word game participation was one of the leisure activities most clearly linked to better cognitive scores.

But there's a catch in each of those studies that most coverage skips, and it changes what you should take from them. Here's what the evidence supports, and what it doesn't.

What does the research actually show?

Three pieces of evidence carry most of the weight here, and they are useful precisely because they have different strengths and weaknesses.

The first is a large survey. Tyran Terada and Yeonjung Lee, publishing in Innovation in Aging in 2021, analysed 3,271 adults aged 65 and over from the Health and Retirement Study. Participation in word games showed a significantly positive association with cognitive functioning, with a beta coefficient of .117 at p below .01. Their model explained 31 percent of the variance in cognitive scores after controlling for demographic and socioeconomic factors.

That is a big sample and a clear signal. It is also cross-sectional, meaning everything was measured at one point in time. The authors are explicit that this cannot establish causation. Word games might sharpen cognition, or people with sharper cognition might simply enjoy word games more, or something else entirely might drive both. The study cannot tell you which.

The biggest dataset says the same thing, with the same caveat

If 3,271 people sounds like a lot, the PROTECT study is an order of magnitude bigger. Researchers led by the University of Exeter and King's College London looked at more than 19,000 adults aged 50 and over, which the Alzheimer's Society described as the largest online study of its kind. The finding was clean and in the direction you would expect: the more regularly people played puzzles like crosswords and Sudoku, the better their performance on tests of brain function.

Nineteen thousand people is a serious sample. But read what the Alzheimer's Society's own Dr James Pickett said about it, because it is more careful than most of the coverage was. The research can only show that puzzling and thinking skills are linked, he noted, not that puzzling will improve thinking skills. And the study did not look for a link with dementia risk at all.

Now the two numbers from that study you have almost certainly seen quoted, because they are the reason it went everywhere. Exeter reported that people who regularly did word puzzles performed on grammatical reasoning tests at a level equivalent to someone ten years younger than their actual age, and on short-term memory tests equivalent to someone eight years younger. The findings came out as two linked papers in the International Journal of Geriatric Psychiatry in May 2019.

Two things to hold onto there. The first is that those are two different numbers for two different tests, and the headline version usually keeps the bigger one and drops the other. Ten years applies to grammatical reasoning. Eight applies to short-term memory. Nothing in the study says your brain is uniformly a decade younger.

The second matters more, and it's the caveat you just read from Pickett applied to these specific figures. "Equivalent to ten years younger" is not a before-and-after measurement of anyone. Nobody got younger. It's a way of expressing how big the gap was between regular puzzlers and non-puzzlers on a single sitting of a test, translated into age units because that reads better than a raw score. If sharper people are the ones who take up crosswords, that gap appears without a single puzzle having improved anything.

And check the year if you see this quoted. The study is from May 2019, but it gets attributed to 2023 fairly often, sometimes bolted onto a specific claim about playing four or more times a week. The underlying numbers are real. The date and the framing wrapped around them frequently are not.

So now you have two large observational studies, one with 3,271 people and one with more than 19,000, both pointing the same way. Here is the thing worth understanding: that does not add up to proof. Stacking bigger and bigger cross-sectional samples makes the association more certain, not more causal. If sharper people gravitate toward puzzles, you would see exactly this pattern at any sample size. Twenty thousand people cannot fix a design that measures everyone once.

Which is why the next study matters more than either of them, despite being far smaller.

The third piece is a proper randomised trial, and it is the one worth knowing about.

Why did crosswords beat computer brain games?

Because someone finally ran the comparison, and the result went the opposite way to what the brain training industry would predict.

D.P. Devanand at Columbia and Murali Doraiswamy at Duke ran a 78-week, two-site, single-blinded trial published in NEJM Evidence and funded by the National Institute on Aging. They randomised 107 people with mild cognitive impairment, mean age 71.2, into two groups. One got 12 weeks of intensive home-based training on web-based cognitive games. The other got web-based crossword puzzles. Both then had six booster sessions.

Crosswords won on the primary outcome. On the ADAS-Cog scale, where a higher score means worse impairment, the crossword group improved from 9.59 to 8.61 while the games group got worse, drifting from 9.53 to 9.93. The adjusted difference at week 78 was 1.44 points in favour of crosswords, with a confidence interval of 2.83 to 0.06 and a p value of 0.04.

Two further results are worth knowing. Crosswords also came out ahead on the Functional Activities Questionnaire, which tracks everyday functioning rather than test scores, by 1.08 points. And on brain imaging, decreases in hippocampal volume and cortical thickness were greater in the games group than the crossword group, which the authors suggest points to a possible neuroprotective effect from the crossword training.

So the humble crossword outperformed software built specifically to train cognition, on memory, on daily function, and on brain structure. That is a genuinely surprising result and it deserves more attention than it gets.

What does that trial not prove?

This is the part that gets left out, and it matters more than the headline.

The authors list their own limitations, and the important one is that there was no control condition without cognitive training. Everyone in the study did something. So the trial shows crosswords beat computerised brain games. It does not show that crosswords beat reading a book, doing the gardening, or sitting quietly. Nobody in that study was doing nothing, so we cannot know how either group compares to a person who simply carried on with their life.

They also flag that participants had an unusually high average level of education, which limits how far the finding generalises. And it is 107 people in one clinical population, adults who already had mild cognitive impairment. Whether the same holds for a healthy forty-year-old is an open question the study was never designed to answer.

Put all three studies side by side and you get the honest picture. The two big surveys, more than 22,000 people between them, show association but cannot prove direction. The trial shows causation between two options but had no do-nothing comparison. None of this is weak research. It is just answering narrower questions than the headlines suggest.

The same caution applies to commercial brain training generally. The National Institute on Aging has said there is not enough evidence to suggest commercially available computer-based brain training applications have the same impact as structured trial-based cognitive training. We go through that evidence properly in our guide on whether brain training actually works.

And that caution has already been tested somewhere firmer than a journal. In January 2016 the US Federal Trade Commission settled deceptive advertising charges against Lumos Labs, the company behind Lumosity, for $2 million in redress. The complaint was that the company had claimed its games would improve performance at work and school, delay age-related cognitive decline, and protect against mild cognitive impairment, dementia and Alzheimer's, and that scientific studies proved all of it. The regulator's own summary is worth quoting: Lumosity "preyed on consumers' fears about age-related cognitive decline" and "simply did not have the science to back up its ads." The settlement also covered testimonials the company had solicited through prize contests without disclosing it.

Hold that next to any brain-game marketing you read, this site's category included. The claim that a game protects your memory is precisely the claim a regulator has already fined a company for making.

Do word games build sharper brains, or do sharper brains pick word games?

Nobody knows, and that gap is the single biggest reason to be careful with the survey findings. Every study showing that regular players have better cognition is compatible with two opposite stories: the games improved them, or people who already had unusual mental machinery were the ones who stuck with the games and got good.

The clearest look at that problem comes from studying people at the very top. Toma, Halpern and Berger, publishing in Applied Cognitive Psychology in 2014, tested 26 nationally ranked Scrabble experts and 31 crossword experts. The control group is what makes this study useful. Instead of comparing champions against average adults, they used college students matched on SAT scores, with quantitative and verbal averages both above 700. These were not slow controls.

The experts still significantly outperformed them on every cognitive measure, including a symmetry span task for visuospatial working memory and a grid-recall task tracking where blue squares had flashed. So competitive word game players really do have exceptional working memory, well beyond even high-achieving students.

But look at what the study found between the two expert groups. Crossword experts beat Scrabble experts on analogies, which makes sense given how cryptic clues work. On the working memory measures, though, the two groups did not significantly differ. And that is strange if the games are doing the building, because Scrabble and crosswords demand different things. Scrabble is tile shuffling and board geometry. Crosswords are retrieval from a clue. If years of practice sculpted the specific abilities each game trains, the two groups should have drifted apart. They had not.

The authors read that as support for domain-general working memory, a single system handling both verbal and visuospatial material rather than two separate ones. There is a second reading that matters more for your purposes: it looks a lot like both groups arrived with high general working memory capacity already, then chose games that reward it.

This study cannot settle it, because it measured everyone once. Cross-sectional designs photograph a difference. They cannot tell you which way the arrow points. To answer it properly you would need to measure working memory in people who have never played competitively, wait years, and see who rose to the top and whether their scores moved. That study does not exist.

Keep this in mind whenever you see a headline about puzzle players having younger-seeming brains. The finding is usually real. The causal claim stacked on top of it usually is not, and selection effects are the most likely alternative explanation. If you want the mechanics of the memory system all this hinges on, our guide to what working memory is covers how it gets measured and why capacity varies so much between people.

Could the benefit just be expecting a benefit?

Partly, and it is the third problem stacked on the two above. Selection explains who signs up. Causation explains what a correlation cannot tell you. Expectation is the sneakier one: people who believe something will help them often score better on the test afterwards, whether or not the thing did anything at all.

Foroughi and colleagues built a study to catch exactly that, published in PNAS in 2016 and available via PubMed Central. They recruited 50 people using two flyers. One openly advertised brain training and told readers that numerous studies have shown working memory training can increase fluid intelligence. The other looked much the same and mentioned nothing about brain training. Both groups then did the same hour of the same task on the same day.

Only the group recruited by the suggestive flyer improved. Their gain on Raven's matrices came out at an effect size of 0.50, which the authors put at the equivalent of a 5 to 10 point rise on a standard IQ test, after a single hour. For the control group, the before and after scores were statistically indistinguishable.

Read that carefully, because it is not a claim that training does nothing. It is a claim that how you recruit people can manufacture an effect roughly the size of the one researchers are hunting for. Advertise for volunteers who want to improve their memory and you have pre-selected for belief, and belief on its own moves the numbers.

Two honest limits before you carry this too far. The sample was 50 people, which is small. And their mean age was 21.5, so this is young adults on a dual n-back task rather than retirees on crosswords, and nobody has run the same trick on the crossword literature. But the mechanism does not care which puzzle is involved, and a good many of the observational studies on this page recruited people who already liked puzzles and knew exactly what was being measured.

Which is the practical reason to keep the bar where the rest of this article puts it. When a study finds that people who chose a brain activity did better on a test they knew they were being scored on, expectation is sitting in the room next to selection. Neither of them is the puzzle.

Does getting better at word games make you better at anything else?

Mostly no, and this is the finding that should shape your expectations more than any other.

The crossword trial had no do-nothing group, as covered above. But there is a much larger trial that did, and it is the one that answers the transfer question. ACTIVE, the Advanced Cognitive Training for Independent and Vital Elderly study, enrolled 2,832 adults aged 65 and over and randomised them into three training arms, one each for memory, reasoning and processing speed, plus a no-contact control arm. It is the largest trial of its kind and it ran for two decades.

The Agency for Healthcare Research and Quality, in a comparative effectiveness review published in March 2017, assessed the evidence and put it plainly. Cognitive training improved performance in the domain that was trained. But transfer to other domains, in the review's own words, was rare.

The review also graded how much confidence to place in the results over time: moderate strength of evidence at two years, dropping to low strength at five and ten years. Processing speed training was associated with better everyday functioning, but that finding was also rated low strength because so many participants dropped out along the way that the numbers get hard to trust.

ACTIVE is about cognitive training generally, though. For word games specifically there is a sharper test, and it used the best possible subjects: people who have spent years doing nothing but this.

Van Hees and colleagues, publishing in Frontiers in Human Neuroscience in November 2016, compared 19 competitive Scrabble experts against 19 age-matched controls. They ran two tasks. A lexical decision task, judging words and non-words, tested near transfer. A symbol decision task, using unfamiliar symbols, tested far transfer.

On the word task the experts were exactly what you would expect: notably better, including at reading vertically, which is a Scrabble-specific skill nobody else practises. On the symbol task they were no faster and no more accurate than the controls. Years of competitive play bought them nothing at all on a non-linguistic problem.

But the brain scans complicate it in a way worth knowing. The experts solved the tasks using different machinery, leaning on posterior visual and temporal-parietal regions where controls used temporal-frontal ones. The authors' summary is the memorable line here: the neuroimaging gave "evidence of brain transfer in the absence of behavioral transfer." Their brains had genuinely reorganised. Their performance outside the game had not.

That distinction is worth holding onto, because it explains how honest researchers and breathless headlines can both cite the same study. "Word games change your brain" is true. "Word games make you sharper at other things" did not follow from it.

Applied to word games, that means something specific. Do crosswords regularly and you will get better at crosswords. Your retrieval for the kind of clue crosswords use will sharpen, and that is real. What you should not expect is for it to make you better at remembering where you parked, following a complicated recipe, or keeping track of a conversation in a noisy room. Those are different abilities, and the evidence that training one moves the others is weak.

This is not a reason to stop playing. It is a reason to play because you enjoy it, and to treat any wider benefit as a bonus you cannot count on.

Do word games work as therapy after a stroke?

Here's the one place where structured word practice has genuinely strong trial evidence behind it. And the interesting part is that even here, with clinical design and real supervision, the pattern looks exactly like everything else on this page.

Aphasia is language loss after brain injury, most often a stroke. Marian Brady and colleagues reviewed the treatment evidence for Cochrane in 2016, pooling 57 randomised trials covering 3,002 people. Compared with no therapy, speech and language therapy produced clinically and statistically significant benefits in functional communication, reading, writing and expressive language across 27 studies and 1,620 participants.

So this is real, and it's a different order of evidence from the crossword studies further up.

Now the study that maps most directly onto word games. Palmer and colleagues ran the Big CACTUS trial, published in Lancet Neurology in 2019. It was a three-arm randomised controlled trial across 21 community therapy departments in 20 UK NHS trusts, with 278 participants. People used software to practise retrieving 100 words they had chosen themselves as personally useful, aiming for 20 to 30 minutes a day over six months. On average they clocked 28 hours of practice.

The word-finding result was emphatic. The therapy group improved 16.4 percent on their practised words against 1.1 percent for usual care, a 16.2 point gap at p less than 0.0001, and it was still there at 9 and 12 months.

Then read the rest of it. Functional communication in actual conversation did not move at all (a change of minus 0.03 points, p = 0.709). Self-rated communication and quality of life showed no meaningful change. And crucially, the gains did not generalise to words the participants had not practised.

Sit with that last one, because it's the whole argument of this article delivered by a large clinical trial. Twenty-eight hours of focused, supported, individually tailored word practice made people substantially better at those specific words and did not spread to anything else. If that's the ceiling under ideal conditions, a Wordle habit is not going to clear it.

The Cochrane review has one more finding worth knowing. Across nine studies and 447 participants, there was no evidence of a difference in functional communication between formal therapy and plain social support. More people dropped out of the social support arm, but the communication outcome held level. Higher intensity therapy did produce better functional communication, though it also came with a significantly higher dropout rate, which is its own kind of finding about what people will actually stick to.

Two things to take from this, and they pull in different directions:

• Practice works on what you practise. That's not a consolation prize. If there are specific words you want back, or want to hold onto, targeted repetition is the thing with evidence behind it.
• It stays where you put it. Nothing here suggests word drills rebuild general communication, and the trial that looked hardest for that found nothing.

One important line to draw. Aphasia is a medical condition and this is a clinical treatment, delivered with a speech and language therapist involved. If you or someone close to you has language difficulty after a stroke, that's a referral to a speech and language therapist, not a puzzle app. The trial evidence above describes supervised therapy programmes, not a substitute for them.

Which brain skills do word games actually train?

Different word games work different things, and the differences are bigger than people assume.

• Crosswords lean on retrieval and general knowledge. You are searching memory for a word that fits a definition and a letter pattern at once, which is a genuine dual constraint.
• Anagram and scramble games train working memory and mental manipulation. You hold letters in mind and rearrange them without writing anything down, which is closer to a pure working memory task than a vocabulary one.
• Word searches mostly train visual scanning and sustained attention. There is very little vocabulary work happening, because the word is already given to you.
• Daily guessing games combine deduction with vocabulary retrieval, and the fixed daily format does something the others do not: it builds a habit.

If you want to understand the memory system doing most of the lifting in the middle two, our guide to working memory covers what it is and why it fills up so fast.

Do word games help because of vocabulary or something else?

Probably not mainly vocabulary, and this is the most useful thing to understand about the whole category.

Psychologists separate crystallised ability, meaning the knowledge you have accumulated, from fluid ability, meaning your capacity to reason and process new information. Vocabulary is the classic example of crystallised ability. It is also the ability that holds up best with age, often improving into your sixties and seventies while processing speed and working memory decline.

Which creates an awkward question. If word games mostly exercise the one ability that was already ageing well, why would they help?

The likely answer is that the vocabulary is the wrapper rather than the mechanism. What a crossword actually demands is sustained attention across a long session, working memory to hold partial answers, flexible switching between clues when you get stuck, and retrieval under constraint. Those are fluid abilities wearing a vocabulary costume. That would also explain why crosswords beat the computerised games in the trial: a crossword keeps you engaged for thirty minutes of genuine effort, while a brain training app often gives you sixty seconds of a narrow drill and a congratulatory noise.

Can word games help with tip-of-the-tongue moments?

Maybe, but only the kind where you have to come up with words, not just recognise them. And nobody has run the trial that would settle it. What the research does show is why these moments happen and one surprisingly simple thing that makes them less likely.

First, the reassurance. Psychologist A. S. Brown reviewed decades of research on this in Psychological Bulletin in 1991 and found tip-of-the-tongue moments are nearly universal. They happen about once a week, they increase with age, they're most often triggered by proper names, and you often get the first letter even when the rest won't come. They also resolve on the spot about half the time. So if you blank on an actor's name at dinner, you're in very ordinary company.

Why do they get more common with age?

The problem isn't that the word has gone. You know it. You can often describe it, give its first letter and name three similar words. What fails is pulling up the full sound of the word so you can say it.

Shafto, Burke, Tyler and colleagues tested this in people aged 19 to 88, publishing in the Journal of Cognitive Neuroscience in 2007. Participants named celebrities from photos and descriptions while the researchers counted tip-of-the-tongue moments and scanned their brains. The number of those moments rose with age, and it tracked grey matter loss in the left insula, a region involved in producing speech sounds. That link held even after age itself was taken out of the picture. And here's the telling part. Scores on a reasoning test also dropped with age but had nothing to do with that same region. So it's a specific sound-retrieval problem, not a general sign of your mind slipping.

What actually helps when a word won't come?

Saying words that sound like it. James and Burke showed this in the Journal of Experimental Psychology: Learning, Memory, and Cognition in 2000. Young and older adults read aloud a list of words before answering general knowledge questions. When some of those words shared sounds with the answer, people had fewer tip-of-the-tongue moments and got more answers right. When people were already stuck, reading related-sounding words helped them retrieve the answer. The effect was the same size for older and younger adults, even though the older group got stuck more often.

The authors' explanation is that the weak connections behind these blanks get stronger each time you actually produce related sounds. Which points to a useful distinction between word games.

• Recognition games give you the word and ask you to spot it. Word search is the classic example. It's relaxing, but you're rarely producing anything.
• Production games make you generate words from sounds or letters. Anagrams, word chains, Scrabble-style tile games and our word scramble all ask you to pull up a word from partial information. That's closer to the exact step that fails during a tip-of-the-tongue moment.
• Saying it out loud may matter too. The priming study had people read words aloud. If you play alone, muttering your guesses isn't silly. It's closer to the mechanism.

Keep that in proportion. The James and Burke study measured an effect within a single session, not a lasting improvement from months of play. No trial has shown that regular word games reduce tip-of-the-tongue moments over time. It's a sensible bet based on how the problem works, not a proven treatment. If word-finding problems are getting noticeably worse, or you're losing common words rather than names, that's worth mentioning to a doctor, because that's a different pattern from the ordinary kind. For the memory side of all this, our guide to working memory explains how it differs from the long-term store your words live in.

Does being bilingual change the picture?

Possibly, and the evidence splits along exactly the same fault line as everything else on this page. Worth covering properly, because a large share of people reading this speak two or three languages daily without ever thinking of it as brain training.

The famous finding came from Ellen Bialystok's group. Reviewing the records of 184 patients at a memory clinic, they found bilingual patients reported the onset of dementia symptoms around four years later than monolingual patients. A follow-up published in Neurology put it at 4.3 years later to diagnosis and about five years later to first symptoms. Those are big numbers. No drug currently does that.

Then the same problem you have already met twice on this page. Those are retrospective studies, looking backwards at people who already had a diagnosis. When researchers ran the other kind, following healthy people forward and waiting to see who developed dementia, the effect went away. A systematic review with meta-analyses pooling prospective studies covering 5,527 bilingual participants found an odds ratio of 0.96 for developing dementia, with a confidence interval running from 0.74 to 1.23. That interval sits squarely across 1.0, which is the statistical way of saying no detectable effect.

The review's own title carries the resolution, and it is a genuinely useful distinction: bilingualism is associated with a delayed onset of dementia but not with a lower risk of developing it. A separate systematic review reached the same unsatisfying place, reporting mixed results with some studies finding a protective effect and others failing to, and pointing at differences in study design to explain the split.

So what should you take from it? Two things, and they pull in opposite directions.

The delay is real enough to matter even if the risk isn't lowered. Cognitive reserve is the usual explanation: a brain with more routes to the same answer keeps functioning normally for longer while the underlying pathology builds. Living well for four extra years is worth having, whatever the eventual diagnosis rate. But the flip side is uncomfortable and rarely mentioned. If symptoms surface later while the disease has been progressing all along, decline after diagnosis tends to be faster, because more damage had already accumulated by the time anything showed.

And for the practical question people actually want answered: none of this says that starting a language app at 55 buys you four years. The research looked at lifelong bilinguals, people who had been switching between languages every day for decades. That is a different exposure from a hobby. If you already live that way, it may be doing more for your brain than any puzzle on this site. If you don't, learning a language is a genuinely good and demanding activity, which is the standard the next section sets anyway.

What actually lowers your dementia risk?

Not word games. That's the uncomfortable part of an article like this, and you deserve it straight rather than buried.

The reference point here is the 2024 report of the Lancet standing Commission on dementia prevention, intervention and care, led by Gill Livingston and published in The Lancet in August 2024. It's the closest thing this field has to a consensus document. The Commission identified 14 modifiable risk factors and estimated that addressing all of them could prevent 45 per cent of dementia cases globally, a figure restated by Burley and colleagues in Alzheimer's and Dementia in 2025.

Here's the list: low education, hearing loss, high LDL cholesterol, depression, traumatic brain injury, physical inactivity, diabetes, smoking, hypertension, obesity, excessive alcohol, social isolation, air pollution, and untreated vision loss. Read it twice and notice what isn't there. No crosswords. No Sudoku. No puzzle habit of any kind.

The closest entry is low education, and it's worth being precise about what that means, because people reach for it as proof that puzzles count. It refers to years of formal schooling early in life, which is thought to build cognitive reserve, meaning a brain that copes with damage for longer before symptoms appear. That's decades of sustained learning, not twenty minutes with a word grid. If word games contribute at all, they're a rounding error next to it.

But what about the study that says crosswords buy you 2.5 years?

This is the one people cite back at you, and it's a real finding from a decent study. It also gets quoted with the second half chopped off, which changes what it means quite a lot.

Pillai, Hall and colleagues worked from the Bronx Aging Study, which followed 488 cognitively normal community residents with assessments every 12 to 18 months. They took the 101 participants who went on to develop dementia and modelled the point where each person's memory decline sharply accelerated, measured on the Buschke Selective Reminding Test. Writing in the Journal of the International Neuropsychological Society in 2011, they reported that crossword participation delayed the onset of that accelerated decline by 2.54 years.

That is the number you see in headlines. Here is the sentence that tends to get cut. Once the decline did start, it ran 3.31 test points per year faster in the crossword players than in the non-players.

The authors read the whole pattern as support for cognitive reserve. Stimulating activity helps the brain keep compensating for damage that is already building, so the symptoms surface later. But nothing has slowed the underlying pathology. When the reserve finally runs out there is more ground to lose, and it goes faster.

So the honest version is deferral, not prevention. That is still worth having, because 2.54 more years of ordinary memory is not nothing at all. It just is not the thing most people think they are buying when they read the headline.

Two of these are worth flagging because they're easy to ignore and unusually fixable. Hearing loss and untreated vision loss both sit on the list, and both have straightforward interventions that most people put off for years. If you're going to act on one thing after reading this, getting your hearing and eyes checked has considerably more evidence behind it than any puzzle you'll play today.

How good is the evidence on hearing, actually?

Better than for puzzles, and worth walking through rather than just asserting, because it's the one recommendation on this page with a large randomised trial behind it.

The ACHIEVE trial, published in The Lancet in 2023, randomised 977 adults aged 70 to 84 who all had untreated hearing loss. Half received a hearing intervention with hearing aids and audiologist support. Half received health education as a control. Both groups had their thinking and memory tracked for three years.

Now the honest headline, because it is not the one usually quoted. Across the full cohort, the hearing intervention did not reduce three-year cognitive decline. The primary result was null.

But the trial contained two distinct groups of participants, and a prespecified analysis found the effect differed between them. Among the 238 people recruited from an existing heart-health study, who were at higher risk of cognitive decline to start with, hearing intervention slowed the loss of thinking and memory by 48 percent over three years. The authors' reading is appropriately narrow: a hearing intervention might reduce cognitive change in populations at increased risk, but not in populations at lower risk.

So this isn't a blanket promise that hearing aids protect everybody's memory. It's a large randomised trial showing a substantial effect in the people most at risk and no effect in those who weren't. Now compare it with word games, where the equivalent trial has never been run at all. A null primary result with a genuine subgroup finding still outranks an association nobody has tested.

The practical read lands the same either way. Untreated hearing loss is on the Lancet list, treating it is straightforward, and if you're in the higher-risk group it's the option with actual trial evidence attached. A crossword isn't competing with that.

Why does this matter more in Asia?

Because the numbers here are moving faster than almost anywhere else. Burley and colleagues looked at 19 countries in the Western Pacific Region and found it currently holds the third-highest dementia prevalence globally, and is projected to have the highest prevalence in the world by 2050, with more than 76 million cases.

That's a regional problem being met, in a lot of places, with puzzle apps marketed as brain training. The Lancet list is unglamorous by comparison. Blood pressure, hearing aids, exercise, not smoking, staying socially connected. None of it sells an app subscription, and all of it has better evidence than the thing that does.

So keep playing. Word games are enjoyable, they're mildly good for you on the evidence we walked through above, and enjoying something is a perfectly good reason to do it. Just don't let a daily crossword stand in for a blood pressure check. And nothing here is medical advice, so take the actual health questions to a doctor.

Does it matter whether you play alone or with someone?

It might be the most important thing on this page, and it falls straight out of the list you just read.

Go back to the 14 modifiable risk factors. Crosswords are not on it. Social isolation is. Now notice what a game of Scrabble across a kitchen table actually is: a word puzzle and an hour with another person, at the same time. Only one of those two things has a matching entry on the Lancet list, and it is not the puzzle.

The evidence on that half is substantial. Desai and colleagues, publishing in the American Journal of Geriatric Psychiatry Open Science, Education and Practice in 2025, followed 10,572 adults from the Chicago Health and Aging Project for a mean of 8.8 years, with some tracked as long as 18.7.

For each one-unit increase in social engagement, the rate of cognitive decline slowed by roughly 13.5 percent. Then they split the sample, and the split is the interesting part.

• Lonely participants, 1,757 people or about 17 percent of the sample, declined 16.7 percent more slowly.
• Participants who were not lonely, 8,804 people, declined 11.8 percent more slowly.
• People with more depressive symptoms, 2,481 people or 23 percent, declined 14.3 percent more slowly, against 12.2 percent for those with few or none.

Read that pattern properly. The people who had the least social contact got the most out of adding some. If you are already surrounded by people, a weekly games night is a smaller marginal gain. If you are not, it is the largest lever in this article.

One detail about the sample is worth flagging, because this article has criticised generalisability twice already. This cohort was 63 percent Black and 61 percent female, which is unusual for a study this size and makes it a genuinely useful counterweight to the crossword trial, whose participants were unusually highly educated.

Does this one escape the causation problem?

No, and it would be dishonest to pretend otherwise after spending half this page on exactly that objection.

CHAP is observational. The arrow could run the other way: people whose thinking is starting to slip often withdraw from company, cancel arrangements and stop going out, which would produce this same pattern with decline causing isolation rather than the reverse. The design cannot separate them, same as the puzzle surveys.

But there is a real difference in standing, and it is worth being precise about it. Social isolation made the Lancet Commission's list of 14 factors. Puzzle habits did not. That list is the Commission's judgment about which evidence is strong enough to act on, drawn from far more than one cohort. So while this particular study has the same limitation as the word game surveys, the underlying claim sits on much firmer ground than they do.

What that means practically:

• Play across a table when you can. Scrabble, Boggle, a crossword you argue over with someone. You get the puzzle either way and the company only one way.
• A regular fixture beats an occasional one. A standing game with the same person or a club does the social work through repetition, which is what the engagement measure was actually capturing.
• Online counts more when there is talking in it. A live game with conversation is closer to the thing being measured than trading asynchronous turns with a stranger.
• Don't let solo play stand in for contact. A daily puzzle on your phone is a fine five minutes. It is not social engagement, and quietly filing it as such is the mistake worth avoiding.
• Watch the streak trap again. If a daily game is the reason you skipped a phone call, the trade has gone the wrong way.

If mood rather than memory is what brought you here, our guide to games for stress relief covers that side, and the depressive-symptom finding above is a reminder that the two questions overlap more than people assume.

Do word games make you feel better?

This is the claim people make most casually and the evidence supports it least, at least for word games specifically. Games as a category do seem to help mood. Word games are in the part of that category that did not.

The clearest read comes from a meta-analysis in the Journal of Medical Internet Research in 2022, pooling 15 randomised controlled trials across 1,280 older adults. Overall, games reduced depression with a standardised mean difference of -0.54, which is a medium effect and a real one.

But the breakdown is the whole story. Games involving physical activity drove it, at -0.60. Games combining physical activity with a cognitive task did better still, at -0.73. Games that were purely cognitive showed no significant effect at all. A crossword is purely cognitive. So the headline number is not yours to claim.

Take the quality seriously before you take any of it too seriously. The authors report that just 1 of the 17 studies they reviewed was high quality with a low risk of bias, and about 5 of 17 carried a high risk. That is a shaky base for a confident statement in either direction.

A trial that looked at puzzles directly landed in the same place. A pilot published in JMIR Aging in 2023 put 12 adults aged 46 to 75 through eight weeks of tablet puzzle games against newspaper reading, in a crossover design. Visual attention improved significantly. Mood disturbance, trait anxiety, state anxiety and quality of life did not shift at all. The authors name the tiny sample as their main limitation themselves, so read it as a hint rather than a verdict.

None of that means word games make you feel worse, and it is not an argument against playing them. If a crossword over coffee is the calmest fifteen minutes of your day, that is a real thing and you do not need a trial to justify it. Enjoying something is reason enough.

What it does mean is that you should not pick a word game as your treatment for low mood. If that is the goal, the evidence points at whatever gets you moving, ideally with other people, and both physical inactivity and depression sit on the Lancet Commission's list of 14 anyway. Play the puzzle because you like the puzzle. Go for the walk because it is the walk that does the work.

Do the benefits last if you stop playing?

Some of them, for a surprisingly long time. But not the one most people are hoping for, and the split is the useful part.

The trial that can actually answer this is ACTIVE, which the sources below already lean on. Rebok and colleagues published the ten-year follow-up in the Journal of the American Geriatrics Society in 2014, and you can find it on PubMed. It followed 2,832 older adults across six US cities, average age 73.6 at the start. Each person got ten training sessions in one of three things, memory, reasoning or speed of processing, with booster sessions at 11 months and 35 months. Then researchers came back a decade later.

Reasoning and speed training still showed an advantage on the ability they trained, ten years on. Memory training did not.

Read that split against what a crossword actually is. Word puzzles sit closest to the verbal and memory end of that spectrum, which is the arm where the ten-year advantage did not hold. That is not a reason to stop doing them. It is a reason to be careful with anyone selling you a word game as a durable memory investment.

One more finding needs handling honestly. Participants in all three groups reported less difficulty with everyday activities like managing money and medication. That sounds like the result that matters most, and it might be. But it was self-reported, and people who know they spent weeks in a training programme tend to report that it helped. Set it beside the measured cognitive scores and it is the softer of the two.

The honest summary is narrower than the headlines. Structured training on a specific ability can leave a trace a decade later. Nobody has shown that doing puzzles for fun does the same thing, because nobody has run that trial. Which is an argument for keeping the habit going rather than doing a burst and expecting it to bank, and it lines up with the consistency point in the next section.

Does it matter what age you start?

Later than most people assume, and the trials are unusually clear on this one.

Look at who was actually enrolled. The Columbia and Duke crossword trial had a mean age of 71.2, and every participant already carried a diagnosis of mild cognitive impairment. These weren't healthy people getting ahead of a problem. They were people in whom the problem had already started, and the crossword group still moved in the right direction on the ADAS-Cog while the comparison group drifted the other way. Whatever the training did, it did it after the decline was underway.

ACTIVE tells the same story across a wider span. It enrolled adults aged 65 to 94, with a mean baseline age of 73.6, and the gains in the trained domain showed up across that whole range rather than clustering in the youngest participants. Ninety-four is not an age at which the research says do not bother. The PROTECT sample stretched down the other way, taking in adults from 50 up.

So there's no deadline. Now the uncomfortable half of the answer.

The 2024 Lancet Commission sorts its 14 risk factors across three life stages, and there is exactly one factor in the early-life bracket: educational attainment. A long duration of education is protective. That's the cognitive reserve argument in its plainest form, and the thing it points at got built decades before anyone picks up a crossword. The reserve that seems to matter most is laid down in your teens and twenties, through schooling, not through puzzles in retirement.

Put both halves together and you get something more useful than either alone. No trial has found an age past which mental effort stops doing anything, so starting at 60 or 75 is not too late in any sense the evidence supports. But the head start that shows up most strongly in the risk models isn't purchasable later. If you're reading this wondering whether you left it too long, the answer is no. If you're reading it hoping a puzzle habit will close a gap that opened forty years ago, that's a bigger ask than the research can carry.

The practical version is the same at any age: pick something hard enough to be work, keep at it, and don't let it crowd out the things in the risk-factor list that have better evidence behind them.

How should you play to get any benefit?

Four things, and only one of them is about which game you choose.

Play at the edge of your ability. This is the whole game. A puzzle you finish comfortably is entertainment, which is fine but is not training. If you are getting through it without pausing, move up a difficulty level. Struggle is the active ingredient.

Go for consistency over intensity. Ten to fifteen minutes most days beats an hour on Sunday. The crossword trial ran intensive training for 12 weeks with boosters afterwards, so sustained practice over months is what the evidence actually tested.

Rotate what you play. Getting better at one puzzle format mostly makes you better at that format. Mixing crosswords with anagrams and a daily guessing game covers more ground than drilling one thing.

Keep the basics in place. Sleep, aerobic exercise, blood pressure control and social contact all have stronger evidence behind them than any puzzle. Word games sit on top of those, not instead of them.

If you want somewhere to start, our free word games include a daily word guess that works well as an anchor habit, plus word scramble for the working memory side and word search when you want something lower effort. All free, no signup.

And if you would rather see the wider picture across puzzle types, our guide to the benefits of puzzle games covers the non-verbal side.

Does a daily game like Wordle count?

As a habit, yes. As the thing the crossword trial measured, no, and the gap between those two answers is mostly about dose.

Run the comparison honestly. The Columbia and Duke trial gave people 12 weeks of intensive home-based crossword training followed by booster sessions, tracked over 78 weeks. A daily word puzzle takes most people three to six minutes. That's a real habit and it is nowhere near the same exposure. Nobody has run a randomised trial on three minutes a day, so anyone telling you a daily game protects your memory is extrapolating well past the evidence.

The more useful question is whether the thing you're doing is still hard. Everything on this page points the same way: the benefit, such as it is, tracks effort rather than time spent. A puzzle you solve on autopilot is a pleasant routine, not training. And short daily games are unusually prone to going stale, because the format never changes and you get efficient at it fast.

Which gives you a straightforward test. If you're still occasionally stuck, still occasionally failing, and still having to work at it, the game is doing something. If you clear it in ninety seconds without effort most days, you've stopped training and started collecting a streak. Neither is bad. Just don't confuse them.

Three ways to keep a daily habit worth having:

• Raise the difficulty when it gets easy. Move to a harder variant, a bigger grid, or a timer. The point of a puzzle is the part where you're stuck.
• Rotate the format. Anagrams, crosswords, word searches and cryptics ask for genuinely different things. A word search in particular is mostly visual scanning rather than retrieval, so it's the one least like the crossword that was actually tested.
• Don't let the streak become the goal. Once you're playing to protect a number rather than because the puzzle is interesting, the habit has quietly inverted.

And keep it in proportion. A daily game is a good five minutes. The things with the strongest evidence behind them are in the previous section, and none of them fit in a browser tab.

What else do people ask about word games and the brain?

Do word games actually improve your brain?

They reliably improve the skills you practise, and one randomised trial found crosswords outperformed computerised brain games in people with mild cognitive impairment. But that trial had no do-nothing group, so it shows crosswords beat one alternative rather than proving they beat doing nothing at all. Treat word games as a good habit with genuine but modest evidence behind them.

Are crosswords better than brain training apps?

In one 78-week trial of 107 people with mild cognitive impairment, yes. Crossword training beat computerised cognitive games on the primary memory measure by 1.44 points on the ADAS-Cog scale, and the games group showed greater shrinkage in hippocampal volume. That is one trial in one population, not a settled verdict, but it does undercut the idea that a purpose-built app automatically beats an old-fashioned puzzle.

Does doing word games prevent dementia?

No, and nobody has shown that. The studies point to better cognitive scores and slower decline in specific groups, not prevention. Sleep, regular aerobic exercise, blood pressure control and social contact all have stronger evidence than any puzzle does. Word games are a reasonable addition to those, not a substitute for them.

How often should you play word games?

Ten to fifteen minutes most days beats a long session once a week. The crossword trial ran intensive home training for 12 weeks followed by booster sessions, so consistency over months is what the evidence actually tested. Pick a game you enjoy enough to keep returning to, because an abandoned routine produces nothing regardless of what the research says.

Which word game is best for your brain?

The one that stays slightly too hard for you. Crosswords have the most direct trial evidence behind them. Word searches train visual scanning more than vocabulary, and anagram games work retrieval speed. Whichever you pick, step the difficulty up once it starts feeling automatic, because comfort is the point at which the training stops.

For general information and self-reflection only. Not medical advice. If you are worried about changes in your memory or thinking, speak to a doctor.

Sources: Brown A.S., A review of the tip-of-the-tongue experience, Psychological Bulletin 109(2), 204 to 223, 1991. Shafto M.A., Burke D.M., Stamatakis E.A., Tam P.P. and Tyler L.K., On the tip-of-the-tongue: neural correlates of increased word-finding failures in normal aging, Journal of Cognitive Neuroscience 19(12), 2060 to 2070, 2007. James L.E. and Burke D.M., Phonological priming effects on word retrieval and tip-of-the-tongue experiences in young and older adults, Journal of Experimental Psychology: Learning, Memory, and Cognition 26(6), 1378 to 1391, 2000. Devanand D.P., Goldberg T.E., Qian M. et al., Computerized Games versus Crosswords Training in Mild Cognitive Impairment, NEJM Evidence, 2022, 107 participants, funded by the National Institute on Aging. Toma M., Halpern D.F. and Berger D.E., Cognitive Abilities of Elite Nationally Ranked SCRABBLE and Crossword Experts, Applied Cognitive Psychology, 2014, volume 28, pages 727 to 736, comparing 26 Scrabble and 31 crossword experts against college students matched on SAT scores averaging above 700. Terada T. and Lee Y., The Association Between Leisure Games and Cognitive Functioning Among Older Adults, Innovation in Aging, 2021, 3,271 adults aged 65 and over from the Health and Retirement Study. PROTECT study led by the University of Exeter and King's College London, more than 19,000 adults aged 50 and over, reported by the Alzheimer's Society, alzheimers.org.uk, including Dr James Pickett's caveat that the research shows a link rather than an effect. The underlying findings were published as two linked papers in the International Journal of Geriatric Psychiatry in May 2019 (PubMed 30443984, doi 10.1002/gps.5033), and the University of Exeter announcement, news-archive.exeter.ac.uk, reports performance equivalent to ten years younger than actual age on grammatical reasoning and eight years younger on short-term memory. Both are cross-sectional associations measured at a single sitting, not before-and-after gains, and the study is frequently misdated to 2023 in secondary coverage. Pillai J.A., Hall C.B., Dickson D.W., Buschke H., Lipton R.B. and Verghese J., Association of Crossword Puzzle Participation with Memory Decline in Persons Who Develop Dementia, Journal of the International Neuropsychological Society, 2011, drawn from the Bronx Aging Study cohort of 488 cognitively normal residents assessed every 12 to 18 months, analysing the 101 who developed incident dementia, reporting a 2.54 year delay in the onset of accelerated memory decline and a subsequently faster decline of 3.31 points per year on the Buschke Selective Reminding Test. van Hees S., Pexman P.M., Hargreaves I.S. et al., Testing the Limits of Skill Transfer for Scrabble Experts in Behavior and Brain, Frontiers in Human Neuroscience, November 2016, 19 competitive Scrabble experts and 19 age-matched controls. Rebok, G. W. et al. (2014). Ten-year effects of the Advanced Cognitive Training for Independent and Vital Elderly cognitive training trial on cognition and everyday functioning in older adults. Journal of the American Geriatrics Society, 62(1), 16 to 24. 2,832 participants, mean baseline age 73.6, ten-year follow-up (PubMed 24417410). Agency for Healthcare Research and Quality, comparative effectiveness review on cognitive training, March 2017, covering the ACTIVE trial of 2,832 adults. National Institute on Aging guidance on cognitive health and commercially available brain training applications. Federal Trade Commission, Lumosity to Pay $2 Million to Settle FTC Deceptive Advertising Charges for Its Brain Training Program, press release January 2016, ftc.gov, on the claims alleged, the quoted finding that the company preyed on consumers' fears and lacked the science to back its ads, and the undisclosed prize-solicited testimonials. Lin F.R., Pike J.R., Albert M.S. et al., Hearing intervention versus health education control to reduce cognitive decline in older adults with hearing loss in the USA (ACHIEVE): a multicentre, randomised controlled trial, The Lancet, 2023, 977 adults aged 70 to 84 with untreated hearing loss, no reduction in three-year cognitive decline in the primary analysis of the total cohort, and a 48 percent slowing over three years in the prespecified higher-risk subgroup of 238 participants drawn from an existing observational heart-health cohort. Bialystok E., Craik F.I.M. and Freedman M., Bilingualism as a protection against the onset of symptoms of dementia, Neuropsychologia, 2007, records of 184 memory clinic patients, with the follow-up in Neurology reporting diagnosis 4.3 years later. Bilingualism Is Associated with a Delayed Onset of Dementia but Not with a Lower Risk of Developing it: a Systematic Review with Meta-Analyses, pooling prospective studies covering 5,527 bilingual participants, odds ratio 0.96 with a confidence interval of 0.74 to 1.23. A Systematic Review on the Possible Relationship Between Bilingualism, Cognitive Decline, and the Onset of Dementia, on the mixed findings and the study-design differences behind them. Desai P., Ng T.K.S., Krueger K.R., Wilson R.S., Evans D.A. and Rajan K.B., Social Engagement, Depressive Symptoms, and Loneliness, and Longitudinal Cognitive Decline in a Population-Based Cohort of Older Adults, American Journal of Geriatric Psychiatry Open Science Education and Practice, 2025, volume 6, pages 39 to 48, covering 10,572 participants in the Chicago Health and Aging Project over a mean 8.8 years, observational and therefore subject to the same direction-of-causation limits as the survey evidence above. Zhang J. et al., Effects of Serious Games on Depression in Older Adults: Systematic Review and Meta-analysis of Randomized Controlled Trials, Journal of Medical Internet Research, 2022, 15 randomised controlled trials pooled across 1,280 older adults, standardised mean difference -0.54 overall, -0.60 for games involving physical activity and -0.73 for combined physical and cognitive games, with no significant effect for purely cognitive games, and only 1 of 17 reviewed studies rated high quality with a low risk of bias. Tablet-Based Puzzle Game Intervention for Cognitive Function and Well-Being in Healthy Adults: Pilot Feasibility Randomized Controlled Trial, JMIR Aging, 2023, 12 adults aged 46 to 75 in a crossover design over 16 weeks, significant gain in visual attention and no significant change in mood disturbance, anxiety or quality of life.