Keyword Density Checker
Your top keywords ranked live by count and share of total words — stop-words excluded — so you can see at a glance whether your draft actually says what it's supposed to be about.
Keywords appear once you type.
The table above is arithmetic, not judgement: it counts how often each word appears, divides by the total number of words, and sorts. Everything on this page explains that arithmetic — what the base is, which words are left out and why, why word forms are never merged, how many rows the engine will return, and why no target percentage is printed anywhere on this site. The companion guide, Keyword density explained, covers the same mechanics at greater length; the word frequency counter shows the same ranking read as an editing tool rather than a topic check.
What the percentage measures: count divided by all words
Each row of the table carries two numbers. The count is literal: the number of times that word occurs, after case is folded so that "Coffee" and "coffee" are one entry. The density is the count divided by the total number of words in the text, multiplied by a hundred. That total is the same figure the Total words tile reports, and it includes every word — the stop words filtered out of the ranking are still in the base. A word that appears 3 times in a 40-word paragraph therefore shows 7.5%, whatever else the paragraph contains.
That base is the single most important convention to know, because it is not universal. A checker that removes stop words before dividing reports the same count against a smaller base and therefore a larger percentage. Neither convention is wrong; they answer slightly different questions — what share of the text is this word, versus what share of the content words is this word — but figures from the two are not comparable, and a target expressed in one convention is meaningless in the other. When a style guide or a client hands you a density figure, the first question is which base it assumes. The worked examples below compute both, so you can see the size of the gap on a real paragraph.
Worked example 1: a 40-word paragraph
"A burr grinder crushes coffee beans between two steel plates, and the gap between them sets the grind size. A finer grind exposes more coffee to the water, so espresso wants a fine grind and a press wants coarse grounds."
The engine finds 40 words, of which 13 are on the stop list and 27 are candidates for the ranking. Those candidates collapse to 22 distinct keywords. (The Unique words tile would read 29 for this paragraph — larger than 22, because it counts distinct stop words too.) Asked for its default number of rows, the engine returns the top 10:
| Keyword | Count | Density |
|---|---|---|
| grind | 3 | 7.5% |
| between | 2 | 5% |
| coffee | 2 | 5% |
| wants | 2 | 5% |
| beans | 1 | 2.5% |
| burr | 1 | 2.5% |
| coarse | 1 | 2.5% |
| crushes | 1 | 2.5% |
| espresso | 1 | 2.5% |
| exposes | 1 | 2.5% |
Read the first row: "grind" occurs 3 times, and 3 ÷ 40 × 100 = 7.5%. The next three rows — "between", "coffee", and "wants" — tie at 2 occurrences each, and the tie is broken alphabetically, which is why "between" outranks "coffee" despite having nothing to do with the subject. That is worth internalizing: the order within a tied group carries no information about importance. Below the ties come the single-occurrence words, also alphabetical, and the table stops at the requested number of rows even though 12 more distinct keywords exist in the paragraph.
Now the same rows against the other base — the 27 content words alone:
| Keyword | Count | ÷ all 40 words | ÷ 27 content words |
|---|---|---|---|
| grind | 3 | 7.5% | 11.11% |
| between | 2 | 5% | 7.41% |
| coffee | 2 | 5% | 7.41% |
| wants | 2 | 5% | 7.41% |
Same text, same counts, and the top word reads 7.5% under this site's convention but 11.11% under the other. The gap grows with the share of function words in the text, so it is widest in exactly the conversational prose where people most often compare tools. The rule for this page is simple: never compare a percentage from here with one from a tool that documents a different base — or documents no base at all.
Worked example 2: case, hyphens, and word forms
"Our e-commerce team optimizes checkout pages. E-commerce optimization starts with speed: optimize the images, then optimize the scripts, and measure every optimization."
22 words, 14 distinct keywords. The engine returns:
| Keyword | Count | Density |
|---|---|---|
| e-commerce | 2 | 9.09% |
| optimization | 2 | 9.09% |
| optimize | 2 | 9.09% |
| checkout | 1 | 4.55% |
| every | 1 | 4.55% |
| images | 1 | 4.55% |
| measure | 1 | 4.55% |
| optimizes | 1 | 4.55% |
| our | 1 | 4.55% |
| pages | 1 | 4.55% |
Three mechanics are visible here. First, case folding: "E-commerce" at the start of the second sentence and "e-commerce" in the first merge into one row of 2. Second, hyphens hold a word together: "e-commerce" is one candidate keyword, not "e" (which the single-character rule would drop anyway) plus "commerce". Third, and most consequential, word forms are not merged. "optimize", "optimizes", and "optimization" occupy three separate rows with 2, 1, and 2 occurrences; no single row shows more than 9.09%, yet the three forms together account for 5 of the 22 words — 22.73% of the text. If your question is whether a draft stays on its subject, you may need to add related rows by hand, because the engine deliberately will not do it for you.
Why word forms stay separate
Stemming — collapsing "run", "runs", and "running" into one entry — would make the second example's topic look better represented in a single row. The engine does not do it, and the reason is the other tool that shares this ranking. The word frequency counter exists to expose repetition, and repetition hides in word forms: a writer who leans on one verb usually leans on all of its forms, and a stemmed table would show one tidy row where three revealing ones belong. The same engine serves both pages, so the density table inherits the choice. There is a quieter benefit too: without stemming, every row is a string that literally occurs in your text, which makes the table predictable and easy to verify with a search. The cost is that you sum related rows yourself when the question calls for it, as the second example shows.
Worked example 3: what never reaches the table
"I bought 2,500 tickets at $4 a ticket, and I don't regret it."
The engine counts 14 words here, which is more than a glance suggests: "2,500" tokenizes as two words, "2" and "500", because a comma ends a run of characters. The ranking then contains 6 rows:
| Keyword | Count | Density |
|---|---|---|
| 500 | 1 | 7.14% |
| bought | 1 | 7.14% |
| don't | 1 | 7.14% |
| regret | 1 | 7.14% |
| ticket | 1 | 7.14% |
| tickets | 1 | 7.14% |
What is missing tells you the rules. "2" and "4" are gone because single-character tokens are excluded, while "500" stays because it has more than one character — numbers are keywords like any other token. "I" is gone twice over: it is a single character, and it is also on the stop list. "a", "at", "and", and "it" are stop words. The dollar sign is not a word at all. An apostrophe does not break a word, so "don't" charts whole rather than as "don" and "t". And "ticket" and "tickets" are separate rows, per the no-stemming rule above. Every one of the 14 words, excluded or not, is still in the base, which is why each surviving row shows 7.14% rather than a larger share.
The 74 stop words and the single-character rule
The ranking excludes a fixed list of 74 common English function words — the, and, of, that, and their relatives — plus any token of a single character. The exclusion exists because "the" tops nearly every English text, and a table led by articles and conjunctions says nothing about the text in front of you. Two properties of the list matter in practice. It is fixed: the same 74 words on every page and every visit, never adapted to your text. And it is an English list: a text in another language is filtered only by the single-character rule, so its own function words rank as keywords.
Some entries surprise people. Negations ("not", "no") are on the list, so a draft that argues against something shows no trace of the argument in the table. So are the small words that complete phrasal verbs ("up", "out", "over"), the hedges "just" and "like", and the comparatives "more" and "most". If one of those words is the very thing you want to count — an editor hunting for "just", say — a search in your editor is the right tool, not this table. The single-character rule removes "I" and "a" (both also on the list) along with initials, list markers, and lone digits. None of the exclusions touch the base: the Total words tile, and every percentage, still count them.
How many rows: one to a hundred
The engine's keyword function takes a second argument, the number of rows to return, and accepts any whole number from 1 to 100; its default is ten, and the live table on this page uses that default. Outside the range it refuses with a plain message rather than guessing: "Choose between 1 and 100 keywords." Asking for more rows than the text has distinct keywords returns as many as exist — the paragraph in the first example yields 22 rows when a hundred are requested, 10 at the default, and 3 when three are requested — and the shorter lists are always the top of the longer one, because the sort order never changes: count descending, then alphabetical. An empty text, or one made only of spaces, returns an empty ranking rather than an error, which is why the table above simply waits until you type.
Why this page publishes no target percentage
Most people who open a density checker are looking for the right number to hit. This page does not give one, and the omission is deliberate. The first reason is the arithmetic above: the same draft produces 7.5% or 11.11% for its top word depending on the base, and a different figure again once the tokenizer or the measured text changes, so a target that means the same thing in two tools does not exist. The second reason is scope: this site documents how the number is computed and tests that computation; it makes no claim about what any reader, editor, or ranking system wants from the number, because a counter cannot know that. Any percentage printed here as a goal would be a number without a source.
What the table does support is relative reading. The same draft before and after an edit, measured the same way, is a fair comparison. Whether the subject appears among the top rows at all is a fair question. Whether one word stands far above its neighbours is a fair observation. If a brief hands you a figure, ask which base and which tokenization it assumes, then measure the way it measures. The keyword density guide is the longer treatment of all of this, and the guides index covers the rest of the site's metrics.
Reading the table like an editor
Three patterns worth acting on. The subject missing from the top rows: the draft talks around its topic — add substance, not repetitions. One word standing far above everything else: the draft probably reads as repetitive — vary the vocabulary, or confirm the repetition is deliberate terminology. A spread of related terms near the top: the natural profile of writing that stays on its subject. One practical caveat: this tool measures the text you paste, while a crawler measuring a live page sees navigation, footers, and boilerplate too, so on-page percentages from site-audit tools will differ from what you see here even for identical copy. And the common failure mode remains writing toward a percentage: repetition that exists for the counter rather than the reader reads as repetition to the reader.
Frequently asked questions
What is keyword density?
A word’s occurrences divided by the total word count, stop words included in the base. If "grind" appears 3 times in a 40-word paragraph, its density is 7.5%. The table ranks your top words by that share, with common function words (the, and, of…) filtered out of the ranking so the list reflects substance.
Is there a target percentage I should hit?
This site does not publish one. The percentage depends on which base a tool divides by, how it tokenizes, and what text it measures, so a target that means the same thing in two tools does not exist — and this site makes no claims about what any reader or ranking system wants from the number. Use the table for relative readings: the same draft before and after an edit, whether the subject appears near the top at all, whether one word dominates.
Why are stop words excluded from the ranking?
Because "the" tops nearly every English text. Excluding the 74 function words on the stop list surfaces the words that characterize your content. Densities are still computed against all words, stop words included, so each percentage remains a share of the whole text.
Why does another tool show a different percentage for the same text?
Usually the base: a tool that strips stop words before dividing reports a larger share for the same count. Tokenization differs too — hyphenated compounds, numbers with commas, and apostrophes are split or kept whole differently from tool to tool — and a site-audit crawler measures a whole page, navigation and footer included, not just the draft you pasted. Compare percentages only between tools that document the same rules.
Why doesn’t "not" (or "up", "over", "just") appear in my table?
They are on the stop list, which includes negations, the small words that complete phrasal verbs, and common hedges as well as articles and conjunctions. Excluded words still count toward the total that every percentage divides by. If you need to count one of them specifically, your editor’s search is the right tool.
Do numbers count as keywords?
Yes, when they are more than one character long. In "I bought 2,500 tickets at $4 a ticket, and I don't regret it." the token "500" ranks, while "2" and "4" are dropped by the single-character rule. A comma ends a word, so "2,500" is two tokens, not one.
Does this handle phrases (two-word keywords)?
This ranks single words. Multi-word phrases ("espresso machine") show up through their components ranking together — if both halves chart, the phrase is well represented. A phrase-level view is on our roadmap.
How many rows can the table show?
The engine returns between 1 and 100 rows and defaults to ten, which is what the live table on this page requests; outside that range it answers "Choose between 1 and 100 keywords." instead of guessing. Ties are broken alphabetically, and you never get more rows than the text has distinct keywords.
How is this different from the word frequency counter?
Same engine, same ranking, same exclusions. This page reads the ranking as a topic check — is the draft about what it is supposed to be about? — and leads with each word’s share of the text. The word frequency counter reads the same ranking as an editing tool, leading with raw counts and the unique-word ratio to expose habits.
Analysis happens locally in your browser — drafts are never transmitted or stored. See the methodology page for the stop-word list and counting rules.