Why word counts differ between tools, and which one to trust

The same text, four rules

A word count is only as definite as the rule behind it, and tools use different rules. Counted four ways, a plain paragraph of nineteenth-century prose gives the same number every time, and a sentence of modern office English gives four different numbers, 26 to 39. The rules are:

• Spaces only: every run of characters between spaces is a word. This is the simplest rule and the one many quick counters use.
• Unicode word boundaries: the rules in Unicode Standard Annex #29, Unicode Text Segmentation, which browsers build in and programs call when they need words. A hyphen, a slash and a colon between digits all break a word.
• Hyphens split: spaces, dashes and slashes separate, and so does every hyphen, as some older word processors counted.
• Simple Online Counter: a word is a run of letters or digits; an apostrophe, a hyphen or a full stop inside joins it, and so does a comma or colon between digits. Dashes, slashes and spaces separate. It is the rule of the word counter on this site.

The Austen paragraphs have no hyphens, dashes, slashes, times or links, so every rule agrees. The office sentence has all of them, and the spread is half its length again. Most real writing sits between the two: a blog post or an essay usually lands within a few words in a hundred, and a technical document full of dates, versions and compound terms can differ by much more.

Do hyphenated words count as one word?

Usually yes, but not under every rule. Word processors and most online counters treat a hyphenated compound such as well-known or mother-in-law as one word, because no space falls inside it. The Unicode word-boundary rules do not: Annex #29 gives the hyphen-minus no joining role, so a program that segments text with them, such as a browser's built-in Intl.Segmenter, counts well-known as two words and mother-in-law as three.

The disagreement is old. A 2007 OpenOffice.org bug report titled "Word count differs from MS Word" records Writer counting hyphenated words as two where Word counted one. Style guides do not settle it either, since they decide whether to hyphenate, not how to count the result.

Hyphens also join things that are not compounds. A date written 2026-11-04 is one word to a space counter and to Simple Online Counter, and three to the Unicode rules. A phone number written with hyphens, a product code such as A-113 and a page range such as 12-14 typed with a hyphen instead of an en dash all behave the same way.

Dashes, slashes and symbols

A dash separates words; it is the spaces around it that change the count. An em dash typed without spaces, as in rock—paper, leaves one unbroken run of characters, so a space counter counts one word where the other rules count two. Typed with spaces, rock — paper, the dash stands alone, and a space counter counts it as a word of its own: three.

LibreOffice's Help documents its fix for the first case: Writer's word count has an Additional separators setting, which by default holds the em dash and the en dash, so rock—paper is two words. Its options let you add more characters to that list.

Slashes follow the same pattern. And/or, his/her and either/or are one word to a space counter and two to the Unicode rules and to Simple Online Counter. A lone symbol such as & or a bullet character typed as text is a word to a space counter and nothing to the others, because it contains no letter or digit.

Numbers, times and web addresses

Numbers count as words under every rule; the separators inside them are where rules part. The Unicode rules join digits across a comma or a full stop, so 1,000 and 3.14 stay whole, but they do not join digits across a colon: 10:30 is two words, 10 and 30. A counter that splits only on spaces, and Simple Online Counter, keep a time as one word.

Web addresses vary the most. https://example.com/a-b is one word to a space counter and to Simple Online Counter, which treats any web or email address as one word, and four to the Unicode rules: https, example.com, a and b. A reference list full of links can move a count by dozens of words.

Chinese, Japanese and languages without spaces

Chinese and Japanese put no spaces between words, so a space counter counts a whole sentence as one word. The other approaches give very different numbers for the same text. For the Japanese sentence 東京は大きい, "Tokyo is big":

Counting each character is the predictable choice, because a dictionary's answer depends on the dictionary. Microsoft Word keeps a separate statistic for it: the Word object model's WdStatistic list includes a count of characters for Asian languages alongside the word count. Translators quote Chinese and Japanese by the character for the same reason. Korean uses spaces between words and is counted by them. Thai, Lao, Khmer and Burmese, written without spaces, need a dictionary; a counter without one counts a run of them as one word.

Characters differ too

A character count depends on what is counted as a character. Most counters now count what you see, a grapheme, but text is stored in smaller units, and a counter that reads raw text length counts those instead.

Platforms set their own rules on top. X counts a post against 280 characters but weighs Chinese, Japanese and Korean characters as two and every link as 23, as its counting documentation sets out. A text message holds 160 characters of the GSM alphabet, or 70 if one character, an emoji included, falls outside it. The character counter counts the same text each of these ways at once and names the characters that make the counts differ.

Writing to a hard limit

When a form refuses text over its limit, write to the highest count you can get, or test in the form itself. The Common App personal essay box takes 650 words, the University of California's personal insight answers 350 words each, and the UCAS personal statement 4,000 characters across its three answers, spaces and line breaks included. None of them publishes how it treats a hyphen or a dash. To plan a character limit in words before you write, the words and characters estimator puts 4,000 characters at about 714 words of ordinary prose.

1. Count the text under more than one rule. The word counter on this site shows its own count, the spaces-only count and the count with hyphens split, and names the hyphens, dashes, slashes and symbols in the text where they part.
2. Take the highest of them as your count. If that is under the limit, the text fits whatever rule the form uses.
3. Paste into the form early. The form's own counter is the only one that decides. A draft pasted in a week before the deadline leaves time to cut.
4. Remove what inflates one rule. Spaced dashes, & signs and pasted links are the usual causes of a count that jumps between tools.

A safety margin of a word or two in a hundred covers ordinary prose. A text dense with dates, compound adjectives and links needs a wider one, or a rewrite of the parts that split.

Reading time is a rule too

A reading-time estimate is a word count divided by an assumed speed, and tools rarely say what speed. The best evidence is a 2019 review of 190 studies and 18,573 readers by Marc Brysbaert, How many words do we read per minute? in the Journal of Memory and Language: adults read English non-fiction silently at about 238 words a minute, fiction at about 260, and read aloud at about 183.

So 1,000 words is about 4 minutes 12 seconds of silent reading at 238 words a minute, or 5 minutes 28 seconds read aloud at 183. A tool assuming 200 or 300 words a minute gives 5 minutes or 3 minutes 20 seconds for the same text. The word counter on this site states its rates and lets you type your own.