What actually changes about memory as you age

People who take a test like number memory and score below what they expect often reach for age as the explanation. It is worth knowing which parts of that is supported.

Two kinds of ability, moving in different directions

Psychologists separate what they call fluid and crystallised abilities, and the distinction does most of the work here.

Fluid covers processing speed, working memory and reasoning about novel problems — the machinery. It peaks in the twenties and declines gradually and steadily thereafter.

Crystallised covers vocabulary, general knowledge and accumulated expertise — the contents. It holds up and often keeps improving into the sixties and beyond.

So the honest answer to “does memory get worse” is that one part does and one part does not, and which one dominates depends entirely on the task. A crossword rewards crystallised ability. A digit span test measures the fluid part almost purely, which is why it is one of the less flattering things a person over fifty can do.

Processing speed explains a surprising amount of the rest

One of the more striking findings in cognitive ageing is how much of the age-related decline across many tasks statistically disappears once processing speed is accounted for.

The interpretation is debated, but the practical reading is straightforward: a lot of what looks like “worse memory” is the same memory operating with less time. If encoding is slower and the material arrives at the same rate, less gets encoded — and the failure shows up at recall, where it looks like forgetting.

This is why age effects are much larger on timed tests than untimed ones, and why the number memory test — which gives you roughly a second a digit whoever you are — is harsher than a version that let you take your time.

What holds up better than expected

Vocabulary, reliably. It often peaks in the sixties or later.

Procedural skill. Typing, driving, an instrument. Well-practised motor sequences are resilient, which is why typing speed declines much less than reaction time does.

Recognition, relative to recall. Producing a name is harder than knowing it when you hear it, and the gap widens with age. That is why the tip-of-the-tongue experience gets more common while the underlying knowledge is intact.

Emotional regulation and, on some measures, judgement. Not memory, but worth saying, since the popular picture of ageing is one of uniform decline and the evidence does not support that.

What declines, and roughly when

Processing speed begins its slow decline earliest — measurable from the late twenties, which is a long way from the ages people worry about.

Working memory — holding and manipulating at once — declines gradually from around the same point.

Recall of new episodic detail. Where you left something, what someone just said. This is the one people notice and worry about, and it is the most normal.

Reaction time, gradually. Across a decade the effect is smaller than the effect of one bad night’s sleep, which is worth keeping in proportion when a browser test looks disappointing.

Ordinary change against something to ask about

This distinction is the most useful thing on the page, and it is not subtle once stated.

Ordinary: forgetting where you put your keys; needing longer to recall a name that then arrives; losing your thread when interrupted; being worse at all of this when tired or stressed.

Worth a GP appointment: forgetting how to use familiar objects; getting lost somewhere well known; struggling to follow a conversation you would previously have followed; other people noticing before you do; a change that is getting steadily worse over months rather than fluctuating.

The pattern that reassures is fluctuation — good days and bad days, worse when tired. The pattern that warrants a conversation is steady progression, and someone else noticing it first.

This site cannot help you tell the difference. These are consumer tests on uncalibrated hardware, and a poor score means you had a poor run. If the second list describes you or someone you know, that is a conversation for a doctor, and an early one is better than a late one.

Cross-sectional against longitudinal, and why it matters

Much of what is popularly said about age and memory comes from cross-sectional studies — testing people of different ages at one moment and comparing the groups.

That design has a well-known problem: it compares people born decades apart, who differ in education, nutrition, health care and familiarity with the format being tested. Someone of eighty taking a computerised test in 2026 grew up without computers, and some of the gap attributed to their age is attributable to that.

Longitudinal studies follow the same people over years and generally find gentler decline, starting later. They have their own problem — the people who stay in the study are the healthier ones — so the truth sits between the two designs.

The practical reading: the steep decline curves people have seen are probably steeper than reality, and the honest summary is gradual change in some abilities, stability or improvement in others, with enormous variation between individuals.

Stereotype threat is a real and measurable confound

One of the more uncomfortable findings in this literature: telling older adults that a task measures memory decline makes them perform worse on it.

The effect has been demonstrated repeatedly. The same people, the same test, described either as a memory assessment or as a neutral puzzle, produce different scores. The size of the effect is disputed and the direction is not.

That matters directly for a site like this one. If you sit down to a memory test already braced for a bad result because of your age, some of the bad result is the bracing. It is one more reason the only comparison worth making is you against your own history, and one more reason not to read a single score as a verdict.

Why practice helps more than it should

Most of what people gain from repeated memory testing is technique rather than capacity, and that is genuinely good news for anyone worried about age.

Chunking, rehearsal, building images for words — these are learnable at any age, and they are where most of the variance between people on a digit span test actually lives. Someone of seventy with a deliberate grouping strategy will out-score someone of twenty-five without one.

The underlying machinery is not what the technique changes. What it changes is how much each slot has to hold, which is the same lever chunking pulls at any age.

What the evidence supports doing

Nothing here is a memory technique, which is the honest part.

Sleep. Consolidation happens during it, and sleep quality declines with age in ways that plausibly account for some of the memory change attributed to age itself.

Aerobic exercise. The best-supported intervention for cognitive function in older adults. Modest, real, repeatedly found.

Treating hearing loss. Appears consistently in the dementia-risk literature. Straining to hear consumes the resources that would otherwise go on understanding and remembering.

Staying socially and mentally engaged with things that are genuinely difficult — which is a weaker finding than the enthusiasm around it, but has the advantage that the activity is worth doing regardless of whether it transfers.

Not brain-training games. Including these ones. They make you better at themselves, which is a narrower claim than the category markets. The value of the tests here is measurement — a number you can watch over time — not training.