Counting Repeated Words, Not Keyword Density
Rank repeated word-like segments, compare case and export the visible table. This is an editing aid for repetition, not a keyword-density target or a grade for vocabulary quality.
Paste text or choose Sample. Choose whether to remove the fixed English stop-word list and compare case exactly, then set whole-number minimum length and occurrence thresholds and choose the ranking order. Copy the displayed table as tab-separated text or download quoted CSV; the selectable table is available if Copy fails. Read a repeated term in context before replacing it, and record the input and filters next to an exported table.
What counts as a word, and what the percentages divide by
Intl.Segmenter with locale en identifies word-like Unicode segments and numbers. Map-based counts support words such as constructor. With case sensitivity off, tokens are lowercased; there is no stemming, accent removal or Unicode normalization. Minimum word length uses code points, not visible grapheme clusters. Equal frequency ties keep first-appearance order; alphabetical order uses en collation.
Total counts include all word-like segments. Retained and unique counts, plus their type-token ratio, follow stop-word removal but precede minimum thresholds. Row percentages use the original total. Filtered percentages can sum to less than 100 percent.
A six-word sample, ranked three ways
Sample contains six words. With default settings, café and cat each occur twice (33.3 percent), and dog once (16.7 percent). The is excluded; retained words are five, unique words three, and type-token ratio 0.60. Minimum occurrences 2 hides dog without changing those summary values.
| Setting | Effect on the Sample |
|---|---|
| Default | café: 2, cat: 2, dog: 1 |
| Case sensitive | cat and Cat are separate entries. |
| Minimum occurrences 2 | Only café and cat remain in the displayed ranking. |
What the numbers do not mean
- Stop words are an optional English list, not an automatic relevance filter.
- Removing no or not can hide important context.
- Type-token ratio changes with sample length and token rules, so it does not rate writing quality.
- Composed and decomposed spellings can remain separate.
- Browser Unicode/ICU data can change segmentation or collation.
- Input is capped at 1,000,000 UTF-16 units; thresholds must be whole numbers from 1 to 1,000,000.
- Unsupported Segmenter browsers or invalid settings block the ranking and exports.
- CSV uses quoted fields and CRLF rows; quoting is formatting, not a promise about spreadsheet formula handling.
Keep the draft and settings for comparisons. MDN documents locale-sensitive segmentation. For a brief summary with multiple character units, use Text Stats; its smaller stop list and top-five view differ from this full ranking.