To check keyword density in content is to count how often a token appears among all recognized words in the text, then divide that count by the total word tokens and multiply by 100 — a single formula the Keyword Density Checker applies directly to pasted copy, with every percentage built on a disclosed denominator and no ranking promise attached. The first sentence matters because it pins down the formula a reader has likely seen written vaguely across many guides, and it states up front that no tool on this page can convert a number into a search position. A density percentage is the share of total tokens that one word, or one exact phrase, occupies in the text you paste; nothing more, nothing less. The check works as a reading aid, not a scoring engine. With that framing in place, the rest of this article walks through what the tool counts, the three concrete steps to run a check, the tokenization rules that quietly shape every row, the math behind one worked sentence, and the limits that prevent the percentage from being mistaken for a ranking target.
What the checker counts when you check keyword density in content
The report you see after pressing the button is built from a small, fixed set of measurements. Total words is the count of every token the tokenizer recognizes in your pasted text. Unique words is the number of distinct normalized tokens — lowercased through Unicode rules — that survive the same pass. The frequency table lists at most 50 single-word rows, sorted by descending count and then alphabetically for ties, so the interface stays readable even on long drafts. If you enter an exact keyword or phrase, the same tokenizer scans for contiguous matches, and phrase matches may overlap by design.
None of those counts are estimates or rounded guesses. The denominator for every percentage shown is the total tokens figure, including words that the minimum-length filter may hide from the table; this means a filtered row does not quietly inflate the percentages of the words you do see. Pasted text is bounded at 500,000 characters, which covers very long articles and most landing pages without ever sending the text to a server. For a closer look at how the same counts behave on a wider variety of inputs, the guide on how to find keyword density in any text follows the same logic across longer samples.
How to check keyword density in content in three steps
The tool exposes only the controls it actually needs, so the workflow stays short. Three steps take you from a pasted draft to a transparent frequency table you can read alongside the source text.
- Paste the visible text you want to inspect and optionally enter one exact keyword or phrase. Type or paste the body copy into the input area. If you have a specific phrase whose repetition you want to inspect — a product name, an offer, a question readers ask — type it into the keyword field as it appears in the text, including any apostrophes or accents. The checker does not stem, lemmatize, or expand synonyms, so entering running shoes will not match shoe or ran.
- Choose the minimum word length for displayed single-word rows, then select Check keyword density. The minimum-length control affects only which single-word rows appear in the frequency table. Total words, unique words, and the density denominator still include every recognized token, so raising the minimum does not pretend those short tokens have disappeared from the source.
- Review counts and surrounding prose, using percentages as descriptive evidence rather than a ranking target. Read the total and unique counts first to confirm tokenization matches your reading of the draft, then scan the frequency table for repetition, missing terminology, or accidental imbalance. Open the Keyword Density Checker in a fresh tab whenever you want to compare a revision; the same tokenizer and denominator let you see whether an edit changed the picture.
Tokenization rules that shape your numbers
Two drafts with identical prose can produce different percentages if their tokenizers disagree on what counts as one word. Three rules drive every count on this page, and once they are visible, the table is easier to read against the source.
Unicode letters and numbers are one token
Tokenization uses Unicode letter and number properties, so words written beyond basic ASCII remain countable. French l'été stays as one token, German compounds like Donaudampfschifffahrtsgesellschaft stay as one token, and accented brand names do not fracture into prefix and suffix fragments. Matching is case-insensitive through Unicode lowercase conversion, so Red, red, and RED all land in the same bin.
Internal apostrophes keep tokens together
An internal straight or typographic apostrophe can stay inside a token, which keeps forms such as don't and l'été as single units. Punctuation, dashes, symbols, and whitespace separate tokens, so the hyphenated phrase red-blue tokenizes into two tokens — red and blue — each contributing to its own row. Commas, periods, question marks, and the like terminate tokens rather than ride along with them.
The display filter never rewrites the denominator
The minimum-length control is a display filter, not a deletion step. Setting it to a minimum of two hides one-character rows but leaves the total tokens and the density denominator untouched. A higher minimum therefore shortens the visual list without inflating the percentages of the longer words that remain visible, which is the kind of quiet distortion a reader might miss if the table alone is trusted.
A worked sentence with transparent math
The same formula drives every row in the table, and it is worth one worked example so the numbers stop feeling arbitrary. Take the test sentence Red blue red. RED green blue red-blue. The tokenizer reads it as eight total tokens after normalization, and the frequency table breaks down as follows.
| Token | Count in sentence | Density |
|---|---|---|
| red | 4 | 50.0% |
| blue | 3 | 37.5% |
| green | 1 | 12.5% |
The arithmetic behind those rows is straightforward. Red appears four times out of eight total tokens, so 4 ÷ 8 × 100 = 50.0%. Blue appears three times, giving 3 ÷ 8 × 100 = 37.5%. Green appears once, for 1 ÷ 8 × 100 = 12.5%. The hyphenated red-blue splits into two separate tokens, which is why red and blue each pick up a count from that final span rather than being counted as a single hyphenated form.
If you enter the exact phrase red blue, the same tokenizer counts two contiguous matches — the opening Red blue and the red-blue hyphenated token at the end of the sentence, which separates into the two tokens red blue — for a phrase density of 2 ÷ 8 × 100 = 25.0%. Phrase matches may overlap by design, and the phrase denominator is the total tokens, not the number of possible starting positions or the phrase length. The same style of calculation appears in a longer write-up on calculating keyword density without guesswork.
What the report deliberately does not do for you
The checker has a short list of things it intentionally cannot do, and each one is a deliberate design choice rather than a missing feature. Treating the percentages as if any of those features existed is the fastest way to misuse them, since most of those choices exist because they would require language- and corpus-specific reference data the tool deliberately avoids embedding.
| Task | What the checker does instead |
|---|---|
| Remove stop words | No stop-word list is embedded; every recognized token appears in the total rather than hiding behind an invisible filter. |
| Stem, lemmatize, or expand synonyms | No morphological normalization is applied, so forms like ran, runs, and running stay separate tokens you can inspect. |
| Detect language or strip boilerplate | No language detector or HTML, footer, or navigation removal runs, which is why pasting the rendered visible text matters more than pasting the page source. |
| Fetch a URL or call Search Console | No network requests are issued; only pasted text is analyzed, and the live URL is never contacted. |
| Predict ranking impact | No ranking guarantee, target percentage, or difficulty score is produced, in line with published search guidance. |
| Send the draft to an AI model | All processing runs in the current browser tab; no AI model is called and no text leaves the device. |
Reading those limits alongside Google Search Central's spam policies on keyword stuffing helps settle expectations. Search engines evaluate usefulness, intent satisfaction, originality, structure, links, reputation, and many other signals beyond raw repetition. The checker deliberately avoids embedding a stop-word list or ranking-threshold reference data because those choices require language- and corpus-specific decisions that would distort the count. A clean frequency report supports editorial decisions; it cannot replace them.
Mistakes that skew a check before you paste
Most inaccurate density reports start with the input rather than the formula. Three preparation habits keep the numbers honest, and they matter long before the first percentage appears.
- Paste only the visible text you intend to inspect. Navigation, footer, repeated templates, markup, and shared sidebars all produce tokens that have nothing to do with the article body. A 600-word post surrounded by 2,000 words of header and footer text will report densities that are only a fraction of what the draft itself contains, which then tempts writers to pad repetition back up to a target they should never have been chasing.
- Recognize an empty or token-free input rather than treating zero as data. The checker refuses to produce a misleading zero-percent report when the pasted text contains no recognized tokens. If that rejection appears, the input is almost certainly markup, a stripped template, or whitespace — clean the source and try again before drawing any conclusion about the draft itself.
- Compare revisions using the same tokenizer and denominator. A draft before and after an edit produces comparable percentages only when the same minimum-length filter and the same Unicode tokenization rules are in place for both runs. Switching the minimum length between checks hides short tokens in one report and not the other, which can make an edit look more or less effective than it really was.
Once the input is clean, the percentage column rewards a specific reading habit: treat each row as a clue rather than a score, and open the surrounding sentences in the source draft before changing a word. That habit is the difference between using density as evidence and using it as a target. Density percentages are best read as a description of what is already on the page, not a forecast of where it will rank. The Keyword Density Checker gives you a transparent local count to compare edits against: paste the visible copy, optionally enter an exact phrase, run the analysis, and read the surrounding prose before deciding what to revise. Treat the report as one piece of evidence among many — alongside editorial judgment, real search performance data, and the actual experience of the reader trying to complete a task on the page.