How to improve working memory: what the research actually supports

Working memory is the system that holds information in mind while you do something with it. It is what lets you keep a phone number in mind while you dial, follow a multi-step recipe, or hold the start of a sentence while you finish reading it. The capacity is small, roughly four to seven items, and it can be improved on trained tasks reliably. Whether that training transfers to real-world cognition is contested. This guide covers what the evidence supports doing, what to skip, and why.

In 30 seconds: Train daily and briefly, with progressive difficulty, on tasks that match the working-memory demand you actually want to improve. Sleep enough. Move your body. Be skeptical of any product or supplement claiming a broad cognitive boost.

What working memory actually is

The term gets used loosely. Cognitively, working memory has three properties:

1. Limited capacity. Nelson Cowan's 2010 Current Directions review put the working figure at four chunks, give or take, in young adults under controlled conditions. The classic "magical seven" applies to short-term memory more broadly, including chunking.
2. Active manipulation. Working memory is not just storage. It is storage plus operation: holding the digits while reversing them, holding the meaning while integrating new information, holding a list while comparing options.
3. Domain dispatch. Verbal working memory and visuospatial working memory rely on partly separable systems (Baddeley's classic model). Training one does not automatically improve the other.

Most everyday cognitive complaints attributed to "memory" are actually working-memory complaints: losing your place in a sentence, forgetting why you walked into a room, struggling to follow a complex conversation. The doorway effect is a working-memory failure mode, not a long-term-memory one.

What the central debate is

Working memory training is the most studied corner of cognitive training, and it sits at the heart of the field's biggest argument: near transfer versus far transfer, which we cover in detail in the cognitive training guide.

• Near transfer is robust. Practice the n-back task, get better at the n-back task. Practice digit-span, get better at digit-span. Replicated reliably across studies and ages.
• Far transfer is fragile. Practice n-back, do you get better at fluid intelligence, reading comprehension, or daily-life multitasking? The evidence is mixed at best.

Susanne Jaeggi's 2008 PNAS paper kicked off a decade of optimism by reporting that dual n-back training improved fluid intelligence in young adults, with a dose-response relationship. The original effect was striking. Subsequent replication attempts have produced mixed results, and the most rigorous meta-analysis to date (Melby-Lervåg, Redick & Hulme, 2016) covering 87 studies concluded that working memory training does not reliably transfer to general intelligence, reading, arithmetic, or other measures of far transfer.

"There was no convincing evidence of any reliable improvements."

Melby-Lervåg, Redick & Hulme, 2016, Perspectives on Psychological Science

That does not mean the training is useless. It means the right framing is specific, not general: train the working memory you want to improve, in a form similar to where you want to use it.

Step-by-step: what to actually do

Step 1: Train a specific working-memory task daily

The protocol is simple:

• Five to fifteen minutes per day, every day. Daily-and-brief consolidates better than long-and-occasional, for reasons we cover separately.
• Adaptive difficulty. The task should adjust to your current level so you stay at the edge of competence, not coasting and not failing.
• Match the modality. If you want better verbal working memory (following arguments, holding multiple ideas), train verbal tasks. If you want better visuospatial (driving, navigating), train visuospatial. The transfer is far stronger within modality than across.

Example tasks with reasonable evidence:

• Dual n-back (Jaeggi-style). Strong near-transfer evidence, contested far transfer. Useful as a workout.
• Operation span / running memory (Daneman-Carpenter style). Closer to real-world verbal working memory.
• Adaptive Corsi block for visuospatial.

The protocol shape, not the specific task, is the point. Brief, daily, progressive, modality-matched.

Step 2: Use retrieval practice on real material

The single most under-used working-memory tool in everyday life is retrieval practice. Roediger and Karpicke's 2006 Psychological Science paper showed that retrieving information from memory strengthens it more than re-studying it does. The effect is especially strong for material being held actively in working memory.

In practice:

• Self-quiz a minute after reading, instead of re-reading. Force the working-memory-to-long-term-memory handoff.
• Pause and retrieve mid-conversation instead of trying to track everything passively. "What was their main point?" reactivates the trace.
• Build the habit in low-stakes contexts first. Retrieving the title of the article you just read, or the name of the cafe you went to yesterday, builds the muscle for higher-stakes use.

The same principle drives the four-step method in how to remember names: retrieval, not exposure, is what consolidates.

Step 3: Reduce concurrent cognitive load

Working memory is finite. Most everyday "I forgot what I was doing" is not a memory deficit; it is a capacity problem. The fix is removing competing demand:

• Single-task during important encoding. Reading, listening to a presentation, having a serious conversation. Phones away, browser tabs closed.
• Externalize where possible. Notes, lists, calendar. Working memory is for active manipulation, not for storage that can be offloaded.
• Manage interruptions. Each interruption forces a working-memory reset. Cluster them where possible.

Klingberg's 2010 Trends in Cognitive Sciences review noted that working-memory load can be the rate-limiting step in many real-world cognitive tasks. Reducing load is sometimes more effective than training capacity.

Step 4: Sleep, move, manage cardiovascular risk

The non-app levers are larger than the app levers, by a margin most people underestimate.

• Sleep. Working-memory capacity drops sharply with sleep restriction. Two consecutive nights of less than 6 hours can reduce working-memory performance to a level equivalent to mild alcohol intoxication. We cover the sleep-memory mechanism in detail.
• Aerobic exercise. Erickson et al.'s 2011 PNAS trial showed that one year of moderate aerobic exercise grew the hippocampus by 2% in older adults, with corresponding memory gains. Aerobic capacity is one of the most reliable correlates of working-memory performance across the lifespan.
• Cardiovascular risk management. Hypertension, untreated diabetes, and high LDL cholesterol all degrade brain vascular function and working-memory performance. Per the Lancet Commission's 14-factor framework, these are some of the highest-leverage modifiable interventions.

Skip these and any working-memory training is fighting headwind. Address them and modest training has a chance.

What does not work (or works less than marketing suggests)

A few interventions have weaker evidence than their popularity implies:

• Generic "brain training" puzzles with no progressive adaptation. They make you better at the puzzle. Transfer is poor, per the Simons et al. 2016 consensus review.
• "Memory" supplements. Ginkgo, lion's mane in healthy adults, B-complex megadosing in non-deficient adults, "memory" blends. The clinical-trial evidence ranges from null to mixed.
• Massed-practice cram sessions. A 90-minute working-memory blitz once a week is consistently outperformed by 10 minutes daily, for the same total minutes. The Cepeda et al. 2008 spacing-effect meta-analysis covers the mechanism.
• Apps marketed without specific transfer claims. "Train your brain" without naming what skill, what dose, and what evidence is the marketing equivalent of "wellness."
• Multitasking practice. Practicing multitasking does not increase working-memory capacity. It increases the speed of task-switching for the specific tasks practiced, with corresponding accuracy costs. Most workplace multitasking is just degraded sequential processing.

Common mistakes

Habits that feel productive and are not:

• Training the same task at the same difficulty for weeks. Plateau. The whole point of progressive adaptation is to keep working memory at the edge.
• Switching apps every two weeks. No protocol works in two weeks. Pick one, stick with it for at least six.
• Stopping after a "good" session. Inconsistency does more damage than low intensity.
• Treating it as the main intervention when sleep is broken. Five hours of sleep cannot be compensated for by twenty minutes of n-back.
• Expecting general cognitive benefit. The specific-tasks-only finding is the modal scientific result. Plan accordingly.

Why this works

Each step targets a different bottleneck:

• Daily training keeps the trace fresh and progressively expands the trained representation.
• Retrieval practice consolidates working-memory traces into long-term storage, which frees capacity.
• Reducing load preserves the limited capacity for the task that matters now.
• Sleep, exercise, vascular care maintain the underlying neural infrastructure on which working memory runs.

The honest framing: working memory is a small, hard-trained capacity. You can squeeze a bit more out of it on tasks similar to your training. The bigger lever is using the capacity you have well.

A practical bottom line

• Train daily, briefly, progressively, in a modality matched to your everyday demand.
• Use retrieval, not re-exposure, to convert working-memory traces into durable knowledge.
• Reduce concurrent load. Single-task what matters.
• Sleep, exercise, and manage cardiovascular risk. These do more than any app.
• Be skeptical of broad "cognitive booster" claims. The specific-near-transfer pattern is the modal scientific finding.

For the wider context on what cognitive training works and what doesn't, see the cognitive training guide. For how memory works underneath all of this, see memory 101. For the brain-health stack that working-memory training sits inside, see our brain health guide.