Word counting rules: why Word, your editor and your limit disagree

Is "1,200" one word or two? Does a hyphenated compound count once? Word counts are conventions, not measurements, and different tools apply different conventions. This guide states the common rules, shows where counters diverge, and tells you how to hit a limit safely when the counter is not yours.

Ask three word processors to count the same essay and you may get three numbers, all of them defensible. Counting words sounds like measuring length, but it is really applying a definition — and the definitions differ on hyphenated compounds, numbers, punctuation clusters, URLs and what to do with headers, footnotes and captions. When a submission limit, an exam rubric or a translation quote rides on the number, "approximately right" stops being good enough.

The good news: the divergences are systematic, small and predictable once named. This article states the whitespace rule most simple counters use, catalogues the edge cases where counters part company, and gives a margin strategy for writing to a limit enforced by someone else's counter.

The whitespace rule, and what it does to real text

The simplest honest counter splits text on whitespace and counts the pieces. Under that rule, state-of-the-art is one word, 1,200 is one word, an em dash standing alone ( — ) is, rather embarrassingly, one word, and don't is one word. Microsoft Word's counter is close to this but not identical: it treats some punctuation-joined forms differently, counts certain symbol strings as words, and — importantly — its status-bar count includes text boxes, footnotes and endnotes, while some "body only" counters exclude them.

Scope: the invisible half of every limit

More submissions are lost to scope than to arithmetic. A journal's "6,000-word limit" may include the abstract, exclude the references, and say nothing about tables — three scope decisions that can swing the total by a thousand words. Examination boards publish their own rulings (footnotes of citation usually out, footnotes of content usually in). Translation pricing counts source words by the CAT tool's segmentation, which is yet another definition. The rule for writers is unglamorous: find the organiser's stated scope before you trim a single sentence, and if none is stated, ask or assume the strictest plausible one — everything in.

Structure your document so scope is easy to apply: keep the abstract, body, tables and reference list as separable blocks, and you can produce any scope's count in seconds rather than by surgery.

Counting like a professional: characters, reading time and other measures

Words are not the only unit in play. Character counts (with or without spaces) govern social posts, metadata fields and SMS segments — and "without spaces" exists because some historical billing and typesetting systems used it. Reading-time estimates assume roughly 200 words per minute for adult prose read silently, and speaking time about 130–150 words per minute aloud; a counter that reports both is translating your count into the units audiences actually experience. For editing, the most useful secondary statistics are sentence count and average sentence length: a 4,000-word piece averaging 28 words per sentence reads very differently from the same count averaging 16, and no word limit will warn you about it.

Writing to the limit without gambling

1. Identify the enforcing counter and its scope — the submission system's own counter, where one exists, outranks every tool you own.
2. Draft to 97–98% of the limit by your counter. The 2–3% buffer absorbs definitional differences on hyphens, numbers and dashes.
3. Reduce edge cases in your own text: fewer free-standing dashes, fewer numeric strings, consistent hyphenation — each shrinks the variance between counters.
4. Final check in the destination: paste into the actual submission box or the specified processor and read its number, not yours. If it is over, trim there, where the truth lives.

Mistakes and pitfalls

Trimming prose to satisfy a counter that excludes references

Cutting good body text while a two-page bibliography sits outside the count is the classic self-inflicted wound of word limits. Establish scope first; edit second.

Padding to reach a minimum with edge-case tricks

Splitting compounds and sprinkling free-standing dashes inflates naive counters and fools nobody reading. Minimum limits exist to force substance; markers spot padding instantly and the count discrepancy invites scrutiny, not credit.

Comparing counts between a PDF and the source document

PDF text extraction inserts and removes spaces around ligatures, line breaks and hyphenation, shifting any count made from it. Always count from the editable source.

Common questions

Do contractions count as one word?

Yes, under every common convention — don't, we're and o'clock are single words. If a piece reads as though it needs them to be two, the definition being applied is not one any mainstream tool uses.

How are words counted in Chinese, Japanese or Korean text?

Whitespace rules collapse entirely, because these scripts do not separate words with spaces. Counters switch to characters or segmentation dictionaries, and limits for such texts are usually given in characters. Never apply an English word count to a CJK document.

My counter says 998 and the portal says 1,004 against a 1,000 limit. What happened?

Most likely a handful of edge cases: the portal splits numbers on commas, counts your free-standing dashes differently, or includes a header block you excluded. The fixes are the buffer strategy above — and, immediately, trimming six or more genuine words in the destination itself until its counter passes.

To wrap up

A word count is a definition wearing a number's clothes. Learn the whitespace baseline, the handful of edge cases where definitions split, and above all the scope of the limit you are writing to — then keep a small buffer and the arithmetic of other people's counters stops being a risk to your work.