What Load sample turns up
Load sample fills the box with a 117-word travel note from a weekend in Lisbon. Two paragraphs, short sentences, the kind of draft people type on a phone. With stop words hidden, the top three rows read:
- view, 5 uses, 4.27 percent of the text
- tram, 4 uses, 3.42 percent
- really, 3 uses, 2.56 percent
Click view in the list. The In context panel prints all five lines and reports the closest repeat as 8 words apart: "The view from Miradouro da Graça was better than the view from Alfama." A reader hears the echo inside one sentence. Swap one of them for "the lookout" and the line reads clean with the same meaning.
Really is the row to act on. View and tram belong in a note about riding trams to lookouts. A filler word in the top three is a habit, and three uses in 117 words is enough for a reader to notice.
Which words the stop list hides
Hide stop words removes 171 common English words from the ranking: articles, pronouns, forms of be, have and do, prepositions, and contractions such as don't and it's. They still count toward the Words total, so every share stays a share of the full text. The Hidden by stop list box shows how many were skipped. In the sample, 56 of 117 words sit on the list.
Turn the option off and the sample ranking changes shape:
- view 5
- tram 4
- really 3
- alfama 2
- got 2
- the 19
- and 5
- view 5
- was 5
- we 5
The list follows the NLTK English set, the stop list most text-analysis code starts from. The set includes just and very, two of the filler words editors cut most. To audit filler, uncheck Hide stop words and type the word in the filter box.
Keep the list off for work where small words carry the signal. Authorship studies compare how often two writers use the, of and upon. A cover letter review counts I against we. Both vanish with the filter on.
Phrases show repeated wording
Single words miss repetition spread over two or three words. Switch to 2-word phrases and the sample returns five pairs used more than once: the view (5), the tram (3), view from (2), was really (2) and we took (2). Phrases seen once are left out. In a long text, the full list of pairs runs to thousands of rows of noise.
In phrase mode the stop list removes only pairs made entirely of stop words, such as and the or of the. The view stays because view is a content word. Phrases never cross a sentence end, a colon or a blank line, so "rain. On" never counts as a pair.
Three-word phrases pay off on longer copy. Paste a landing page and the stock lines surface: "sign up today", "in this guide", "at no cost". In a contract, the same mode flags boilerplate clauses repeated across sections.
How the counter splits words
| You type | Counted as |
|---|---|
Don't, don’t | don't, with straight and curly apostrophes merged |
e-mail | e-mail, one word |
café | café, accents kept |
Tram, TRAM | tram, while Match case is off |
3.5, 1,000 | one number each, only with Count numbers on |
U.S. | u and s, two rows |
run, runs, running | three separate rows |
A word here is a run of letters or digits, joined by apostrophes or hyphens. Slashes, underscores, emoji and every other symbol act as gaps. So email/password counts as two words, and well-known and well known land in different rows.
What the share column measures
For single words, share is the count divided by total words. For phrases, the tool multiplies the count by the phrase length first, the formula most keyword density checkers use:
share = count × words in phrase ÷ total words × 100
In the sample, the view appears 5 times in 117 words. 5 × 2 ÷ 117 gives 8.55 percent.
Big shares for small words are normal. In the Brown Corpus, a million-word collection of American English from 1961, the alone makes up close to 7 percent of all words. The sample hits 16 percent because a short diary puts the in nearly every clause.
Google publishes no target keyword density, and its spam policies list keyword stuffing as a violation. Our rule when editing web copy: once a content word other than the page topic passes 2 percent of a 1,000-word article, read every line in the In context panel before publishing. Five repeats inside one paragraph read worse than the same five spread over ten.
Distinct ratio drops as text grows
Distinct ratio divides distinct words by total words. Linguists call this the type-token ratio. The sample scores 59.8 percent, 70 distinct words out of 117.
The figure falls as a text gets longer, since common words keep coming back while new vocabulary arrives slower. A 300-word product blurb and a 5,000-word chapter from the same writer score far apart. Compare ratios only between texts of similar length, for example the first 500 words of each draft.
Where this tool stops
- No stemming. Run, runs and running fill three rows. Type
runin the filter box to see the forms side by side. - English stop list only. French le and la or Spanish el and los stay in the ranking. For other languages, read past the top rows.
- Scripts without spaces. Chinese, Japanese and Thai write words without gaps, so an unbroken run of characters counts as one word. Results for those scripts are not usable.
- Dotted abbreviations split. U.S. and e.g. turn into single letters. Write US and eg first if those counts matter to you.
- Plain text only. Open .txt reads .txt, .md and .csv files up to 5 MB. For Word files or PDFs, copy the text out and paste.
- Speed. In testing, 1 MB of text (about 197,000 words) ranked in about a tenth of a second. Two-word phrases on a 5 MB file took about 1.5 seconds, so on long inputs the list refreshes once you pause typing.