Word counter

There is no universal definition of a word

Ask ten people to count the words in "a half-baked compromise that won't survive the week" and you will not get one answer. Is "half-baked" one word or two? Is "won't" one or two? Nothing outside the sentence settles it — the language has no committee that ruled on the question.

So every counting tool picks a rule and lives with the consequences. There are two families of rule. One breaks the text at spaces and calls each surviving chunk a word. The other breaks at anything that is not a letter, which pulls hyphenated compounds and contractions apart. Word processors mostly sit somewhere between the two, carrying special cases accumulated over decades.

This page uses the space rule with nothing bolted on. The advantage is not that it is more correct than the alternative — neither is. The advantage is that it is predictable. You can look at any string in your document and know how it will be treated before the number appears.

The cases where two tools give two answers

The disagreements are concentrated in a handful of constructions, and they are the same handful every time. Here is what this page reports for each, and what a tool that breaks on non-letters would report instead.

Where sentence counting surprises people

A sentence ends at a full stop, question mark, exclamation mark or ellipsis — but only when the next thing along is a space or the end of the text. Closing quotes and brackets are allowed in between, so He said "stop!" closes a sentence rather than leaving one hanging open.

That single condition does most of the work. The dot in 3.14 has a digit after it. The dots in example.com and in .co.uk have letters after them. None of them ends a sentence, and no list of exceptions was needed to get there.

Where the rule fails is the abbreviation that ends in a dot and is followed by a space. "Dr. Okonkwo arrived at 9 a.m. sharp." is one sentence to a reader and three to this counter, because "Dr." and "a.m." both look exactly like endings. Any tool without a dictionary of abbreviations behaves this way, and every such dictionary is incomplete. If your text is thick with them, treat the sentence count — and the average sentence length that depends on it — as approximate rather than exact.

• A run of terminators is one ending. "Really?!" and "wait..." each close a single sentence, not two or three.
• Text with no terminator at all counts as one sentence. That is why a headline or a bare list item reports 1 rather than 0.
• A line of nothing but punctuation counts as nothing, because a sentence has to contain a word to exist.

Paragraphs are blank-line blocks, not line breaks

Press Enter once and you begin a new line. Press it twice and you begin a new paragraph. This page follows that convention, and it is by far the most common reason a paragraph count comes out lower than expected.

The reason for the rule is pasted text. Copy a page out of a PDF and it usually arrives hard-wrapped, with a real line break at the end of every visual line whether the sentence ended there or not. Treating each of those as a paragraph would turn one page of prose into forty paragraphs and make the paragraph average meaningless. The line count is reported separately for anyone who wants the other number.

Two consequences follow. A bulleted list written on consecutive lines is one paragraph rather than six, so add blank lines between items if you want them counted apart. And a block holding no letters or digits — a row of asterisks used as a scene break, say — is not counted as a paragraph at all.

What the structural numbers are for

Words per sentence and words per paragraph are not scores. Nothing is being graded, and there is no target the page wants you to hit. They are here because they make a shape visible that is genuinely hard to see from inside a draft.

Prose written for a general audience tends to average somewhere from the mid-teens to the low twenties. Academic writing, legal drafting and technical documentation commonly run higher, and that is a property of the genre rather than a fault to correct. The figure is worth attention mainly when it moves: a section averaging 34 words per sentence inside a draft that otherwise sits near 18 is usually the section you wrote fast and have not reread.

An average also hides its own distribution. A page alternating four-word sentences with forty-word ones averages twenty-two, and reads nothing like a page of steady twenty-two-word sentences. If the average looks reasonable and the writing still feels heavy, the variation is where to look, not the mean.

Paragraph balance behaves the same way, which is why the longest and shortest paragraph appear next to the average — the distance between them says more than the average does. A single 280-word paragraph among paragraphs of 70 usually holds two ideas that have not been separated yet, or one idea that arrived with everything you know about it still attached.

Counting in a second language, or a script without spaces

The space rule assumes spaces. Chinese, Japanese, Thai, Lao, Khmer and Burmese do not put them between words, and a space rule would report an entire paragraph as one word. Where those scripts appear, the page hands the run to the browser's own word segmentation instead, which knows where the boundaries fall.

That gives a defensible word count, but be careful what you do with it. Length in Chinese and Japanese is conventionally stated in characters rather than words, so a limit expressed in words for a Japanese text was usually written with English in mind. If a submission system asks for 2,000 words and your text is in Japanese, that is a question for whoever set the limit, not for a counter.

Between two spaced languages the arithmetic differs but the caution is identical. The same argument written in Portuguese and in English does not come out at the same length; Portuguese generally needs more words to say it. A draft that fits a 2,000-word ceiling in one language may need cutting in the other, and translating is not a way to make a word count go down.