Does Working Memory Training Actually Work? A Sceptical Look at the Evidence

Working memory is the cognitive system that lets you hold a few items in mind while doing something with them. The phone number you remember just long enough to type it. The list of three things you walked upstairs to grab. The opening of a sentence you are still trying to finish.

It is one of the strongest predictors of academic and professional performance. Better working memory correlates with better reading comprehension, better mathematical reasoning, and better problem-solving in nearly every domain that has been measured. Which is why "train your working memory" has become a small industry. Cogmed, Lumosity, Peak, BrainHQ, and dozens of smaller apps all promise that practising specific exercises will increase your working memory capacity, and that this will then transfer to better performance in real life.

The honest version of the evidence: the first claim is roughly true. The second claim is mostly false.

If you want to test your own working memory before reading on, our Number Memory test and Sequence Memory test take about a minute each.

What working memory actually is

The technical model that dominates research separates working memory into three components: a phonological loop for verbal information, a visuospatial sketchpad for visual information, and a central executive that allocates attention and coordinates the others.

In practical terms, working memory is the holding bay for whatever your mind is currently working with. It is small (around four to seven items for most adults), brief (lasting a few seconds without rehearsal), and bottlenecked (you can only juggle so much at once).

People who score higher on working memory tests are also, on average, better at the harder tasks that require holding partial results in mind while continuing to compute. This is why working memory shows up as a predictor of so many real-world cognitive outcomes.

What "training" usually means

The standard working memory training paradigm asks you to do a working-memory-intensive task repeatedly, day after day, for several weeks. The two most studied tasks are:

N-back. You see a stream of items (letters, positions on a grid). After each new item, you indicate whether it matches the item from N items ago. Easy versions are 1-back. Hard versions are 4-back or higher. Dual N-back makes you track two streams (visual and auditory) simultaneously.

Complex span. You alternate between storing items (a digit, a word) and processing other information (solving a small arithmetic problem). After several rounds, you recall the stored items in order.

Both tasks push the limits of working memory and force adaptation. After a few weeks of practice, performance on the trained task improves substantially. The question is whether anything else improves with it.

The transfer problem

This is where the field has been arguing for two decades.

In 2008, Susanne Jaeggi and colleagues published a paper in PNAS claiming that dual N-back training improved fluid intelligence (the ability to reason and solve novel problems). The paper kicked off a wave of commercial training products and a wave of replication attempts.

The replication record has been disappointing.

Melby-Lervåg and Hulme's 2013 meta-analysis pooled 23 studies of working memory training across age groups. Their finding: trained tasks improved substantially (large effect sizes), closely related tasks improved modestly (small effect sizes), and unrelated cognitive abilities like fluid intelligence, attention, and reading comprehension did not improve in any reliable way.

Sala and Gobet's 2017 meta-analysis of working memory training in children reached similar conclusions. Training improves the trained task. It does not reliably improve other things.

Simons and colleagues' 2016 paper in Psychological Science in the Public Interest reviewed the entire brain-training industry's evidence base. The conclusion: there is good evidence that brain-training programs improve performance on the specific tasks practiced, modest evidence for transfer to closely related tasks performed in the lab, and little evidence that they improve real-world performance.

Translation: you will get better at the test. You will get marginally better at very similar tests. You will not get more generally intelligent, better at school, better at work, or sharper in the broad way the marketing suggests.

This is the honest standing of the field.

What is actually happening when you practise

When you do an N-back task every day for three weeks, your score on N-back climbs. The question is what is causing that climb.

Three things contribute, in roughly this order of importance:

Task-specific learning. You become more familiar with the format. You learn faster what to look for, how to encode items, what strategies work. Most early improvement is this.

Strategy refinement. You discover, often without noticing, that certain mental approaches work better than others. Chunking items, using verbal labels for visual ones, rehearsal techniques. These strategies are specific to the task structure.

Modest capacity expansion. A small portion of the improvement may reflect genuine expansion of the underlying working memory capacity. This is the part that should, in theory, transfer. In practice, it does not transfer well.

The mismatch between trained-task gains and real-world gains is the heart of the problem. You are mostly getting better at the task, not at the underlying ability that the task was supposed to measure.

Where the picture is slightly more positive

A few specific populations and outcomes have shown more promising results.

Children with ADHD. Some studies have shown modest improvements in attention and academic outcomes after working memory training, though replication has been mixed. The most commonly used commercial program in this space (Cogmed) has a more substantial evidence base than most consumer apps, but the effects are still modest and contested.

Older adults experiencing mild cognitive decline. Working memory training appears to slow some aspects of decline in older adults, though the effect is modest and unclear whether it generalises beyond trained tasks.

Healthy young adults. This is the population that buys most brain-training apps. The evidence for benefits in this group is the weakest.

What is more likely to actually improve real-world performance

If your goal is to think more clearly in daily life, the evidence supports several interventions much more strongly than brain training.

Sleep. Working memory is exquisitely sensitive to sleep loss. A single night of poor sleep can shrink your effective working memory by 15 to 20 percent. The Lim and Dinges meta-analysis covered in our reaction time piece makes this clear.

Aerobic exercise. Cardiovascular fitness is one of the most robust correlates of cognitive function across the lifespan. Regular aerobic exercise improves working memory, attention, and processing speed in ways that brain training does not.

Learning something hard. Picking up a new language, a musical instrument, or a complex skill produces broader cognitive benefits than doing the same brain-training game repeatedly. The novelty and depth of engagement matter.

Reducing chronic stress. Cortisol over time damages the hippocampus and impairs working memory. Practices that reliably lower stress (meditation, regular exercise, social connection) protect cognitive function.

Diet and metabolic health. Insulin resistance, high blood sugar, and chronic inflammation all impair cognition. Improving these helps working memory more than any app.

None of these are as easy to package as a 5-minute daily game. They are also far more effective.

When working memory training might still be useful

A few honest reasons to do these games.

For diagnostic feedback. Tracking your performance on a simple test like our Sequence Memory over time gives you a low-stakes window into how rested, hydrated, and alert you are. Big day-to-day changes in score often correlate with sleep, stress, or illness.

For modest task-specific improvement. If your job involves specific working memory demands (interpretation, simultaneous translation, complex multi-step calculations), practising similar tasks may help with those specific tasks.

For warming up the brain before demanding work. A few minutes of working memory exercise as part of a morning routine can prime attention for the day. The effect is small and short-lived.

For curiosity. Watching your own cognitive limits is interesting. Trying to push them is harmless.

The framing that helps most: treat these as games to enjoy and tests to track, not as training that will measurably make you smarter.

What about dual N-back specifically

Dual N-back has the strongest claim to producing some transfer, originating from the original Jaeggi work. Subsequent replication has been mixed. Some careful studies have shown small fluid intelligence improvements after extensive training (20+ sessions of 20+ minutes). Others have not.

If you want to try it, the honest expectation is that you may see a small improvement on similar tests if you train consistently for several weeks. You probably will not see meaningful real-life cognitive change.

Our piece on n-back versus dual n-back covers the practical differences and what the evidence actually shows.

The most honest summary

The strong claim ("brain training makes you generally smarter") is not supported by the evidence.

The weak claim ("you can get better at the specific task you practise") is well-supported but commercially unimpressive.

The useful framing ("these are tools for testing and tracking your current cognitive state, not for training your underlying intelligence") is closer to the truth.

The boring evidence-based playbook for actual cognitive improvement is: sleep, exercise, learn hard things, manage stress, eat well, stay socially connected. None of it is as exciting as an app. All of it works better.

FAQ

For more on individual cognitive tests, see our pieces on reaction time training and the Stroop effect. The full menu of cognitive tests sits in our brain games hub.