Last year I helped a senior labmate check a manuscript before submission. He kept saying the main text was "about twenty thousand characters", and Word's status bar showed 19876, so I entered that number in the word-count field of the submission system. A week later the editor returned it with a very specific reason: the journal required no more than twenty thousand characters, and the count including punctuation and spaces was already over twenty-three thousand. That was the first time I realized that "word count" has at least four or five definitions, and submission guidelines usually assume the reader knows which one is meant.
1. First work out what the other side is actually counting
Chinese counts by character and Western languages count by word, so mixed text splits the counting rules: the same sentence produces completely different numbers under different conventions. This sentence comes from the abstract at the time (three full-width punctuation marks, two half-width symbols, three spaces, and two continuous Western strings each counted as one word):
结论：样本量约 210 户，占总体的 12.6%。
| Convention | Count for this sentence | What it counts |
|---|---|---|
| Chinese characters | 11 | Han characters only |
| Word count (this article's convention) | 13 | Han characters one by one; strings such as "210" and "12.6%" each count as one word |
| Character count (excluding spaces) | 22 | Han characters, punctuation, digits and letters counted individually, spaces removed |
| Character count (including spaces) | 25 | The previous item plus the spaces |
| Paragraphs / lines | 1 / 1 | Paragraphs split on Enter, lines split on line breaks |
One caveat first: 13 is this article's convention, obtained by counting Han characters one by one plus runs of Western text; editors do not guarantee that the same name means the same value, and whether Word counts full-width punctuation in "characters" varies by version. Treat field names as clues, not definitions.
25 and 13 differ by nearly a factor of two. If the guidelines say "characters including spaces" and you submit by "word count", your own number will deceive you; that was the entire reason for my rejection, and it had nothing to do with the content.
2. Four sources of mismatched numbers
Once a manuscript travels between Word, a web editor and a submission system, differences are almost inevitable, and they usually come from these places:
- Whether punctuation counts. Any "character count" convention includes full-width punctuation and half-width symbols; some guidelines' "word count excluding punctuation" is a separate item and has to be counted on its own.
- Soft line breaks. Shift+Enter produces a line break, not a paragraph break, so line count exceeds paragraph count. Copy a passage out of a typeset PDF and often the whole passage is one or two paragraphs but occupies a dozen lines.
- Invisible spaces. Chinese copied from a PDF or a web page often has half-width spaces inserted between characters. The mechanism deserves spelling out: Word already counts CJK characters one by one, and its statistics dialog lists "Chinese characters and Korean words" as a separate item, so spaces do not push that item up; what they push up is "characters (with spaces)". My sample's 3 spaces correspond exactly to the 25 minus 22 difference. What really multiplies the number is whitespace-tokenising counters and plagiarism checkers: they treat a space as a word boundary, so once Han characters are split apart, each becomes a word of its own and the word count jumps. So when "the word count is clearly over the limit", suspect this first.
- Whether ancillary parts are included. Word's word-count dialog has a checkbox for "include textboxes, footnotes and endnotes", and ticking it or not makes a difference of hundreds of words; whether figure titles, footnotes and references count as main text can only be settled by the wording of the guidelines.
There is one more detail about paragraph count: the status bar shows only the word count by default, and you have to right-click it to display paragraph and line counts, so you need not open the statistics dialog every time.
3. My current fixed routine
- Start with the wording of the guidelines themselves: does it say "word count", "number of characters" or "length". In English submissions word count is usually tokenised on spaces; in Chinese submissions it is counted by Han character. The two are not interchangeable.
- Export the main text separately as plain text and remove headers, footers, author details and template placeholders so they are not counted along with it.
- Paste the whole piece into Word Count and see everything at once: Chinese character count, character count with and without spaces, lines and paragraphs are all there, and conventions that exclude spaces and punctuation can be read directly; it is steadier than opening three programs and counting each way.
- Reserve headroom against the strictest convention. My habit is 90% of the limit: if 20,000 characters are required, I stop at 18,000, because the other side's system usually counts more coarsely than your local one (including table text and figure captions).
- After cutting, paste the before and after versions into Text Diff and read line by line to confirm that what went was padding and not main text damaged by mistake, and check at the same time whether the change in length meets the target.
These typical magnitudes for common requirements can serve as a reference before writing (always defer to the current official guidelines; journals differ widely):
| Type | Common length | Usually counted as |
|---|---|---|
| Chinese core journal paper | 6000-10000 characters | Chinese characters including figures, tables and references |
| Degree thesis (undergraduate) | main text 8000-15000 characters | Main text, excluding appendices |
| Chinese abstract of a paper | 300-500 characters | Han character count |
| English abstract / conference abstract | 200-250 words | Word count tokenised on spaces |
| Keywords | 3-5 | Number of items, no character count |
| WeChat public account article | 1200-2000 characters | Character count from a third-party editor (varies by vendor) |
4. A few corners that are easy to hit
Character count is not byte count. In UTF-8 a Chinese character usually takes 3 bytes, so for limits such as an API field of varchar(255), a database column width or a form maxLength, whether the limit is measured in bytes or characters can make the same Chinese text pass or overflow. When writing a data table specification, state the unit.
Plagiarism reports count differently from Word. Most plagiarism systems count characters, and the system decides how to treat punctuation and line breaks, which is why schools usually write "the plagiarism system's result prevails". Do not try to convert Word's number; the two conventions are not comparable.
Using too much jargon is also a problem. I run Text Frequency Counter over it as a matter of course; if a concept word appears dozens of times, that is usually a sign the same sentence is being written again and again, and cutting those passages saves the most characters without damaging the argument.
The WeChat backend does not give a character count. Its built-in editor has no statistics, so most people estimate from a third-party editor's character count, and conventions differ; the same article can differ by hundreds of characters between editors. If the limit really matters, fix on one convention and count to the end, instead of picking whichever editor shows the friendlier number.
5. Common questions
Q: The guidelines say "within 8000 characters"; which number should I use?
Use the convention least favourable to you: first check whether it says "including references and figures"; if it does not say so, assume it does, and leave 10% headroom on a character count that includes punctuation. Better to write eight hundred characters fewer than to gamble on the other system's convention.
Q: Why do two programs give different word counts?
The difference is almost always in four things: spaces, punctuation, soft line breaks and ancillary parts such as footnotes and text boxes. Different numbers are normal; the key is to confirm which convention the recipient uses, then use only that one throughout.
Q: Does the word count of an English manuscript include references?
Journals differ. Some state "main text only", others require table notes to be reported as well. The submission system usually has separate fields; fill each according to its label and do not substitute the whole-manuscript count.
Q: Why does Chinese pasted from a PDF give an absurdly high count?
Eight times out of ten there are half-width spaces between characters. Word's Chinese character count does not change, but whitespace-tokenising counters and plagiarism systems treat a space as a word boundary, so every Han character becomes a word and the number jumps. Clear the spaces before counting, or read the Chinese character count item directly.
Q: Is it normal for paragraph count and line count to differ?
Yes. Only a hard return starts a new paragraph; Shift+Enter only breaks the line. In typeset material such as poetry, lists and song lyrics, lines often far exceed paragraphs; report whichever item the guidelines ask for.
After that rejection was revised, the editor replied in an email, "the length is compliant this time." Looking back, the real problem was not that I had written too much but that I used a number I had never verified to satisfy someone's one word, "length". So now anything with a length limit, whether a paper, marketing copy, a contract clause or subtitles, gets one question first: is it counting Chinese characters, words, or characters including spaces and punctuation. Ask that one question and most rework disappears.