Why Microsoft Word, Google Docs, and WordPress count your words differently
Paste the same paragraph into four tools and you can get four different word counts. Across four counting rules calibrated to Microsoft Word, Google Docs, WordPress, and our own baseline, we measured a 52% gap on one 21-word paragraph — because each rule defines "a word" differently.
One paragraph, four answers
Here is a perfectly ordinary English paragraph. It contains one hyphenated compound, three em dashes, one URL, one decimal, one email address, and one slash-joined word:
The well-known editor—writer said yes—or no—to the plan. See https://glyphcount.com/regex-tester/ and 3.14 notes. Don't email support@glyphcount.com about cat/dog data... or do.
Feed it to the four counting engines behind the glyphcount word counter — a baseline rule plus simulations calibrated to Microsoft Word, Google Docs, and WordPress — and you get counts from 21 to 32 words (21, 21, 28, and 32 across the four engines). The gap between the lowest and highest count is 11 words — a 52% difference.
The gap is structural, not a rounding bug. Each engine encodes a different definition of word boundary, and twelve short probes are enough to map precisely where those definitions stop overlapping. One caveat before the numbers: the three vendor columns are simulations — approximations calibrated to each product's documented English counting behavior, labelled as such in the tool itself. The real products can vary by version and language setting, so read the table as a map of the rules, not a lab report on specific app builds.
The five rules that diverge
1. Dashes: em, en, and hyphen are three different decisions
In the simulations, an em dash (teacher—author) splits the word in the Google Docs and WordPress columns but not in the baseline or MS Word columns. An en dash (teacher–author) goes one step further and splits in the MS Word simulation too. A hyphen (well-known) does not split in the baseline, MS Word, or Google Docs simulations; only the WordPress letter-run rule fragments it, turning well-known editor-fixed into four letter runs where the baseline counts two words.
2. Slashes and paired dashes
The Google Docs simulation treats a slash as a hard word boundary, so cat/dog is two words there but one in the baseline. Paired em dashes compound the effect: yes—or not—about reads as one flowing phrase to a human, yet the Google Docs simulation counts 8 words where the baseline counts 6.
3. URLs and email addresses
This is the biggest single divergence in the bake-off. The whole URL line (See https://glyphcount.com/word-counter/ for details) counts 4 / 4 / 7 / 8 across baseline / MS Word / Google Docs / WordPress simulations. The URL alone tells the story more sharply: https://glyphcount.com/word-counter/ counts 1 word in the baseline (edge punctuation stripped, letters inside), 4 in the Google Docs simulation (shredded on every dot and slash), and 5 in the WordPress simulation (each letter run, including the split at the hyphen). An email address shows the mirror image: support@glyphcount.com alone counts 1 in the baseline and 3 letter runs in the WordPress simulation.
4. Numbers, decimals, and symbols
The WordPress simulation ignores digits entirely — not just when they stand alone: under a letter-run rule, a1b splits into two runs and 3.14 vanishes, taking Pi is about 3.14 in two decimals from 7 words down to 6. Currency is harsher — $50 and €40 vanish, cutting that line from 6 to 4.
5. CJK text: the clearest point of agreement
All four engines count 寿司 as one word per character — one of only two rows in the bake-off where every column agrees (the other is the ellipsis probe). With no spaces to split on, per-character counting is the standard approach for unsegmented CJK text, and every rule in our set follows it for this probe.
The bake-off, all twelve probes
Every number below was computed by the four engines at build time — the same code you run when you use the word counter. 10 of 12 probes diverge across engines.
| Case | Text | Baseline | MS Word sim | Google Docs sim | WordPress sim |
|---|---|---|---|---|---|
| em dash | The teacher—author meeting ran late | 5 | 5 | 6 | 6 |
| en dash | The teacher–author meeting ran late | 5 | 6 | 6 | 6 |
| slash | We asked whether cat/dog owners differ | 6 | 6 | 7 | 7 |
| paired em dashes | I said yes—or not—about the plan | 6 | 6 | 8 | 8 |
| decimal number | Pi is about 3.14 in two decimals | 7 | 7 | 7 | 6 |
| hyphenated words | A well-known editor-fixed style guide | 5 | 5 | 5 | 7 |
| apostrophe | Don't stop believin' in counts | 5 | 5 | 5 | 6 |
| currency symbols | It cost $50, or €40, roughly | 6 | 6 | 6 | 4 |
| email address | Mail support@glyphcount.com anytime | 3 | 3 | 3 | 5 |
| URL | See https://glyphcount.com/word-counter/ for details | 4 | 4 | 7 | 8 |
| CJK mixed text | I ate 寿司 and ramen today | 7 | 7 | 7 | 7 |
| ellipsis | Wait... what happened exactly | 4 | 4 | 4 | 4 |
Baseline keeps em dashes and hyphens inside a word and requires at least one letter or digit; the MS Word simulation additionally splits on en dashes; the Google Docs simulation adds slashes and URL fragmentation; the WordPress simulation counts letter runs only. Full rule set: the counting methodology.
Where a 52% gap actually costs you
For casual writing, none of this matters. It matters when a number on screen is a contract:
- Freelance billing per word. Using our paragraph's ratio as an illustration: a text that measures 1,000 words under the baseline rule would measure roughly 1524 under the highest-count rule in the set. These are illustrative extrapolations from one sample, not conversion factors — the practical takeaway is to state the counting rule in the contract, not on the invoice.
- Hard limits. Abstracts, grant sections, and personal statements with a word ceiling punish whichever engine you happened to paste from. Count in the tool whose rule the recipient uses.
- Platform limits are a different measuring problem. SMS and X limits are character-based, not word-based, and both use weighted counts: one non-GSM character switches a single-segment SMS from 160 to 70 units (multipart segments hold 153 GSM-7 or 67 UCS-2 units each), and on X a CJK character or emoji counts as 2 of 280. The character counter measures those in units for exactly that reason.
Which count should you trust?
The honest answer: the one whose rule you can read. The glyphcount word counter shows all four engines side by side instead of silently picking one, and publishes the exact rule behind each metric on the methodology page. When a count has consequences, paste the text, compare the engines, and quote the one your counterparty uses.
Frequently asked questions
Which word count is the "correct" one?
There is no universal correct count — only a declared rule. The baseline keeps em dashes and hyphens inside a word and requires at least one letter or digit; the MS Word simulation splits on en dashes; the Google Docs simulation also splits on slashes and fragments URLs; the WordPress simulation counts pure letter runs. For a paid agreement, use the rule both parties accept; for a submission with a word ceiling, use the counting method the recipient requires.
Are these simulations exactly what Word, Docs, and WordPress do?
They are approximations calibrated to each product's documented English counting behavior — that is why the word counter labels them "simulation". The real products can differ by version and language setting (WordPress, for example, supports locale-dependent counting strategies), so treat per-tool counts on a specific document as that tool's answer, and treat this table as a map of how the rules differ.
Why does the WordPress simulation count numbers lower but fragment some other tokens?
Its letter-run rule matches only continuous runs of A–Z letters, plus one per CJK character. Digits never count on their own — a1b splits into two runs — so $50 and 3.14 vanish, cutting the currency line from 6 to 4. But fragmentation can also raise the count: support@glyphcount.com is one baseline token yet three letter runs, and the email row rises from 3 to 5. Disappearing numeric tokens push the count down; split punctuation pushes it up.
Does the discrepancy matter for SEO meta descriptions?
Only at the margin. Google states meta descriptions have no fixed length limit — displayed snippets are truncated to fit the device, with roughly 150–160 characters as the practical desktop ceiling. Word-count rules rarely move that. Where the gap really bites is freelance billing per word and academic abstracts with a hard word limit.
How were the numbers in the table computed?
Every figure is generated at build time by the same versioned code that powers the glyphcount word counter, and automated tests fail the build if the table ever disagrees with the engines — the article cannot silently drift from the tool.