Why Word Counters Disagree: We Tested 24 Texts in Microsoft Word
Paste the same text into Word and into a website and you can get 12 words in one and 20 in the other. Both are right; they count different things. We measured where and by how much.
In November 2026 we typed 24 short texts in eight languages into Microsoft Word and read the word and character counts Word reported. Then we ran the same texts through 79 smaller edge cases, one rule at a time. Every number on this page comes from that test, and you can download all of it at the end.
The short answer
Microsoft Word counts everything between two spaces as one word. Many online counters instead split text at punctuation, so a hyphen, a slash or a dot starts a new word. On ordinary prose the two agree. On text with hyphens, links, numbers or code they drift apart quickly, and on Chinese, Japanese and Thai they are not even close.
Results: Word against a punctuation-splitting counter
The middle column is Word. The right column is a counter that follows Unicode word boundaries, the method behind the Words figure in our Word Counter.
| Text | Microsoft Word | Split at punctuation |
|---|---|---|
| Plain English paragraph (3 sentences) | 38 | 38 |
| German paragraph (3 sentences) | 34 | 34 |
| “state-of-the-art, well-known, user-friendly …” | 12 | 20 |
| An email address and a URL in one sentence | 9 | 17 |
| “3.5 kg, $20, 1,000,000 units at 9:30 on 2026-11-15” | 19 | 22 |
| “and/or”, “km/h”, “1/2” | 17 | 20 |
| “J. R. R. Tolkien met Dr. Smith at 5 p.m. in the U.S.A.” | 18 | 22 |
| German: „E-Mail-Adresse … z. B. … 2–3 Tagen“ | 19 | 22 |
| Portuguese: “Dá-lo-ei … guarda-chuva … arco-íris” | 13 | 17 |
| French: “l’homme qu’on attendait … qu’est-ce que” | 19 | 15 |
| Japanese, 2 sentences | 39 | 22 |
| Chinese, 2 sentences | 32 | 18 |
| Thai, one sentence with no spaces | 1 | 11 |
| Hashtags, a mention and three emoji | 12 | 9 |
Across all 24 texts the two methods gave the same number only 8 times. The gap was largest exactly where people tend to care most: academic writing full of abbreviations, technical text with links and code, and Asian languages.
How Microsoft Word counts words
These rules come from the edge cases, each typed into Word on its own:
- Ordinary spaces, tabs, line breaks and the no-break space separate words. Thin spaces and the narrow no-break space (often used in “10 000” or before French punctuation) do not: “a b” joined by a thin space is one word.
- Hyphens, slashes, dots, apostrophes, @ and + never split a word: “E-Mail-Adresse”, “and/or”, “help@example.com” and “O’Brien’s” are one word each.
- The en dash and em dash do split, even without spaces around them, and are not words themselves: “late—very late” is three words, and “2–3 days” is three.
- A standalone hyphen, asterisk or arrow at the start of a list line counts as a word. The bullet character • does not.
- Every Chinese, Japanese and Korean character is a word of its own, and so is CJK punctuation such as 。、「」!, along with full-width letters and digits. Korean is counted per syllable, not per space-separated word: “한국어 문장” is five words in Word.
- Thai, which is written without spaces between words, is counted by spaces only, so a whole sentence can be one word.
Persian and Arabic: the half-space problem
In Persian, the parts of a compound like «میخواهم» or «کتابها» are joined by a zero-width non-joiner (ZWNJ), an invisible character that is not a space. Many people type a normal space instead. The text looks almost the same, but the count does not:
| The same Persian sentence | Words in Word | Characters in Word |
|---|---|---|
| Typed with ZWNJ («میخواهم این کتابها …») | 13 | 81 |
| Typed with spaces («می خواهم این کتاب ها …») | 18 | 81 |
Five extra words in a 13-word sentence is a 38% difference. A space always starts a new word, in Word and in our counter alike, so if a limit is strict, check how the text was typed before you argue about the tool. Arabic has a related effect for characters: Word counts every short-vowel mark (ـَ ـُ ـِ) as a character, so a vocalised sentence of 54 visible characters showed as 59.
Character counts differ too
Word counts characters as stored code points and leaves line breaks out. A counter that counts what you see, as ours does, gives the same result on most text but not on these:
| Text | Word (with spaces) | Visible characters |
|---|---|---|
| A seven-line shopping list | 62 | 68, because line breaks count |
| Thai sentence with vowel marks | 42 | 31 |
| Arabic with short-vowel marks | 59 | 54 |
| Persian with ZWNJ | 81 | 76 |
| Text with a flag emoji 🇩🇪 | The flag counts as 2 | The flag counts as 1 |
What we changed after the test
Our Word Counter shows two word counts: Words, split at punctuation, and Words (as in Word), meant to match Microsoft Word. Before this test the second one matched Word on 19 of the 24 texts. It got the dashes, the bullet character and CJK punctuation wrong. We fixed the rules above, and it now matches Word on all 24 texts and on all 79 edge cases.
Which number should you use?
| You need | Use |
|---|---|
| A word limit for a university, journal or client who works in Word | Words (as in Word) |
| A count that does not reward or punish hyphens and links | Words |
| Japanese or Chinese length | Characters, which is how both languages are usually measured |
| German translation pricing by Normseite or Normzeile | Characters with spaces; our counter shows both units |
| A limit on X, a meta description or an SMS | Characters, checked in the tool that enforces the limit |
How we tested
- Microsoft Word, version 16.0.17932 (Office 16), driven through its automation interface so that no number was read off the screen by hand. Each text was put into an empty document and measured with Word’s own statistics function.
- The 24 texts were written for this test: plain prose in English and German, and one text per tricky case (hyphens, numbers, links, dashes, apostrophes, lists, code, emoji, French, Spanish, Portuguese, Japanese, Chinese, Persian, Arabic, Thai).
- The punctuation-splitting counts are the Words figure of our Word Counter in Chrome, which uses the browser’s Unicode word segmentation.
- We did not test Google Docs or LibreOffice. They have their own rules, and we will add them in a later round.
All 24 texts with every count are in one file: download the data (CSV).