What you remember is never the items themselves. It's the "because… so" between them.
A bit of gossip you heard over dinner three years ago, you can still tell start to finish. The five key numbers from last week's meeting are all gone. Almost everyone knows this feeling, and the internet has a slogan for it: stories are 22 times more memorable than facts.
The number is usually credited to the psychologist Jerome Bruner, but nobody can find where he said it. The real picture is that the effect exists but isn't that dramatic. The more interesting question is why it exists. The answer isn't really about "stories" as such. The brain was never built to store things as list items. It stores events that hang together.
In 1969, Gordon Bower and Michael Clark at Stanford ran a blunt experiment. People had to learn 12 lists of 10 unrelated nouns. Half used whatever method they liked; the other half were told to weave each list into a little story. After all 12 lists, the researchers gave only the first word of each list as a cue and asked for the rest.
The story group recalled 93% of the words on average; the control group recalled 13%, roughly a sevenfold gap. Both groups spent about the same time studying the same words. The only difference was whether the words had been linked to each other.
The reason is that recall runs on cues. Whether a memory comes back depends on whether something else pulls it out. Isolated facts have no links to each other: forget one, and nothing reminds you. In a story, every event is the result of the last one and the cause of the next, so one memory leads you to the next. The psychologist Tom Trabasso analysed large numbers of story recalls and found that the more causal links an event has in a story, the more likely it is to be remembered, and to be judged important.
The key part of the brain here is the hippocampus, one on each side, roughly seahorse-shaped, the structure that "packages" a stretch of experience. Hippocampus & entorhinal cortex It's good at binding several things into one set of relations; later, give it one corner and it fills in the rest (the pattern completion from Topic 4). A story arrives with the bindings already tied.
A 2021 meta-analysis (pooling many studies to estimate the average effect) covered more than seventy samples and over 33,000 participants. Its conclusion: narrative texts really are easier to understand and remember than expository ones. The finding is robust, but the edge is "clear", not "22 times".
Put another way: what can be predicted can be compressed. Compression means storing just the "rule" plus the "exceptions". A random string of nouns has no rule, so each has to be stored on its own. In a causally coherent story each step can be roughly guessed from the one before, and all you really need to keep are the surprising turns. Large language models (AI that learns by predicting the next word) have a measure called perplexity, which captures how "surprised" the model is by the next word: structured text has low perplexity, a random word list has high perplexity. Information theory has a precise result here: a good predictor can be turned directly into a good compressor; mathematically they're the same thing. The brain is not literally running a compression algorithm, but if it is a prediction machine (Topic 1), then "stories are easy, lists are hard" are two sides of one fact.
You've surely had this happen: in the kitchen you remember you need your keys from the bedroom, you walk into the bedroom, and you stand there with no idea why you came. The psychologist Gabriel Radvansky tested exactly this: people who walked through a doorway forgot more than people who walked the same distance within one room. Doorways aren't magic. A doorway is an event boundary.
Experience is continuous, but the brain doesn't store it second by second. It cuts it into "events". The cuts tend to fall where predictions fail: a new place, a new goal, a new person. At each cut the segment that just ended gets packaged and the mental "workbench" is cleared for the next one.
Brain imaging has caught these cuts happening. In 2018 Aya Ben-Yakov and Richard Henson analysed scans from more than two hundred people watching the same short film: wherever independent viewers agreed that "one part ends and the next begins", the hippocampus produced a burst of activity, and the more people agreed on a boundary, the bigger the burst. They nicknamed the hippocampus the film editor. In 2017 Christopher Baldassano and colleagues had people watch, then retell, an episode of Sherlock. Activity patterns in high-level brain regions "shifted gear" as a whole at boundaries, and the stronger the hippocampal response at a boundary, the more completely that segment was recalled later.
Stories naturally move scene by scene, so the save points come pre-marked. A row of parallel facts has no clear boundaries: either it blurs into one lump, or each item is its own lonely fragment with nothing before or after to hang on.
This idea has already been borrowed for large language models. A model can only "see" so much text at once, its context window, and anything beyond that has to be stored some other way. EM-LLM (2024) lets the model decide where things are "surprising": wherever its prediction fails, it makes a cut, dividing long text into "episodes". When needed, it first finds relevant episodes by similarity, then pulls in the ones right next to them in time, much as remembering one thing tends to bring back what happened just before and after. The authors report that the model's boundaries correlate strongly with the event boundaries humans mark. That shows "cut where surprised" works well as engineering; it can't be turned around to prove the brain does the same.
Different levels of the brain "wait" for different stretches of time. In 2011 Yulia Lerner and colleagues in Uri Hasson's lab took a story told by a real person and chopped it up: scrambled by word, by sentence, by paragraph, plus the intact original. They played each version to listeners and checked whether each brain region's response was reliable across repeated plays.
The result looks like a staircase. Auditory cortex cares only about the sound itself and responds however you scramble it. Language areas need at least whole sentences. The highest parietal and frontal regions respond reliably only when whole paragraphs play in their original order. The researchers called this the temporal receptive window: the higher the region, the longer the stretch it integrates. Many of these high-level regions belong to the default mode network, the set of regions that is especially active when you daydream, reminisce or imagine. Default mode network In other words, the top tier of the brain is built for "plots" lasting minutes.
More striking still, plots look alike in different people's brains. In 2017 Janice Chen and colleagues had 17 people watch a 50-minute episode of Sherlock, then retell it aloud with no prompts. For the same scene, default-mode activity patterns during recall were similar across different people, similar enough to tell scenes apart across individuals. In 2010 Greg Stephens and colleagues went further: one person told an unrehearsed personal story in the scanner, and the recording was played to others. Listeners' brain activity followed the speaker's, usually a second or two behind, with some regions slightly ahead, as if anticipating. The more tightly a listener was coupled, the better they scored on a comprehension test afterwards.
To be clear about the limits: the speaker–listener study had one speaker and about a dozen listeners, a small sample, and "coupling" is a statistical correlation, not telepathy.
Almost every story hangs on the same skeleton: someone wants something, hits an obstacle, tries something, and it ends somehow. A skeleton like this is a schema, a ready-made frame in your head. New information that can hook onto an existing frame is learned remarkably fast. In 2007 Dorothy Tse and colleagues spent weeks teaching rats the overall layout of "which flavour goes with which place" in an arena; after that, rats learned new pairings in a single trial, and within two days no longer needed the hippocampus for them, which normally takes weeks. The prefrontal cortex plays a key role in calling up frames like these. Prefrontal cortex
The cost comes from the same place. In 1932 the British psychologist Frederic Bartlett had Cambridge students read a Native American folk tale, The War of the Ghosts, and retell it after a delay, then again after another. The story got shorter with each retelling; the parts readers found odd were quietly dropped or "made sensible", canoes became boats, the supernatural faded. What survives is the gist, with the details rewritten to fit what you expect, the "recall is reconstruction" of Topic 4.
Stories also raise the "save priority". Suspense opens an information gap, and curiosity engages the midbrain dopamine system, boosting how well the hippocampus encodes. In 2014 Matthias Gruber and colleagues found that while people waited for an answer they were very curious about, they even remembered unrelated faces shown in the meantime better. Reward circuit Emotional arousal, via the amygdala and noradrenaline, strengthens the hippocampus's consolidation of that stretch (James McGaugh's long line of work). Amygdala
So stories cut both ways. One concrete, named case often moves people more than a set of statistics, and is harder to forget. This is the identifiable victim effect: a story about Rokia, a young girl in Mali, raised more in donations than statistics describing millions of hungry people, and adding the statistics next to her story actually lowered giving. A 2024 replication suggests the effect may be smaller than first reported. Memorable doesn't mean true, or representative.
To make facts stick, you can borrow the machinery: string facts into a causal chain, asking "why did this happen, and what did it lead to?"; pose the question before giving the answer, to open a gap first; give abstract knowledge a person and a scene, such as where its discoverer got stuck. For how to practise and review, see Topic 43.