Made with 🧠 and 🫀 by Youssef Bouksim

Back to library
🧩

Chunking

Short-term memory handles chunks (meaningful units) better than raw individual items. Organizing information into meaningful groups — visually and conceptually — reduces cognitive load and improves both comprehension and memory.

5 min readUI · UX

In 1956, psychologist George Miller published a paper called “The Magical Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information.” His core finding was that human working memory can hold approximately 7 (±2) items at once — but critically, the definition of “item” is elastic. A digit is an item. A word is an item. A meaningful phrase is an item. A well-labelled group of related concepts is an item. By grouping individual pieces of information into meaningful units — chunks — people can effectively multiply the amount of information they can hold in working memory.

Miller's later research and subsequent work by Cowan (2001) refined the working memory capacity estimate to approximately four chunks, not seven. But the central insight held: the unit of working memory is the chunk, and the size of the chunk is determined by meaning rather than by the count of underlying items. A phone number formatted as 555-867-5309 is three chunks — area code, exchange, number — and is recalled as easily as three random digits. The same ten digits written as 5558675309 with no grouping are ten separate items — far exceeding working memory's capacity and requiring the reader to impose their own structure to remember them.

For product designers, chunking determines cognitive load across every element of an interface. A navigation with twelve ungrouped items requires the user to hold twelve separate items while scanning for their destination. The same twelve items organised into four labelled groups of three require holding only four chunks — the group names — while scanning, then drilling into the relevant group. A settings form with fifteen sequential fields loads fifteen items into working memory simultaneously. The same form divided into four named sections loads one section at a time. The information content is identical. The cognitive cost is not.

✦ Three things to know
✓
The limit is on chunks, not on items — grouping reduces cognitive load directly. Cowan's (2001) revision of Miller established that working memory holds approximately four chunks. Every label, category, or visual grouping that turns multiple items into a single unit reduces the number of chunks the user must hold simultaneously. A navigation group labelled “Account” that contains five items costs one chunk, not five. The label is the chunk; the items inside it are only loaded when the user selects that group.
✓
Meaning creates chunks — arbitrary groupings do not reduce load. The cognitive benefit of chunking depends on the grouping being semantically meaningful. Items grouped together because they are used together, belong to the same category, or share a common purpose form a genuine chunk — the label becomes a retrieval cue for the group. Items grouped arbitrarily — because there are seven of them and someone decided they should be in a box — do not. The label must describe a real relationship between the items for chunking to work.
✓
Chunking applies equally to sequence and to space. Forms chunked into sections reduce the items in working memory at any given step. Reading chunked into numbered sections reduces the mental model the reader must maintain. Data formatted in groups (credit card numbers in sets of four, phone numbers in country/area/number groups) reduces the working memory load of encoding and recalling the digits. The principle is not limited to navigation and forms — it applies to any structured information a user must scan, process, or remember.
“By organising the stimulus input simultaneously into several dimensions and successively into a sequence of chunks, we manage to break the informational bottleneck.”
— George Miller, Psychological Review, 1956

Navigation — flat list vs grouped chunks

A flat navigation list loads every item into working memory simultaneously during scanning. The user must compare each item against their goal, one at a time, until they locate it. Grouped navigation loads only the group labels during the initial scan — four chunks instead of twelve items — then narrows into the relevant group. The user's eyes need to travel less, hold less, and make fewer comparisons.

Both sidebars below contain exactly the same twelve items. Try to find “Billing” in each. The flat list requires scanning all twelve items in sequence. The grouped version requires scanning four group labels, then the items within the relevant group.

Flat list — 12 items in working memory at once
app.yourapp.com
Dashboard
Projects
Tasks
Timeline
Messages
Notifications
Team
Profile
Settings
Billing
Integrations
Help
Find “Billing” in the sidebar
12 items in a flat list. Your eyes scan from top to bottom, comparing each item against the goal. Working memory holds all 12 labels simultaneously during the scan.
Billing is item 10 of 12. Users must scan through nine other items first. All 12 items load into working memory during scanning.
No structure. Every item is an individual unit, so the user must impose their own grouping while scanning.
Grouped nav — 4 chunks, then 4 items
app.yourapp.com
Work
Dashboard
Projects
Tasks
Timeline
Communicate
Messages
Notifications
People
Team
Profile
Account
Settings
Billing
Integrations
Help
Find “Billing” in the sidebar
4 group labels to scan, then 4 items within “Account.” The group label is a single chunk that predicts where Billing will be found.
Billing sits in the “Account” group. Users scan 4 labels, pick “Account”, then scan 4 items. Maximum items in working memory at any point: 4.
The same twelve items, chunked into four named groups — the labels become retrieval cues rather than the user having to remember position.

The grouped navigation's benefit is not just speed — it is also error rate and cognitive load after the task. A user who navigated to Billing through grouped navigation can reconstruct the path (“it was under Account”) more reliably than one who navigated through the flat list (who must remember item position, not group membership). Chunking reduces the cognitive cost of both the navigation act and the memory of how to repeat it.


Forms — unchunked fields vs named sections

A form without sectioning loads all its fields into working memory simultaneously as the user scans to understand its scope. The user must process the entire form before they can form a plan for completing it. A form divided into named sections allows the user to process one section at a time — each section is a chunk, and the user's working memory is occupied only by the items in the current section, not by the entire form.

Baymard Institute's research on form usability found that users make 40% fewer errors on forms with clear section groupings than on equivalent flat forms, because sectioning reduces the items held in working memory during any individual field completion, which reduces the likelihood of confusion about what belongs in which field.

Flat form — all fields in working memory at once
app.yourapp.com/settings
Settings
Full name
Youssef Bouksim
Display name
ybouksim
Email
you@company.com
Phone
+212 6...
Email notifications
All activity
Push notifications
Enabled
Weekly digest
Enabled
Current password
••••••••
New password
••••••••
Two-factor auth
Disabled
Active sessions
View sessions
Save all changes
10 fields with no grouping. Profile fields sit next to notification fields sit next to security fields — the user must impose their own structure on first scan.
Every field is an individual item. The form's scope cannot be grasped without processing all ten fields simultaneously.
Sectioned form — one chunk at a time
app.yourapp.com/settings
Settings
Profile
Full name
Youssef Bouksim
Display name
ybouksim
Email
you@company.com
Phone
+212 6...
Notifications
Email
All activity
Push
Enabled
Weekly digest
Enabled
Security
Current password
••••••••
New password
••••••••
Two-factor auth
Disabled
Save changes
3 named sections: Profile, Notifications, Security. Each header is a chunk that predicts what fields are inside. The user processes one section at a time.
Section headers act as chunks and as retrieval cues — “I need to update notifications” maps directly to a location, not to a field-by-field scan.

The section headers serve a dual function. First, they reduce the items in working memory to the count within the current section. Second, they serve as retrieval cues when the user returns to update a specific setting — “I need to change my notification preferences” maps directly to the “Notifications” section without requiring a scan of all fields. The label is the chunk; the chunk enables indexed retrieval rather than sequential search.


Data formatting — raw strings vs chunked display

The most literal application of chunking in interface design is the formatting of long number strings. A 16-digit credit card number is impossible to hold in working memory as 16 individual digits — it exceeds working memory capacity by a factor of four. Formatted as four groups of four (4532 1987 6543 2193), it becomes four chunks, each within the four-chunk working memory limit. The same number. Radically different cognitive load during entry, verification, and recall.

The two fields below show the same sixteen digits. The left displays them as a raw string. The right chunks them into groups of four — the same input, with the cognitive work of grouping done by the interface rather than the user.

Raw input — 16 items, no chunks
checkout.yourapp.com
Payment details
Card number
4532198765432193
16 digits · no formatting
Expiry
MM/YY
CVV
000
Pay now
16 unbroken digits. The user must impose grouping mentally during entry and verification. Transposed digits are invisible without structure.
Sixteen separate items — four times working memory capacity. Verification against the physical card requires self-imposed chunking.
Auto-chunked — 4 groups of 4 as you type
checkout.yourapp.com
Payment details
Card number
4532 1987 6543 2193
✓ 4 groups complete
Expiry
MM/YY
CVV
000
Pay now
Spaces are inserted automatically after every 4 digits. 16 digits become 4 chunks — matching working memory capacity exactly.
The interface does the chunking work. Verification happens four digits at a time — a transposed digit becomes visible within its group.

The chunked card field does not change what the user types — it changes what they hold in working memory while typing and verifying. The raw field requires the user to mentally group sixteen digits into a verifiable sequence. The chunked field delivers four groups — four chunks — that can be compared against the physical card one group at a time. Research by Lindgaard and Dudek (2003) found that chunked card fields produced significantly fewer entry errors than unchunked fields for the same card numbers, because the structure made transposed digits visible within a group rather than hidden in an unstructured string.


Applying this to your work

Chunking is one of the most transferable findings in cognitive psychology: it applies to every interface surface where a user must scan, hold, process, or recall structured information. The practical design question is rarely whether to chunk — it is whether the chunks reflect real semantic relationships and whether the labels act as retrieval cues. When both conditions are met, working memory load drops from the item count to the chunk count, and every downstream task — scanning, completing, remembering — gets cheaper.

✓ Apply it like this
→Group navigation items into four or fewer named categories. The group labels become chunks, and users scan four labels rather than the full item count. Labels must describe a real semantic relationship, not an arbitrary grouping.
→Divide long forms into named sections. The section header is the chunk that predicts what fields are inside and serves as a retrieval cue on return visits. Users process one section's worth of fields at a time rather than the full form.
→Auto-format long number strings as the user types — card numbers, phone numbers, IBANs, sort codes. The interface does the chunking work; the user gets the cognitive benefit without the effort of imposing structure manually.
→Use visual hierarchy and whitespace to make chunk boundaries explicit. Section headings, dividers, and spacing signal where one chunk ends and another begins, reducing the user's work in identifying the structure.
✗ Common mistakes
→Flat navigation with more than five or six items. Beyond that range, scanning requires holding items that exceed working memory's chunk capacity. Users miss items, re-scan, and form less reliable mental models of the navigation.
→Forms with no section grouping. Every field becomes an individual item during the initial scan. Users cannot form a plan for the form without reading all fields, which must all be held simultaneously.
→Arbitrary groupings with vague or misleading labels. A group called 'Other' containing settings from several categories does not reduce cognitive load. The label must predict the contents for the chunk to function as a retrieval cue.
→Unformatted long strings. Card numbers, phone numbers, and IBANs presented as raw strings require the user to impose chunking mentally. Any error in the self-imposed grouping becomes an invisible source of mistakes during verification.

Miller, G. A. (1956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review, 63(2), 81–97. · Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24(1), 87–114. · Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4(1), 55–81.