Miller's Law UI Design Examples: A Starter Kit of 9 Chunking Patterns You Can Ship This Sprint

    November 18, 2026
    9 min read
    By Emil Thornovik
    Share this article

    Does your navigation really need to stop at seven items? That question comes up in design reviews more than almost any other "law" of UX, and the answer is usually no. The paper behind it says something narrower and far more useful: people can only hold a handful of unrelated items in mind at once, so anything a user has to remember, compare or retype should arrive in chunks.

    That reframing turns Miller's law from a menu rule into a practical toolkit. Below is the starter kit: what the research measured, nine interface patterns that apply it correctly, a comparison table you can paste into a design doc, and a worked audit of a real screen.

    What Miller measured, and why the menu rule misreads it

    In 1956 George A. Miller published "The Magical Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information" in Psychological Review. He reviewed experiments on absolute judgement and on the span of immediate memory, such as how many digits a person can repeat back after hearing them once. Across tasks, performance tended to collapse somewhere around 5 to 9 items. Miller's own tone was half playful; he noted the recurring number while warning against reading too much into it.

    His more durable idea was chunking. A string like 1 9 5 6 2 0 0 1 is eight digits, but a reader who sees two years, 1956 and 2001, holds only two chunks. Memory capacity is measured in meaningful units, and design can change how many units a piece of information takes. Later work pushed the estimate lower: Nelson Cowan's 2001 paper "The magical number 4 in short-term memory" argued that when rehearsal and chunking are controlled for, capacity is closer to 3 to 5 chunks.

    Now the misreading. A navigation bar is visible. Users scan and recognise its items; they do not memorise them. Recognition draws on very different limits from recall, so a menu with 9 clear labels can outperform one with 6 vague ones. Jon Yablonski's book "Laws of UX" makes the same point: the law is about short term memory, not about how many things a screen can show.

    So the test for whether Miller's law applies is simple. Does the user need to hold this information in their head while doing something else, such as switching screens, typing it elsewhere, comparing it to another value or reading it aloud? If yes, chunk it and reduce it. If the information stays on screen the whole time, look to Hick's law, visual hierarchy and labelling instead.

    Nine Miller's law UI design examples for the starter kit

    Each pattern below handles a moment where users hold something in memory. The order runs from tiny strings to whole flows.

    1. Payment card numbers in groups of four. A 16 digit Visa or Mastercard number is printed as 4 groups of 4, and the input field should mirror that so users can check the screen against the card in four glances. Amex uses 15 digits grouped 4, 6 and 5; a good field detects the brand from the first digits and switches the mask. Storing the value without spaces keeps validation and screen readers clean.

    2. One time passcodes split 3 and 3. A 6 digit OTP is read from an SMS or authenticator and typed into another app. Showing it as 482 917 in the message and offering 2 groups of 3 boxes on the input side cuts the load to 2 chunks. Support pasting the whole code into the first box, since splitting fields breaks paste in naive implementations.

    3. IBANs and long account identifiers in fours. The paper format of an IBAN groups characters in blocks of 4, so a Portuguese IBAN of 25 characters reads as 7 blocks instead of one wall. Copy buttons help, but many users still compare the number against a bank letter, and blocks make a mismatch visible.

    4. Phone numbers formatted on the fly. A US number shown as 415 555 0132 follows a 3, 3, 4 rhythm users already know. Formatting while typing, by country, lets people spot a missing digit before submitting. Avoid forcing one country's pattern on international numbers.

    5. Booking references with a restricted alphabet. Airline record locators are typically 6 characters, short enough to read over the phone. If you generate your own references, drop look alike characters such as 0, O, 1, I and l, and group longer codes in 4s. Fewer ambiguous symbols reduce errors more than clever grouping.

    6. Multi step forms with 3 to 5 stages. A checkout that asks for 18 fields on one page forces users to track where they are. Grouping fields into Contact, Delivery and Payment, with a stepper that names each stage, gives a running mental model of 3 chunks. Five named steps is a sensible upper bound before the stepper itself becomes something to remember.

    7. Pricing tables with 3 or 4 plans. Comparison is a working memory task: users hold plan A's price and limits while reading plan B. Most SaaS pricing pages settle on 3 or 4 tiers, and the rows that matter should be grouped, such as usage limits, collaboration and support, rather than listed as 25 flat feature lines.

    8. Dashboards with a short headline row. A KPI row of 4 numbers can be compared at a glance; a row of 11 tiles cannot, because the user has to remember earlier tiles while reading later ones. Push secondary metrics into a second, collapsible group.

    9. Settings pages grouped by intent. A flat list of 30 toggles forces readers to hold their position. Grouping into 5 or 6 labelled sections, each with 4 to 7 items, turns scanning into a two level search: pick the group, then the setting.

    The comparison table below summarises the formats so you can drop it into a spec.

    Information Raw format Chunked format Chunks held
    Card number, 16 digits 4111111111111111 4111 1111 1111 1111 4
    OTP, 6 digits 482917 482 917 2
    Portuguese IBAN, 25 chars one unbroken string 7 blocks of up to 4 7
    US phone, 10 digits 4155550132 415 555 0132 3
    Checkout, 18 fields one long page 3 named steps 3
    Pricing, 25 features flat list 3 groups per plan 3

    A caveat on the table: the "chunks held" column describes the best case, a user who parses each group as a unit. Unfamiliar formats chunk less well, which is why mirroring a printed source, such as the card or the bank letter, matters more than picking a mathematically optimal group size.

    A worked audit: applying the starter kit to one screen

    Here is an illustrative audit of a bank transfer screen, the kind where Miller's law problems hide in plain sight. The current design asks users to copy a 25 character IBAN from an email, a 12 digit payment reference from an invoice, and an amount, then confirm on a second screen that shows all three in plain text.

    Count the memory load first. The IBAN as one string is 25 items for most people. The reference is 12. On the confirmation screen, the user has to compare both against the source documents, which means holding a segment in memory while switching windows. If a user can reliably carry about 4 chunks, the unchunked IBAN needs at least 7 trips between windows to verify, and the reference another 3. That is roughly 10 window switches for a single transfer, each a chance to lose their place.

    After chunking, the IBAN shows as 7 blocks and the reference as 3 blocks of 4. Comparison now happens block by block: glance, hold one block, check. The user still makes about 10 comparisons, but each holds 1 chunk instead of 4 loose characters, so the error rate per comparison drops sharply. A team that tested this kind of change would track two numbers: the share of transfers returned for a wrong beneficiary and the median time on the confirmation step.

    The procedure to run on any screen:

    1. List every value the user must remember, retype or compare across a context switch.
    2. Count items per value in its current form, treating each character or field as one item.
    3. Mark any value over 4 items that crosses a context switch: those are your candidates.
    4. Choose a grouping that mirrors the source the user reads from, then a fallback of 3 or 4.
    5. Apply formatting visually, keep stored values clean, and check screen reader output.
    6. Remove ambiguous characters from any code your system generates.
    7. Measure one error metric and one time metric before and after the change.

    The audit usually surfaces 2 or 3 fixes per flow, and they tend to be cheap: an input mask, a stepper, a regrouped table. That is the appeal of Miller's law used properly. It points at specific moments of memory strain instead of at an arbitrary item count.

    Questions designers ask about Miller's law

    Is Miller's law the same as Hick's law? No. Hick's law describes how decision time grows with the number of choices, roughly logarithmically. Miller's law is about how much a person can hold in short term memory at once. A long menu can be a Hick's law problem without being a Miller's law problem, because the options stay visible.

    Does chunking help screen reader users? It can hurt them if done with literal spaces inside an input value, since some screen readers announce each group separately or read spaces aloud. Format visually with CSS or a mask, keep the stored value unspaced, and test with VoiceOver and NVDA before shipping.

    Should I cite Miller's law in a design review? Cite the mechanism instead of the law. Saying the user must remember a 12 character code across two screens is concrete and testable; saying it violates Miller's law invites a debate about the number seven.

    What chunk size works for codes people read aloud to support agents? Groups of 3 or 4 characters work well over the phone, and removing ambiguous characters such as 0 and O or 1 and l cuts misreadings more than any grouping does. Many teams pair 4 character groups with a restricted alphabet of about 30 symbols.

    Pick the one flow in your product where users copy a code or number from somewhere else, run steps 1 to 3 of the audit on it tomorrow, and put the count of items crossing a context switch in your next design review.

    Related Articles