What Counts as a Word — and Why Counters Disagree

Ask several tools to count the same document and you will often get several answers. None of them is careless. Each draws the line between "one word" and "two words" in a slightly different place, and the differences pile up on hyphens, numbers, links and headings. This guide sets out the rule the text analyzer applies, walks through each edge case with the engine's own results, and explains why — when a limit is strict — the count that matters is the recipient's rather than anyone else's.

The rule: a run of letters and digits

A word is a Unicode-aware run of letters and digits. Internal hyphens and apostrophes (straight or curly) do not end the run; everything else does — spaces, commas, periods, double quotation marks, slashes, colons, and dashes other than the plain hyphen. Nothing is looked up and nothing is judged: a run is a word whether it is "the", a typo, or a year. The same indifference cuts the other way: a hyphen or apostrophe standing alone between spaces is a run of its own and counts as a word, so text set with spaced hyphens counts slightly high. That makes the rule fast and predictable, and it is the same rule for every script. Accented and non-Latin letters are letters, so The café's naïve résumé impressed them. counts as 6 words with no special handling.

Example 1 — a baseline. The quick brown fox jumps over the lazy dog.

9 words · 1 sentence · 44 characters with spaces · 36 without · 8 unique words

"The" appears twice, once capitalised, and the unique count treats both as the same word — which is why it comes in below the total. Characters with spaces counts every Unicode code point, spaces and the final period included; characters without spaces removes whitespace and nothing else, so punctuation stays in.

Example 2 — hyphens and apostrophes. It's a state-of-the-art design, and we can't wait to ship it.

11 words · 11 unique · 51 characters without spaces

"state-of-the-art", "It's" and "can't" are one word each. Replace every hyphen with a space and run the same sentence again: 14 words. Replace the apostrophes instead: 13. That spread, on one ordinary sentence, is much of the story of why word counts differ between tools. Notice also that "It's" and "it" are different entries in the unique count, because the apostrophe is part of the run.

Numbers, decimals and URLs

The rule has no special case for numbers, and that is the point at which it surprises people. A comma or a period is not a letter or a digit, so it ends a run: "2,500" is two words ("2" and "500"), and "3.14" is two words as well. A URL breaks wherever its punctuation falls, so https://example.com tokenizes as three words — https, example, com. None of this is a defect in the counting; it is the same rule applied without exceptions, and it is documented so that you can predict the result rather than guess at it.

Example 3 — a thousands separator and a decimal. The budget is 2,500 dollars, and pi is roughly 3.14.

12 words · 1 sentence · 11 unique

Join the digit groups by deleting the comma and the decimal point and the count falls to 10. The sentence count does not change, because the period inside 3.14 is followed by a digit, not by whitespace, so it is not read as a sentence ending. Number-heavy text is also where the speaking-time estimate runs long: "2,500" counts as two words but is voiced as four.

Example 4 — a URL and an abbreviation. Visit https://example.com for details, e.g. the schedule.

10 words · 2 sentences · 10 unique

The colon, the slashes and the period split the address into three words — https, example, com. "e.g." contributes two one-letter words, "e" and "g" — and, separately, a false sentence break, because its final period is followed by a space. A reader sees one sentence; the engine reports 2. The period inside example.com ends a word run but not a sentence, since it is followed by a letter rather than a space. Single-character tokens like "e" and "g" are excluded from the rankings on the word frequency counter, but they remain in the total word count that density percentages divide by.

Emoji, symbols and the character counts

Emoji are not words. They are neither letters nor digits, so they never start or extend a run; they are counted as characters, one per Unicode code point. The distinction matters because "character" itself has more than one meaning across tools: this engine counts code points, while a counter built on a raw string length counts UTF-16 code units, in which most emoji cost two. Neither figure is wrong, but they answer different questions, and the SMS guide explains where the code-unit figure is the one that decides a segment boundary.

Example 5 — emoji and punctuation. Great work today 🎉🎉 thank you all!

6 words · 34 characters with spaces · 28 without · 25 letters · 3 other characters

The letter breakdown separates 25 letters and 0 digits from 3 remaining non-space characters — the two emoji and the exclamation mark. Counted the way the engine's SMS segmentation does — in UTF-16 units for any text outside the GSM-7 set — the same text is 36 units: the 2-unit gap comes from the two emoji, which cost two UTF-16 units apiece. If a form somewhere tells you a text is longer than this page says, a different counting unit is the likely reason.

Sentences and paragraphs

A sentence ends on a run of ending punctuation — period, exclamation mark, question mark or ellipsis — followed by whitespace or the end of the text, and a fragment must contain at least one letter or digit to count. Paragraphs are blocks separated by one or more blank lines; a single line break does not split one. The sentence rule is deliberately simple, and it has a known cost: an abbreviation followed by a space over-counts. Mr. Smith arrived. registers as 2 sentences, a limitation shared by most automated counters and stated here rather than hidden. The rule also has a known benefit: decimals, version numbers and domain names do not split, because their periods are followed by a character rather than a space.

Example 6 — headings without punctuation. A heading line followed by a line of prose, then a blank line, then another heading line and its prose: Project update / The build passed. Tests are green! and Next steps / Ship on Friday.

13 words · 3 sentences · 2 paragraphs

Count the sentences by eye, skipping the headings, and you arrive at the same 3 the engine reports. The count agrees; what differs is what each sentence contains. "Project update" has no ending punctuation, so it merges into "The build passed", just as "Next steps" merges into "Ship on Friday" — the line break between them is not a sentence boundary. Strip the two heading lines and the prose alone is 9 words in 3 sentences, an average of 3.0 words each; with the headings folded in, the same 3 sentences hold 13 words, an average of 4.3. The blank line, by contrast, is a paragraph boundary, so the paragraph count is 2. This is why readability scores on outlines and slide notes drift: merged headings lengthen the average sentence, and the Reading Ease formula reads that as harder prose. Score running text, not structure.

Unique words: case and word forms

The unique-word count lowercases every word before comparing, so a word that opens a sentence is not counted twice for its capital letter. It does not stem: "run", "runs" and "running" are three entries, deliberately, because merging forms would hide exactly the repetition that the word frequency counter exists to reveal. The same convention carries into keyword density, where a fixed list of 74 common English function words is additionally left out of the ranking but kept in the base the percentages divide by.

Example 7 — case and inflection. Run, run, RUN — the runs and the running never stop.

10 words · 7 unique

"Run", "run" and "RUN" collapse to a single entry; "runs" and "running" do not join it. The em dash is not a hyphen, so it separates words like any other punctuation. If you expected the unique count to be lower, you were assuming stemming; if you expected it to be higher, you were assuming case sensitivity. Each is a reasonable design, and each produces a different number.

Why counters disagree — and which one is right

Every case above is a place where a defensible rule can go either way: whether a hyphenated compound is one word or several, whether "2,500" is a single token, whether a URL is one word or a handful, whether a heading without a period ends a sentence, whether trailing punctuation belongs to a word or stands alone. Two counters that each apply their rule consistently will still part ways wherever the rules differ, and on a long document those partings accumulate into a small, stable gap between the totals. The gap is a property of the rules, not a sign that either tool has slipped; the methodology page states this briefly in its section on why tools disagree, and the examples on this page are the long version. Reading and speaking times inherit whatever gap the word count carries, because both are the word count divided by a population-average pace — see the guides on reading time and speaking time for how those estimates are built.

When an exact figure is contractual — a strict submission limit, a competition cap, a form that rejects on the count — the figure that decides the outcome is produced by the recipient's software under the recipient's rules. No published rule set can promise to reproduce it, and this one does not try. What it offers instead is a rule you can read in full, results you can predict from it, and a clear signal of whether you are close enough to a limit to care. If you are, trim to a margin rather than to the line, and let the recipient's count be the last word.

Frequently asked questions

Is "state-of-the-art" one word or four?

One, here: internal hyphens keep a run whole, so "It's a state-of-the-art design, and we can't wait to ship it." counts as 11 words. A counter that treats hyphens as separators reports 14 for the same sentence. Neither is wrong — they draw the boundary differently, and the gap is where disagreements between tools usually come from.

Why does "2,500" count as two words?

Because a comma is not a letter or a digit, so it ends the run: "2" and "500" are separate words, and "3.14" splits the same way. "The budget is 2,500 dollars, and pi is roughly 3.14." therefore counts as 12 words; a counter that keeps digit groups together would report 10. The period inside 3.14 does not end a sentence, because it is not followed by a space.

Do emoji count as words?

No. Emoji are not letters or digits, so they never form a word; they count as characters, one per Unicode code point. "Great work today 🎉🎉 thank you all!" has 6 words and 34 characters including spaces. Some tools count UTF-16 code units instead, in which most emoji cost two — counted in UTF-16 units, as the engine's SMS segmentation does, this text is 36 units.

Why is my sentence count higher than I expected?

Abbreviations. A sentence ends on a run of . ! ? or … followed by a space or the end of the text, so "Mr. Smith arrived." registers as 2 sentences and "e.g." adds a false break whenever its final period is followed by a space. Decimals and URLs do not split, because their periods are followed by a character, not a space. The over-count is a stated limitation shared by most automated counters.

Why is the unique-word count lower than the word count?

Unique words are compared case-insensitively, so "The" and "the" are one entry: the pangram "The quick brown fox jumps over the lazy dog." has 9 words but 8 unique words. Word forms are not merged, though — "run", "runs" and "running" stay separate, which is why "Run, run, RUN — the runs and the running never stop." has 7 unique words rather than fewer.

Which counter should I trust for a strict limit?

The one the recipient uses. When a limit is contractual — a submission cap, an application form, a competition rule — the count that decides the outcome is the count made at the other end, on their software, with their rules for hyphens, numbers and headings. Use this counter to learn whether you are close enough to care, then leave a margin.

Every number in this guide was computed at build time by the same engine that runs the text analyzer. When you use the tool itself, counting happens locally in your browser — nothing you paste is transmitted or stored. The tokenization and sentence rules are the engine's own, as documented on the methodology page.