Your working memory can only hold about four to seven meaningful "chunks" at once. Group information well, and people can handle far more than they could item by item.
In 1956, psychologist George A. Miller published a paper that became one of the most cited in cognitive psychology: "The Magical Number Seven, Plus or Minus Two." His finding, in plain terms: people can hold roughly seven items in short-term memory at once, give or take two. Try to hand someone ten unrelated facts in a row and most of them will fall out before they can use any of them.
The twist that makes this useful isn't the number itself. It's what Miller called chunking. Memory isn't limited by how many individual facts you present. It's limited by how many meaningful groups you present. Group ten digits into three or four chunks, and suddenly all ten fit. This is why phone numbers are written as 555-0183 and not 5550183, and why a good presentation has three main points instead of eleven.
This model is about the capacity limit of short-term memory and how chunking works around it. It is easy to confuse with two neighboring models, so the Compass routes you to the right one.
Click through the sequence below. Each step adds one layer, and the third one includes a live memory check so you can feel the effect rather than just read about it.
Miller reviewed experiments on two different tasks: judging a stimulus along one dimension (like pitch) and recalling a list of items in order. Both tasks topped out around the same number: roughly seven, plus or minus two. He thought the overlap was interesting enough to name, but he was explicit that the number "seven" was not some magic constant of the mind. It was a rough empirical pattern, not a law of nature in the strict sense.
Here's the part people miss. Miller found that memory span barely changes whether each item is a single bit (yes/no) or something richer, like a whole word. What matters is the number of chunks, where a chunk is the largest meaningful unit the person already recognizes. A chess grandmaster sees a cluster of twelve pieces as "a known opening" — one chunk. A beginner sees twelve separate pieces. Same board, very different memory loads.
That means you can expand what someone can hold in mind without asking their brain to do anything new: you just repackage the information into fewer, bigger units.
You'll see a string of 10 digits for 4 seconds. Half the time it's shown as one unbroken string; half the time it's grouped into chunks. Then it disappears and you type what you remember. Run it a few times with each mode and compare how it feels.
Later researchers refined Miller's number. Nelson Cowan's 2001 review argued that when you strip out chunking and rehearsal tricks, raw working memory capacity is closer to four chunks for most adults. Other work found the span varies by material: around seven for digits, closer to five or six for unrelated words, and less for complex items. None of this overturns Miller's core point. If anything it strengthens it: the ceiling is tight, often closer to four than seven, which makes chunking even more valuable in practice, not less.
The problem: An AI analytics product asks new users to configure 16 settings on one screen before they can see their first dashboard: data source, refresh rate, timezone, currency, default chart type, team roles, 4 notification toggles, 3 integration keys, export format, and 2 permission flags. Completion rate on this screen is 34%. Support tickets mention "didn't know where to start" and "forgot to come back and finish."
Projected effect, based on how comparable onboarding redesigns typically move: completion on the first-dashboard step improves meaningfully once the real decision count per screen drops under roughly 5, even though the total information collected across the whole flow doesn't shrink at all.
Paste a list you're currently asking someone to remember or choose from (a menu, a form, a set of instructions, talking points). Get back a tailored prompt you can hand to yourself or to an AI assistant to chunk it properly.
Below is a portable, model-agnostic description of Miller's Law written so you (or an AI assistant) can apply it consistently to future work: menus, forms, slide decks, documentation, checklists, anything with a list someone has to hold in mind.
Describe what you're designing, writing, or organizing. This generates a ready-to-use prompt that applies Miller's Law directly to your situation.