In 1973 Lionel Standing showed people 10,000 photographs, one at a time, briefly. Later he tested recognition by showing pairs and asking which one they had seen.
They got about 83 percent right. Roughly 8,300 images, after a single short exposure each. A control condition using words scored around 14 percent.
That gap has a name. Pictures are remembered better than words, and the effect is one of the largest and most reliable in memory research.
Why it happens
The standard explanation is Allan Paivio's dual coding theory, which is also the basis of dual coding as a study technique.
A word arrives through one channel. You encode it verbally, and retrieval has one route in.
A picture arrives through two. It is encoded as an image, and it is also almost automatically labelled ("a dog", "a red door"). Two representations, two retrieval paths, and either one can recover the memory when the other fails.
Concrete words sit in between, which is the tell that makes the theory credible. "Elephant" is remembered better than "justice", because "elephant" spontaneously generates a picture and "justice" does not. The advantage tracks imageability, not word length or frequency.
Later work sharpened the surprise. Brady and colleagues at MIT showed participants 2,500 objects and then tested them not on category but on detail, whether it was this teapot or a slightly different teapot. Performance stayed high. Visual memory does not just store gist. It stores a startling amount of specific detail.
What the effect does not promise
Three limits, because the pop version of this oversells badly.
Recognition is not recall. Standing's participants picked which of two images they had seen. That is a much easier operation than producing something from nothing. You will recognise a diagram you drew last month and still be unable to reproduce it on a blank page. Notes are usually asked to support recall, so do not assume the 83 percent transfers.
Distinctiveness is doing some of the work. Ten thousand varied photographs are ten thousand different things. Ten thousand near-identical diagrams would not behave that way, for the same reason described in the von Restorff effect: what stands out is what survives.
A picture of a word is still a word. Screenshotting a paragraph does not convert it into a visual memory. The effect needs an actual image carrying actual meaning, not text in a different container. This is worth stating because "take a photo of the slide" is the most common way people believe they are using this and are not. See the collector's fallacy.
Using it in notes
The practical form is not "draw more". It is "make the structure visible".
Draw the relationship, not the object. An arrow between two things you already understood separately is worth more than a careful sketch of one of them. This is most of why concept mapping and mind mapping work at all.
Sketch badly and quickly. The memory benefit comes from generating the image, not from its quality. A study by Wammes and colleagues found that drawing a term beat writing it repeatedly, and that artistic ability made almost no difference. Ugly drawings work. See sketchnoting for the disciplined version and the generation effect for why generating beats receiving.
Use spatial position as information. Where something sits on the page is itself a visual cue. This is the entire mechanism of the memory palace, and a weaker version of it operates on any page with deliberate layout, which is part of why the Cornell method is stickier than a continuous transcript.
One image per idea, not per page. A single diagram that carries the argument beats six decorative ones. Decorative images actively cost you, because they add cognitive load without adding a retrieval path.
Where this argues against pure text
Most digital note systems are text systems with images bolted on. Files get attached, screenshots pile up, and none of it is searchable or connected to anything.
That is the real trade-off with plain text notes and Markdown, which are excellent for durability and portability and mediocre for thinking visually. Handwriting apps invert it: good for diagrams, historically bad for search, though handwriting recognition has narrowed that gap considerably. See digital vs paper notes and the best handwriting notes app.
There is no clean answer. Notice which one your system makes hard, because that is the one you will stop doing.
The short version
Pictures are remembered because they are encoded twice. The effect is large, well replicated, and older than most of the advice built on it.
It does not mean illustrate everything. It means that when an idea has a shape, drawing the shape is not decoration. It is a second copy.
More in why writing helps memory, how memory works, and mind map vs outline.