Zipf's Law

Law 06 / 20 Grow the Brand

In many ranked lists a few items are extremely common and a long tail of items is rare, but a long tail does not automatically mean hidden profit.

  • Empirical regularity
  • 6 min read
  • Documented study
  • Reviewed by Ali Dezhami, November 8, 2026

The dilemma

Illustrative scenario

Ines runs content for an online specialty tea shop. Its SEO agency wants to spend the whole quarter's content budget trying to rank for "green tea", by far the biggest query in their keyword tool. The ranking pages are dominated by marketplaces and large retailers, and the shop currently sits on page three.

Ines has been reading the shop's own search data instead. Beyond a handful of broad terms, there are hundreds of small, specific queries such as how long to steep sencha or which green teas are low in caffeine. Each brings only a trickle of visits, but together they add up to a meaningful share of clicks, and the few orders she can trace from them look promising.

The conflicting metric: the agency's report shows a huge prize in one term; her data shows many small ones, with conversion figures based on a sample too small to trust. Where should the budget go?

The short answer

Zipf's law describes a rank-frequency pattern: in many collections, an item's frequency is roughly inversely proportional to its rank, so the second most common item appears about half as often as the first, the third about a third as often, and so on. The linguist George Kingsley Zipf documented it for word frequencies and popularized it in Human Behavior and the Principle of Least Effort (1949).

How strong is the evidence?

Empirical regularity

A pattern repeatedly observed in market data. It describes what usually happens; the exact numbers vary.

This is an empirical regularity, first and most robustly observed in word frequencies in text. Zipf-like and other heavy-tailed patterns have since been reported in many ranked datasets, such as city sizes, website traffic and search queries. Statisticians caution, though, that data which look Zipfian on a chart often fit other heavy-tailed curves about as well, so the exact formula matters less than the practical point: a few items are very common and a very long list of items is rare.

For marketers the regularity connects to the "long tail" debate. Chris Anderson (2004) argued that when shelf space is effectively unlimited online, the many niche products together can form a market that rivals the hits. Anita Elberse (2008) examined sales data from digital media markets and reached a more cautious view: demand remained heavily concentrated in hits, and the tail, though longer, was very flat.

The decision lesson: Zipf tells you how frequencies are distributed. It says nothing about whether tail queries or tail products convert better, cost less to serve or earn more. That is a separate, empirical question for each business.

Investigating an example

Documented study

Should you invest in the long tail?

In a Harvard Business Review article, Elberse tested the long-tail argument against sales data from music and home-video markets. Her reading was that, online as well as offline, a small number of best-sellers continued to account for a large share of demand, while the growing number of niche titles each sold very little. She advised companies not to abandon investment in hits on the assumption that the tail would carry them.

Limits: the research covers media products in a particular period and market structure, and the full findings sit behind a paywall, so we describe them only in general terms. It concerns product sales, not search queries; a business still has to test whether its own long-tail queries are worth serving.

A published study or record, linked below. Its limits are stated.

Choose your move

You are in the dilemma above. Which move would you make? Open each one to see its trade-offs.

A Spend the whole budget trying to rank for the head term.

The biggest prize if it works, but competition from marketplaces is fierce and broad queries often mix researchers with buyers. A small shop may spend the quarter and stay on page two or three.

B Write a separate page for every long-tail question in the keyword export.

Captures specific intent, but many tail queries have near-zero volume, hundreds of thin pages are costly to maintain, and there is no guarantee tail visitors convert better. It can also look like low-quality content to search engines and readers.

C Group tail questions into a few strong pages by intent, and compare them with a modest head-term effort.

Balances the bets and produces evidence. It needs careful clustering work and patience, because tail traffic accumulates slowly and conversion data takes time to become reliable.

No move wins in every business. The law tells you what to check, not what to choose.

How to apply it

  1. Export your query or product data. Take 12 months of search queries, page views or product sales and rank them from most to least frequent.
  2. Describe the concentration. Calculate the share held by the top 10 items, items 11 to 100 and everything after. A log-log chart of rank against frequency shows how heavy the tail is.
  3. Cluster the tail by intent. Group rare queries or products by the need behind them, not by exact wording, so each cluster has enough volume to act on.
  4. Compare value, not just volume. Measure conversion rate and revenue per visit by band, and note how many conversions each figure rests on.
  5. Pilot before scaling. Build pages for a few clusters and compare them with similar clusters you have not yet covered.

What to measure

Share of impressions and clicks by rank band, conversion rate and revenue per visit by band, and the number of tail clusters that reach a minimum level of traffic. Guardrail: judge tail content by combined results per cluster, not page by page, and do not publish pages that exist only to catch a query.

Worked example

Worked example: 1,000 queries following an exact Zipf pattern (illustrative model)
Query rank bandShare of searchesAssumed conversion rateShare of conversions
Rank 1 13.4% 1.0% 7.7%
Ranks 2–10 25.8% 1.5% 22.3%
Ranks 11–100 30.2% 2.0% 34.7%
Ranks 101–1,000 30.7% 2.0% 35.3%

Search shares are calculated from an exact 1-over-rank pattern across 1,000 queries; real query data rarely fits so neatly. Conversion rates are assumptions chosen to show the arithmetic, not evidence that tail queries convert better. The model also assumes you rank equally well for every query, which is rarely true.

Where it breaks

  • Not every heavy tail is Zipf. Many datasets fit other curves; focus on measured concentration rather than the formula.
  • A longer tail is not a fatter tail. More niche items can each sell less, leaving demand still concentrated in hits.
  • Some of the tail is worthless. Many rare queries carry irrelevant or purely informational intent.
  • Your data hides part of the tail. Search tools sample data and may omit rare queries for privacy reasons.
  • Small samples mislead. Tail conversion rates often rest on a handful of orders.
  • Ethics and quality: mass-producing thin pages for every query wastes readers' time and can breach search engines' spam policies.

What if you apply it incorrectly?

A publisher reads "the long tail converts better" as a law of nature and generates thousands of near-identical pages, one for each keyword variant. Traffic rises briefly, then the site's overall quality signals suffer and the team spends months pruning pages. Conversion data later shows most tail visitors were researchers, not buyers. The tail was a hypothesis to test, not a guarantee.

Myth versus evidence

The social-video version

Long-tail keywords always convert better, so the tail is where the money is.

What the evidence supports

Zipf describes how frequency falls with rank; whether the tail converts or pays has to be measured for each business.

Five-minute field assignment

Export the last 12 months of queries from Google Search Console. Sort by clicks and calculate the share of clicks from the top 10 queries, from queries 11 to 100 and from everything after. If you track conversions by landing page, compare pages that mainly serve broad queries with those that serve specific ones, and note how many conversions each figure is based on.

Checkpoint

Two questions: one about the idea, one about a decision. Answer both correctly to master this law. Questions are drawn from a larger bank and graded on our server; a retry brings different questions.

Evidence ledger

Every factual claim in this lesson, where it comes from, and how far it has been checked.

  • In word frequencies, an item's frequency is roughly inversely proportional to its rank; Zipf popularized this in his 1949 book.

    Source cited, awaiting editor check Zipf, G. K. (1949). Human Behavior and the Principle of Least Effort. Addison-Wesley.

    Limit: Approximate pattern; other heavy-tailed distributions often fit similar data.

  • Anderson argued that with effectively unlimited online shelf space, niche products together can form a market rivaling the hits.

    Source cited, awaiting editor check Anderson, C. (2004, October). The long tail. Wired.

    Limit: Magazine argument, not a peer-reviewed study; the link checker could not open the page.

  • Elberse found demand in the digital media markets she studied remained concentrated in hits, with a longer but flatter tail.

    Source cited, awaiting editor check Elberse, A. (2008). Should you invest in the long tail? Harvard Business Review.

    Limit: Article behind a paywall; described only qualitatively; media markets of that period.

  • The tea shop and the Zipf query table.

    Illustrative, not a factual claim

    Limit: Teaching scenario and model; conversion rates are assumptions.

Further reading

Which other law could explain this decision?

Editorial review: Ali Dezhami, November 8, 2026. Last substantive update: November 8, 2026. Our methodology