Miller's law

Miller's law refers to George Miller's 1956 paper The Magical Number Seven, Plus or Minus Two, which observed that people's span of immediate memory and their ability to make absolute judgments both seemed limited to around seven items. Miller also described chunking: grouping items into larger meaningful units so that more information fits in the same limited span. Later research puts the limit lower. Nelson Cowan's 2001 review argues for a central capacity limit "averaging about four chunks".

Why it matters

Short-term memory is small and fragile. When an interface makes people hold information in their heads, such as a code from one screen to type on another, or options from step 1 while deciding in step 3, they forget and make mistakes. Designing for this limit means showing information where it is needed and chunking what people must remember.

How to apply it

• Do chunk long strings. Phone numbers, card numbers and verification codes are easier to read and check in groups.
• Do break long content into short sections with clear headings, and long forms into labeled groups.
• Do keep relevant information visible instead of asking people to remember it, for example a summary of choices on the confirmation step.
• Don't make users carry data between screens. Prefill, copy buttons and "paste from clipboard" beat memorization.
• Don't treat seven as a design rule. Laws of UX warns against using the magical number seven to justify unnecessary design limits.

Common mistakes

The most common misuse is limiting menus, navigation bars or tab sets to seven items "because of Miller's law". The paper is about how many items people can hold in memory, not how many they can choose from on a visible list. NN/g puts it directly: "It's a common misconception that limited short-term memory implies that menus should be similarly limited to 7 items." A menu is on screen, so people recognize items rather than recall them. Menu length should follow from information architecture and scanning, not from a memory limit.

Other mistakes:

• Treating 7±2 as a precise, universal number. Miller himself ended by suspecting the recurrence of seven was "only a pernicious, Pythagorean coincidence", and capacity varies with prior knowledge and context.
• Chunking arbitrarily. Chunks help when they are meaningful to the user, such as grouping settings by task, not by internal system structure.

In AI products

Long AI answers put this limit under strain. A chat assistant that returns twelve steps in one dense paragraph, or a meeting-notes summariser that lists twenty action items without grouping, asks people to hold too much at once. Structure output into a few labeled chunks, lead with the answer, and put details behind expandable sections.

Multi-turn conversations add another load: people have to remember what the AI said several messages ago. Restate key facts when they matter again, and let users pin or summarize earlier results instead of scrolling back.