Keyword Density: What It Measures and What It Does Not

Keyword density is a simple statistic: how often a word appears, divided by how many words there are. It is also easy to misread, because people arrive looking for a figure to hit, and the figure is not a property of the text alone — it depends on what the tool divides by, which words it leaves out of the ranking, and whether it merges word forms. This guide explains how the keyword density checker on this site builds its table, works several examples with numbers computed by the same engine, and explains why the site publishes no target density and what the table is genuinely good for.

The definition, and the base it divides by

For every word in the text, the checker counts its occurrences — case-insensitively, so "Chain" and "chain" merge — and divides by the total word count of the whole text. Total means total: every token the word counter would count, including "the", "and", "of", and the other function words. That base is a deliberate convention documented on the methodology page, and it is the single most important thing to know when comparing this tool's figure with another's.

The ranking is a separate step. Before the table is built, two kinds of token are removed from the candidates: a fixed list of 74 common English function words, and single-character tokens such as "a" or "I". They are removed from the ranking only — they still sit in the base. So a stop word can be the most frequent word in your text and never appear in the table, while every percentage in the table is still computed against a count that includes it. Ties are broken alphabetically, which matters in short texts where most words appear once.

Nine words that show all of it

The classic pangram "The quick brown fox jumps over the lazy dog." has 9 words and 8 unique ones, so exactly 1 word repeats — and that word, "the", is on the stop list. Asking for the top three entries returns:

KeywordCountDensity
brown 1 11.11%
dog 1 11.11%
fox 1 11.11%

Every content word appears once, so all 6 of them tie at one occurrence and 11.11% — one part in 9 — and the alphabetical tie-break decides which three you see. The base of 9 includes the 3 stop-word tokens ("the" twice and "over" once). Had they been stripped before dividing, each remaining word would show 16.67% instead.

A worked example on a full paragraph

Here is an original paragraph of 66 words about bicycle maintenance:

A bicycle chain wears slowly, and a worn chain wears the cassette with it. Check the chain every few weeks with a chain checker: if the tool drops into the links, the chain is stretched and it is time to replace it. Replace the chain before the cassette wears, and the cassette will outlast several chains. Clean the chain after wet rides and oil it lightly.

The top five entries the checker returns:

KeywordCountDensity
chain 7 10.61%
cassette 3 4.55%
wears 3 4.55%
replace 2 3.03%
after 1 1.52%

Read the first row mechanically: "chain" occurs 7 times, and 7 ÷ 66 × 100 = 10.61%. Two things are worth noticing. First, "chains" — the plural, with a count of 1 — is a separate candidate with its own row further down the full ranking, at 1.52%, because the engine performs no stemming; the singular and the plural would merge only if you rewrote one of them. Second, the resolution of the percentage is set by the base: with 66 words, a single occurrence is worth 1.52%, so no row in this paragraph can move by less than that. The gap between "chain" and the second row, "cassette" at 4.55%, is the paragraph's real signal: one subject, mentioned far more than anything else.

The same paragraph on a different base

Suppose a different tool strips stop words before it divides. Of the 66 tokens in the paragraph, 38 survive the stop list and the single-character filter, and 28 do not. On that reduced base, "chain" would be reported at 18.42% — same text, same count, a visibly larger number. Neither figure is wrong; they answer different questions, "share of everything written" versus "share of the content words". But they are not comparable, and a density target expressed without saying which base it assumes is not a target at all. That is half the answer to the question most people bring to this page.

When a repeated word is on the stop list, and when it is not

The stop list changes which words you see, not the base. The two versions of a short passage below are identical except that the verb "like", which is on the stop list, has been replaced by "love", which is not.

I like the sound of rain on a tin roof. I like the smell of the street after rain. I like the way the town goes quiet, and I like that nobody hurries.

I love the sound of rain on a tin roof. I love the smell of the street after rain. I love the way the town goes quiet, and I love that nobody hurries.

Both versions have 33 words. The top three entries for the first version, with "like":

KeywordCountDensity
rain 2 6.06%
after 1 3.03%
goes 1 3.03%

And for the second version, with "love":

KeywordCountDensity
love 4 12.12%
rain 2 6.06%
after 1 3.03%

In the first version "like" appears 4 times — more often than any word that does chart — and is absent from the table; "rain" leads instead at 6.06%. Change the verb and "love" goes straight to the top at 12.12%, while "rain" keeps exactly the density it had before, 6.06%, because the base did not change. This is the mechanism to remember when a table seems to be missing something obvious: a word can be excluded from the ranking without being excluded from the arithmetic. The list is fixed and English-only, and a few of its entries are less obvious than "the" and "and" — "like", "over", "just", "more", "most", "only", "out", and "up" are all on it. If your subject happens to be one of those words, this tool will not chart it, and the word frequency counter, which applies the same list, will not either.

Word forms are not merged

Stemming — collapsing "plant", "plants", and "planted" into one root — is absent here, on purpose. A 36-word garden note shows the consequence:

We planted the beds in April. The plants took a week to settle, and planting the second row went faster. By June every plant had flowered, and the planting plan for next year was already drawn.

Filtering the full ranking for words that begin with "plant" gives 4 rows:

KeywordCountDensity
planting 2 5.56%
plant 1 2.78%
planted 1 2.78%
plants 1 2.78%

The leading form, "planting", has a density of 5.56%. All 4 forms together account for 5 of 36 words, or 13.89% — a share no single row shows. The tool keeps them apart because, for editing, the split is the information: several forms of one verb charting at once means that verb is carrying the paragraph. For the opposite question — is this note about planting? — you add the rows yourself. Either way, a tool that stems would report one row and a larger number for identical text, which is the second reason density figures do not travel between tools.

Why this site publishes no target density

People arrive at this guide, and at the checker, searching for the best keyword density. The honest answer is that this site does not state one, for two separate reasons.

The first is sourcing. Every figure on this site traces to a documented method — the engine's own counting rules, the published Flesch coefficients, the reading-rate meta-analysis behind the timing estimates. There is no equivalent source for a target density, and a percentage without a source is an opinion dressed as a measurement. For the same reason this guide makes no claims about how search engines treat the figure: those are not claims a word counter can verify.

The second is that the number is not well defined without the tool. As the examples above show, the same text yields different percentages depending on whether the base includes function words, which stop list is applied, whether word forms are merged, and whether the measurement covers a draft's body text or a rendered page complete with navigation and footers. A target stated for one convention is meaningless under another. Anyone who hands you a density target owes you the convention it assumes; if they cannot say, the target cannot be checked.

How to read the table

Treat the table as a description of repetition, and compare it with itself rather than with a threshold. Three readings are useful. Compare rows against each other: a subject word trailing words you would not have guessed suggests the draft talks around its topic. Compare drafts of the same piece: paste the text before and after an edit and watch which rows move — because the base is the same text, the change is meaningful in a way that a cross-tool comparison is not. And compare counts, not only percentages: on a short text the percentage is coarse (1.52% per occurrence in the chain paragraph), and the raw count on the word frequency counter is often the clearer number.

What the table cannot tell you matters just as much. It measures occurrence, not relevance: a word can be repeated and still be beside the point, or absent and implied by synonyms the counter does not know. It ignores sentence structure entirely — a text with a tidy keyword profile can still be hard going, which is what Flesch Reading Ease estimates instead. And it ranks single words, so a phrase is visible only through its parts.

Where the count itself breaks

Because density is a ratio of two token counts, every tokenization edge in the word counter reaches the percentage. A number written with a comma, "2,500", is two tokens, and each inflates the base. A pasted URL contributes three tokens — the scheme, the domain, and the top-level domain — and in a link-heavy draft "https" or "com" can climb the table as though they were subjects. Hyphenated compounds stay whole, so "e-commerce" ranks as one candidate rather than two. Emoji are never words and never enter the base. None of these are errors; they are the counting rules laid out in the guide to what counts as a word, and knowing them explains most of the small differences between this table and another tool's. The rest of the guides take the site's other metrics apart the same way.

Frequently asked questions

What is keyword density?

A word's occurrences divided by all the words in the text, times 100. In the bicycle-chain paragraph in this guide, "chain" appears 7 times in 66 words, so its density is 10.61%. The base is every word the word counter would count, including function words such as "the" and "and".

Is there a best keyword density?

This site does not publish one, and this guide does not give one. No source this site relies on states a target, and the figure itself depends on the tool that produces it: the base it divides by, the stop list it applies, whether it merges word forms, and whether it measures a draft or a rendered page. Use the table to compare drafts of the same piece, not to chase a number.

Why is my most frequent word missing from the table?

Because it is one of the 74 English function words excluded from the ranking, or a single-character token. In the pangram example, 9 words contain 8 unique ones, so exactly 1 word repeats — "the" — and it is a stop word, so "brown" leads a table of ties instead. Excluded words still count toward the base, which is why the remaining percentages do not change.

Why do "plant" and "plants" get separate rows?

The engine does not stem. In the garden example, 4 forms of the same root chart separately: the leading form, "planting", at 5.56%, and all forms together at 13.89% of 36 words. Separate rows are deliberate — merging forms would hide the repetition the table exists to reveal.

Why does another tool report a different percentage for the same text?

Most often because of the base. Here density divides by all 66 words of the chain paragraph, giving "chain" 10.61%; a tool that removes stop words before dividing would use a base of 38 and report 18.42% for identical text. Different stop lists, stemming, and measuring a live page instead of a draft account for the rest.

Does the checker measure phrases?

It ranks single words. A two-word phrase shows up through its parts: if both halves chart with similar counts the phrase is probably well represented, but the table cannot confirm that they appeared together. The word frequency counter uses the same ranking and is the better lens when raw counts matter more than shares.

Every number in this guide was computed at build time by the same engine that runs the checker, and analysis on the tool pages happens locally in your browser — nothing you paste is transmitted or stored. The stop-word convention and counting rules are documented on the methodology page.