Does Working-Memory Training Improve General Intelligence?

The important difference is between getting better and transferring

Suppose a person practices remembering the last few locations and letters in a stream. After repeated sessions, the person may respond faster, remember more of the same kind of sequence, or use a better strategy. That is a real performance change. It tells us that the practiced activity has become easier under its practiced conditions.

Transfer asks a different question: does the improvement appear on a task the person did not practice? Near transfer uses a similar demand, such as another working-memory task. Far transfer uses a more distant outcome, such as matrix reasoning, verbal ability, an intelligence composite, school attainment, or an everyday activity. General intelligence is an interpretation of performance across multiple cognitive tasks, not the score from one memory game. The distance between the training task and the outcome therefore matters more than the impressive sound of the training label.

This distinction changes how to read a result. A higher raw score on a familiar exercise is evidence of performance on that exercise. It is not, by itself, evidence that an underlying general ability has increased.

What working memory is, and what the exercises actually train

Working memory is the limited ability to keep information temporarily available while using it. A person might hold a partial calculation in mind, update a sequence after each new item, or remember an instruction while carrying it out. Training programs vary: some use recall, some require continuous updating, and some adapt difficulty as the participant responds.

Those tasks combine several operations. They can reward remembering the material, ignoring distractions, learning the response rhythm, spotting recurring formats, and choosing a strategy. Improvement can therefore reflect task-specific learning without requiring a broad change in reasoning capacity. That is why a fair study tests more than the exercise itself and compares the training with a credible control activity.

The phrase working-memory training does not identify one intervention. Dose, task, age, starting ability, control condition, outcome measure, and time between training and retest all affect what a study can show. A result from one program or population should not be generalized to every app, worksheet, or browser quiz.

Why an early positive result attracted attention

A 2014 meta-analysis by Au and colleagues examined n-back training studies with a control group, healthy participants aged 18 to 50, and a fluid-intelligence outcome. The authors reported a small, statistically significant positive effect on fluid intelligence. This is evidence worth reporting, but it is narrower than the headline that working-memory training improves intelligence.

The review focused on one training family, and its result depended on the studies available at that time and the way the comparisons were combined. A small average effect can also be sensitive to study quality, sample size, choice of control, and whether the outcome is measured immediately after training. The finding supports a hypothesis for further testing. It does not supply a universal score increase, a conversion for an individual, or a guarantee that practice will raise an IQ result.

A useful evidence summary keeps both sentences: an early synthesis found a small effect in a defined set of n-back studies, and that claim has not become a settled general rule.

What broader reviews found when controls became stricter

Melby-Lervåg and Hulme's meta-analytic review included 23 studies and 30 group comparisons involving children and adults. It found reliable short-term improvement in working-memory skills, but no convincing evidence that those gains generalized to other abilities such as verbal or nonverbal ability, attention, word decoding, or arithmetic. The review also warned that benefits may not persist without continued training.

A later meta-analysis by Melby-Lervåg, Redick, and Hulme examined 87 publications and 145 experimental comparisons. Against treated control groups, it found improvements in intermediate transfer measures such as verbal and visuospatial working memory immediately after training, but no convincing reliable improvement on far-transfer measures including nonverbal ability, verbal ability, reading, or arithmetic. The authors also reported that improvement on working-memory measures was not related to the size of far-transfer effects across studies.

The control comparison matters. A no-contact group can improve less simply because it receives less contact, expectation, or practice with testing. An active control group helps ask whether the memory program did more than engaging in another demanding activity. This is not a technical footnote; it is part of the causal question.

Controlled studies make the broad promise harder to defend

Redick and colleagues compared young adults who practiced an adaptive dual n-back task with people assigned to an adaptive visual-search control and a no-contact control. Participants improved on the trained tasks, but the study found no positive transfer to its measures of fluid intelligence, multitasking, working-memory capacity, crystallized intelligence, or perceptual speed. Its design illustrates why an improvement on the training task is not enough.

A two-year study followed students who practiced a varied set of working-memory tasks every two weeks. The researchers found substantial and reliable improvement on practiced tasks and on a latent working-memory factor, but no transfer to latent fluid or crystallized intelligence factors. The long duration makes this a useful test of the idea that a larger dose will automatically produce a broad ability change, although one study still does not settle every possible program or population.

A separate randomized study of healthy middle-aged adults used adaptive n-back training and a non-adaptive working-memory control. It found no near or far transfer relative to the control group despite pronounced improvement on the trained n-back task. The authors discuss optimized task strategies as one possible explanation: people can learn how to handle the exercise without changing a broad cognitive capacity.

A worked example: what a result can and cannot support

Imagine a student completes 20 sessions of a memory exercise. On the first session, the student correctly recalls six positions. On the final session, the student recalls nine. The raw performance change is three positions on that exercise under those testing conditions. It may reflect stronger temporary retention, a more efficient grouping strategy, familiarity with the timing, or some combination of these.

Now give the student an unfamiliar reasoning task that requires comparing visual relations, and give it to a control group as well. If the trained group improves more than the control group, that is evidence of transfer to that particular task. If both groups improve similarly, the memory exercise has not demonstrated a training-specific gain there. If only the original exercise improves, the safe interpretation is task-specific practice.

Notice what this example does not permit. It does not convert six or nine correct responses into an IQ. It does not show that the student's general intelligence, potential, or worth changed. It does not predict classroom or job performance. A normed score requires an instrument, standardization sample, administration conditions, and a supported interpretation.

How to interpret a higher reasoning or IQ-test score

First ask whether the retest used the same or very similar items, format, timing, and instructions. Familiarity can improve raw performance even when the target construct has not changed. Then ask whether the score is raw performance, a domain score, a standard score, a percentile, or an estimate. These are not interchangeable. A standard score is a transformed comparison with a specified norm group; a percentile describes the share of that group scoring below a result. Neither can be inferred responsibly from an item count alone.

Next ask what evidence supports the interpretation. A professional intelligence assessment is administered and interpreted within a defined framework. Test IQ Free's browser assessment is a fixed-form educational assessment with 50 original questions across pattern, quantitative, verbal, spatial, and logical reasoning. Its raw and domain results can help a reader review reasoning performance and worked answers. They should not be treated as a clinical IQ score or as equivalent to a professional instrument.

For any retest, record the conditions: sleep, time pressure, interruptions, device, prior exposure, and whether help or notes were used. This does not eliminate uncertainty, but it makes the comparison more honest.

What practice is still good for

The evidence does not make practice pointless. It makes the target of practice more specific. If you want to improve at holding several steps in mind, practice that operation and review the strategy that helped. If you want stronger performance on reasoning questions, work directly on the operations those questions require: tracking a changing relation, comparing quantities, translating words into conditions, or checking whether a rule fits every part of a problem.

A useful session includes a fresh example after instruction. State the rule in plain language, solve without looking at the answer, and explain why the other options fail. Then change the surface details. This tests whether the operation survived a small change in content. It is evidence of near transfer at most, not proof of a general-intelligence gain.

For learning facts or procedures, retrieval practice and spacing can be sensible study methods. The Institute of Education Sciences practice guide summarizes evidence that spacing study episodes improves later recall when total study time is held constant, and it treats delayed review as a learning recommendation. That is a claim about learning and recall, not a promise that any method raises general intelligence.

Claims to reject when choosing a program

Be cautious when a program treats a score on its own exercises as a general IQ increase. Ask which untrained outcomes were measured, whether there was an active control, whether the assessment used multiple measures, and whether improvement lasted beyond the immediate retest. A before-and-after graph without those details cannot establish transfer.

Also reject false precision. No study can tell an individual that completing a specified number of sessions will add a specified number of IQ points. Even a genuine group difference is an average under defined conditions, not a personal forecast. Claims about diagnosis, gifted identification, employment suitability, or fixed potential require a different level of assessment and should not be drawn from a browser exercise.

The strongest exception is a clearly defined intervention that improves a clearly defined outcome for a clearly defined group. That possibility is why research continues. It still requires evidence for that intervention and use. It does not rescue a broad claim about every working-memory game.

The responsible next step

If your goal is curiosity or low-stakes practice, take the free 50-question Test IQ Free assessment, keep the conditions consistent, and read every worked answer. Use the result as a record of performance on that fixed set of reasoning tasks. Review one error pattern, then choose fresh problems that exercise the same operation with different surface details.

The optional $9 practice report can add structured domain interpretation, an error-pattern review, fresh practice, and a 14-day plan. Its value is organization and explanation. It should not be purchased as a promise of a higher IQ or a prediction about your future.

If a result will affect a consequential education, employment, disability, or clinical decision, use a qualified professional assessment and the other relevant evidence. A browser quiz and a memory-training score are not substitutes for that process.

Questions readers ask