Why you remember things that never happened

If you have played the verbal memory test, you will have had the experience of being certain a word appeared before, and being wrong. That is not carelessness. It is how recognition works, and there is a famous demonstration of it.

The experiment

Read this list, then look away and try to recall it:

bed, rest, awake, tired, dream, wake, snooze, blanket, doze, slumber, snore, nap, peace, yawn, drowsy

Most people recall sleep. It is not in the list.

This is the Deese-Roediger-McDermott procedure, named for the 1959 study it grew from and the 1995 paper by Henry Roediger and Kathleen McDermott that established it as a standard method. Its results are unusually consistent: the unpresented word — the “critical lure” — is recalled at rates comparable to words that actually appeared.

The uncomfortable part is not the error. It is the confidence. People report vivid, certain recollection of an item that was never there, and being told about the effect in advance reduces it only somewhat.

Why it happens

The list is built so that every word is associated with the same absent one. Reading them activates sleep repeatedly without ever presenting it, and by the end that word has been strongly activated by your own mind.

At recall, you cannot distinguish “this feels familiar because I saw it” from “this feels familiar because I generated it”. Both produce the same signal. This is the standard account, called source monitoring — memory stores the content more reliably than the origin.

What this means for the verbal test

The version on this site is a recognition task rather than a recall one, and the same mechanism produces its characteristic error.

Calling a new word “seen” is far more common than the reverse. The reason is that familiarity has multiple sources: you may have seen the word here, or it may simply be a common word you meet daily, or it may resemble one that did appear. Only the first is the question being asked, and all three feel identical.

That is why common words are harder than distinctive ones, and why the middle of a long run is disproportionately difficult — the memorable words have been used and what remains is a soup of ordinary ones.

Two things follow, both useful:

• Trust a strong “new”. Uncertainty concentrates on the “seen” side, so when you genuinely cannot place a word, “new” is the better guess.
• Do not deliberate. Long consideration tends to manufacture familiarity rather than reveal it, because thinking about the word activates it further.

Memory is reconstruction, not playback

The deeper point the DRM effect illustrates is that remembering is not retrieval of a stored recording. It is reassembly from fragments, and the gaps are filled with what is plausible.

That process is usually invisible because it usually gets the answer right — plausible reconstruction of a real event is mostly accurate. It becomes visible when the plausible answer and the true one differ, which is precisely what the DRM list is engineered to produce.

It also means memories change with each recall. Retrieving a memory reopens it, and what is stored afterwards is the reconstruction rather than the original. A memory recalled often is not better preserved for the practice; it is further from the event.

Where this matters outside a browser game

Eyewitness testimony. Confidence and accuracy correlate far more weakly than juries assume, and confidence can be raised by feedback after the fact — being told “you identified the suspect” makes a witness more certain without making them more correct. This is one of the better-evidenced findings in applied psychology and has driven real changes in how police line-ups are run.

Repeated questioning. Asking about a detail repeatedly can create it, particularly with leading phrasing. The effect is strongest in children and is why interview protocols for child witnesses are so carefully specified.

Your own recollections. Family stories, told often, drift. Confident memories of early childhood are frequently reconstructed from photographs and retellings rather than remembered.

None of that means memory is unreliable in general. It means confidence is a poor guide to accuracy, which is a narrower and more useful claim.

Sleep is where memories are sorted

One reason recollection changes rather than simply fading: consolidation happens largely during sleep, and consolidation is not passive storage.

Overnight, memories are reactivated and integrated with what you already know. That process is what makes them durable and what makes them change — a memory that has been integrated is stored in relation to everything around it, not preserved as a separate recording.

The practical consequence sits alongside everything in the sleep guide: a short night costs you both the encoding of the day before and the consolidation of it. And it is another reason recalling something vividly is not evidence it happened as remembered. Vividness tracks how well integrated a memory is, not how accurate.

Recognition and recall are different systems

The DRM effect is easiest to understand alongside a distinction the tests here make concrete.

Recall is production: generate the item from nothing. That is what number memory asks — no options are offered and you either produce the digits or you do not.

Recognition is judgement: decide whether something in front of you is familiar. That is verbal memory, and it is far easier, which is why scores there run into the dozens while digit span stops below fifteen.

The catch is that recognition’s easiness comes from leaning on a feeling rather than a retrieval, and feelings can be produced by things other than the event you are asking about. Recall fails by producing nothing; recognition fails by producing false confidence. Of the two failure modes, the second is much more dangerous outside a game.

It is also why multiple-choice exams feel easier than short-answer ones, and why they are a weaker test of whether you actually know something.

Why the effect is stronger when you are trying harder

A counter-intuitive detail, and a useful one for reading your own results.

Deep processing — thinking about what words mean rather than how they look — improves memory for material that was actually presented. It also increases false recall of the critical lure, because thinking about the meaning of fifteen sleep-related words activates sleep more strongly.

So the same effort that helps you remember correctly also helps you remember incorrectly, and the two are not separable by trying harder. That is a genuinely odd property for a memory system to have, and it is one of the clearer pieces of evidence that recollection is reconstruction rather than retrieval.

For the test here, it is the reason deliberating over a word tends to make you more likely to call it seen, not less.

Three ways familiarity gets manufactured

Useful to recognise, because each one has an everyday equivalent.

Repetition. Encountering a claim repeatedly makes it feel more true — the illusory truth effect — and it works even when the claim was labelled false the first time. Familiarity is being mistaken for accuracy.

Fluency. Anything easy to process feels more familiar: a clear font, a rhyming phrase, a simple name. The effect is small and remarkably general.

Your own imagination. Vividly picturing an event raises the chance of later remembering it as having happened. This is the mechanism behind the DRM list, and it is why leading questions are dangerous — asking someone to imagine a detail is a step towards them remembering it.

The practical upshot

For the test here: expect false alarms, do not treat them as failures of attention, and answer quickly rather than deliberating.

For everything else: when it matters, write it down at the time. A contemporaneous note beats a confident memory, and the more confident the memory the more that is worth remembering.