Does brain training actually work?

This site is a set of brain tests, so it is worth being straight about what the evidence says — including the parts that argue against the whole category.

The distinction everything turns on

Researchers separate two things, and almost all the confusion comes from collapsing them.

Near transfer is improving at the trained task, or at something very like it. Practise a digit span task and your digit span rises. This is real, reliable and uncontroversial.

Far transfer is improving at something different — your memory generally, your concentration at work, your academic results, your risk of cognitive decline. This is what brain training products have historically been sold on.

The evidence for near transfer is strong. The evidence for far transfer is weak, and the gap between the two is where the industry has lived.

What actually happened with Lumosity

In January 2016 the US Federal Trade Commission settled with Lumos Labs, makers of Lumosity, over its advertising. The judgment was $50 million, suspended to $2 million on the company’s financial condition.

The FTC’s complaint was that the company had claimed its games would improve performance at work and school, delay age-related decline and protect against dementia, without adequate evidence for any of it. The order barred those claims unless supported by competent and reliable scientific evidence.

That is the single most useful fact in this area, because it was not a scientific dispute about a subtle effect. It was a regulator finding that the headline claims were not supported.

What the research generally finds

Reviews of the field have converged on a broadly consistent picture over the last decade:

• Trained tasks improve substantially. Often dramatically, and quickly.
• Closely related tasks improve a little. Practise one working memory task and a second, similar one may improve somewhat.
• Unrelated abilities mostly do not improve. Reasoning, general intelligence, academic performance and everyday memory complaints show little to no reliable benefit.
• Studies with active control groups find smaller effects than those comparing against people who did nothing. Some of what looks like improvement is expectation and familiarity with being tested.

That last point matters more than it sounds. If your control group sits at home and your training group plays engaging games and is told they are improving, some improvement will appear whatever the games do.

What practice on these tests genuinely gets you

Being specific about what does improve, because something does.

Technique. Most of the early gain on a memory test is learning to chunk. Reading 4 9 1 2 6 3 as 49 12 63 genuinely lets you hold more, and that skill transfers to any digits you need to hold — a phone number, a door code. It is narrow, and it is real.

Task familiarity. Knowing where to look, when the signal is coming, how the controls behave. Worth 20 to 30 milliseconds on a reaction test in the first few attempts, and then nothing more.

A measurement you can use. This is the honest case for the whole thing. A repeatable test lets you see what a bad night’s sleep costs you, or what your afternoon slump actually looks like as a number. That is not training — it is instrumentation, and it is genuinely useful in a way “improving your brain” is not.

The dual n-back story, which is instructive

If any single task was going to produce far transfer, it was this one.

In 2008 a study reported that training on a demanding working-memory task — dual n-back, where you track two streams of stimuli and report matches from n steps back — improved fluid intelligence, and that the improvement scaled with how much training people did. Fluid intelligence had long been considered close to fixed in adults, so the claim was substantial and it drew enormous attention.

What followed is a good illustration of how this field works. Replication attempts produced mixed results. Studies with active control groups found smaller effects than those with passive ones. Meta-analyses that accounted for control-group design concluded the transfer to fluid intelligence was small at best and plausibly zero, while the improvement on n-back itself was large and unambiguous.

The pattern — impressive near transfer, disappearing far transfer, effects shrinking as designs tighten — is the pattern for the whole category. It is worth knowing about dual n-back specifically because it was the strongest case anyone had, it was studied more carefully than most commercial products ever are, and it still came out this way.

What this means for the tests here

None of the eight tests on this site is a training programme, and describing them as one would be exactly the claim the evidence does not support.

What they are is repeatable measurements. That distinction sounds like hedging and it is actually the whole design:

• The reference ranges are fixed, so your score cannot drift upward because the test got easier or because the population changed. A training product has a commercial reason to let your numbers rise; a measurement does not.
• There is no “cognitive age”, no personalised programme and no assessment of potential. Those are inferences the data cannot support, and they are close to what the FTC action was about.
• The daily round reports a number and a history and stops there. It does not tell you what the number means about you, because a browser test cannot know.

The honest pitch is narrow: if you want to see what a bad night’s sleep costs you, or whether your afternoon slump is real, this will show you. It will not make you sharper, and a site that says otherwise is selling something.

What does have evidence behind it

If the goal is cognitive function rather than a higher score, the interventions with better support are unglamorous:

Sleep. The largest and most consistent effect available, and the one most often traded away. Restriction impairs attention, working memory and reaction time reliably.

Aerobic exercise. Modest but reasonably supported benefits for executive function, especially in older adults.

Not being sedentary, socially isolated or untreated for hearing loss. These appear consistently in the dementia risk literature, which is a different question from short-term performance but the one people usually mean when they worry about it.

Learning something genuinely difficult. An instrument, a language, a craft. The evidence here is weaker than the enthusiasm, but it has one clear advantage over training games: the skill is worth having on its own terms whether or not it transfers.

So why build this site

Because measuring is a different proposition from training, and it is the one we think is defensible.

The daily round exists to give you a repeatable number and a history to put it in. It will tell you whether today was slower than last Tuesday. It will not make you smarter, and nothing on this site claims it will — there are no “cognitive age” figures, no personalised training programmes and no assessments of your potential, because those are the claims the FTC action was about.

The tests are also, straightforwardly, quite good fun for five minutes. That is a sufficient reason for a thing to exist, and a more honest one than most of the category offers.

Why the effects look bigger in some studies

If the evidence is as unimpressive as described above, the volume of positive-sounding studies takes some explaining. Three things account for most of it.

Passive control groups. If your training group plays engaging games while your control group does nothing, some of the difference is expectation and some is simply being more practised at being tested. Studies using an active control — another absorbing activity — consistently find smaller effects.

Testing on tasks that resemble the training. Improvement on a task closely related to the one you trained is near transfer, and it is not in dispute. Reporting it as evidence of general cognitive benefit is where the overreach happens.

Publication bias. Small positive results are more publishable than small null ones, so the visible literature skews positive relative to what was actually run. This is not unique to brain training, but the commercial interest in the field makes it more consequential.

None of this means the researchers were dishonest. It means the design choices that make a study easy to run are the same ones that inflate its apparent effect.

The one thing that reliably improves

Your ability to do the specific thing you practised. That is worth taking seriously rather than dismissing.

If your job involves reading numbers off a screen and holding them briefly, practising a digit span task will make you better at reading numbers off a screen and holding them briefly. If you play a game requiring fast target acquisition, an aim trainer will help a little with that specific component.

What will not happen is a general lift. The honest framing is that these are narrow skills with narrow benefits, which is a perfectly reasonable thing to want — it is only a problem when sold as something else.

How to read your own scores

Compare yourself against your own history, on the same device, at roughly the same time of day. Expect the first several attempts at any test to improve sharply — that is you learning the task, not you improving. The interesting part starts after that plateau, when the remaining variation is mostly about your state rather than your familiarity.

And treat a bad day as a bad day. If you are genuinely worried about your memory or concentration, a browser test is not the instrument for that question, and a doctor is.