Zipf’s Law

A few items carry most of the use.

Rank anything by how often it is used and the counts fall away on a steep, predictable curve: the second item gets about half the traffic of the first, the third about a third. Navigation, search ranking, and feature priority all sit on that curve.

Origin

George Kingsley Zipf, a linguist at Harvard, set out the pattern in The Psycho-Biology of Language in 1935 and expanded it in Human Behavior and the Principle of Least Effort in 1949. Count how often each word appears in a body of text, rank the words by that count, and a word's frequency turns out to be roughly inversely proportional to its rank.

Zipf was not the first to notice it. Jean-Baptiste Estoup had seen it in French shorthand in 1916, and Felix Auerbach found the same shape in city populations in 1913. What Zipf added was a proposed cause, the principle of least effort, and the argument that the curve reaches well beyond language.

The Law

Rank a set of items by how often they are used and the counts fall on a steep curve. The item at rank two is used about half as often as the first, the item at rank three about a third as often. A short head carries most of the traffic and a very long tail carries the remainder.

f(r) ∝ 1 / r — the item at rank r is used about one r-th as often as the first.

Coined by
George Kingsley Zipf (1949)
Source
Human Behavior and the Principle of Least Effort
In practice
Design for the head, keep the tail reachable
When to Use
How to Use
The curve 01 / 10

The Short Head

Rank your pages, features, or queries by use and the top handful will account for most of the total. That head is where layout, speed, and polish repay the effort.

Best for
Prioritising work
Use when
Usage data exists
Avoid when
The product just launched
RelatedPareto Principle
Navigation 02 / 10

Put the Head in the Nav

A navigation bar holds a handful of destinations. Fill it from the top of the ranked list rather than from the org chart, and most people never need to open a menu.

Best for
Primary navigation
Use when
Traffic is measurable
Avoid when
Every section must rank equally
RelatedNavigation Patterns
Search 03 / 10

Ranking Follows the Curve

Query logs are Zipf-shaped too. A small set of searches repeats endlessly, so hand-tuning the top hundred results beats tuning the algorithm for everything else.

Best for
Site search
Use when
Queries are logged
Avoid when
The corpus is tiny
The tail 04 / 10

The Tail Is Long, Not Empty

Items below the head are individually rare and collectively large. Cutting them because each one looks small removes a meaningful share of total use.

Best for
Decisions about pruning
Use when
Removal is on the table
Avoid when
Upkeep is the binding constraint
RelatedProgressive Disclosure
Menus 05 / 10

Rank, Then Cut

Hick’s Law says fewer options are faster to choose from. Zipf tells you which ones to keep: the head stays visible and the tail moves behind a More.

Best for
Menus and toolbars
Use when
The list outgrew its space
Avoid when
Choices are used evenly
RelatedHick’s Law
Words 06 / 10

Write With the Common Words

The pattern began in language. A small vocabulary covers most of what people read, so copy built from frequent words is read faster and translates more cleanly.

Best for
Interface copy
Use when
The audience is broad
Avoid when
Precision needs a rare term
Caution 07 / 10

Traffic Follows Placement

The ranking you measure is partly the ranking you caused. Something buried three levels down is rare because it is buried, not because nobody wanted it.

Best for
Reading analytics
Use when
Deciding what to cut
Avoid when
The layout has never changed
Fit 08 / 10

Not Every Set Is Zipfian

The curve describes large, unconstrained sets: words, pages, queries, cities. A ten-item settings screen used by one team has no head and no tail worth the name.

Best for
Choosing an analysis
Use when
The set is large
Avoid when
The list is short and fixed
✦

Zipf in the Age of AI

Models see the head thousands of times and the tail twice, so the rare case is exactly where a generated answer runs thin without sounding like it.

✦ AI Era 09 / 10

Thin on the Tail

A model has met the common case endlessly and the rare one barely at all. Fluency does not fall off with frequency, so tail answers arrive confident and wrong.

Shift
Frequency → confidence
Use when
Answering long-tail questions
Watch for
Fluent answers to rare questions
✦ AI Era 10 / 10

Prompts Have a Head Too

What people type at an assistant is Zipf-shaped like any other query log. Handling the top hundred prompts deliberately beats tuning for an average nobody sends.

Shift
Average → ranked
Use when
Designing an assistant
Watch for
Tuning for the mean request
Further Reading

Original

Wikipedia

Reference