Sentence Counter

This page explains the rule the counter above uses to find a sentence boundary, why decimal numbers survive that rule while abbreviations do not, and how to read the words-per-sentence average printed beside the count. Every number in the worked examples below is computed at build time by the same engine that runs in your browser, so the copy cannot drift from the tool. That includes each "reader's count": rather than being typed in, it is the engine's own figure for the same text with the abbreviation's final period removed — or, in the quotation example, with the period moved outside the closing mark. Sentences, paragraphs, and lines are the three units that depend on punctuation and line breaks rather than on the words themselves; the guide Sentences, paragraphs, and lines sets the three conventions side by side and shows where they interact. The guides index lists the rest.

The boundary rule

A sentence boundary is a run of one or more ending marks — period, exclamation point, question mark, or the ellipsis character — followed by whitespace or by the end of the text. The text is cut at every such boundary, and each segment that remains counts as a sentence only if it contains at least one letter or digit. That is the whole rule. There is no list of abbreviations, no grammar, and no lookahead for a capital letter, and three consequences follow directly from its shape.

First, a run of marks counts once. "?!" is one boundary, not two, and the same goes for a string of periods typed as an ellipsis, which the engine treats as a single run. Second, the final segment counts even without a closing mark, because the rule cuts at boundaries and then counts whatever is left, so a trailing fragment that contains a word still registers. Third, whitespace is the trigger, not a line break in particular: a period followed by a space, a tab, or a new line all end a sentence, and a line break with no punctuation before it does not.

The counter cuts after the ellipsis and after the "?!" run, then counts the 3 segments that remain, the unpunctuated tail included. The whitespace requirement also means the counter looks only at the character immediately after the run. A closing quotation mark or bracket sitting between the period and the space hides the boundary, because the character after the period is then the quotation mark, not whitespace:

The period after "Stop" is followed by a quotation mark, so no cut happens there, and the 2 sentences a reader sees are reported as 1. Move the period outside the closing mark — "She said "Stop". Then she left." — and the engine reports 2, on the same 6 words. The methodology page names this as the one construction that runs the other way: the abbreviation error described below only ever pushes the count up, while a closing mark placed after the period pulls it down, and it is spelled out here so the count never surprises you. Stray punctuation, by contrast, is handled by the letter-or-digit test:

The spaced dots produce empty segments between them, none of which contains a letter or digit, so none of them counts. The same test is why empty or whitespace-only text reports 0 sentences rather than one.

Why decimals do not split a sentence

A period inside a number is followed by a digit, not by whitespace, so it is not a boundary. "3.5" in the middle of a sentence is invisible to the sentence rule, exactly as the first period in "a.m." is, and the same holds for version numbers, file names typed with their extensions, and any other dotted token — as long as no space follows the dot.

The passage holds 2 decimals and three ending marks of three different kinds, and the count is 3 — the number a reader would give. Look at the word figure, though. The word rule on this site treats a period as a separator and never as part of a word, so each decimal contributes two word-tokens: "3.5" counts as "3" and "5". The passage has 20 words by that rule; strip the decimal points so that each figure is one token and the engine reports 18 words on the same 3 sentences, which moves the average from 6.7 to 6 words per sentence. On prose with an occasional figure the difference is noise; on a table of measurements pasted as text it is worth knowing about. The What counts as a word guide covers the token rule in full.

Abbreviations: the over-count and its direction

The rule cannot tell the period that ends "Dr." from the period that ends a sentence, because both are followed by a space. The methodology page's standard paragraph shows the effect with a single abbreviation:

A reader sees 2 sentences; the counter reports 3. The extra one comes from "Dr." — period, then space — and from nowhere else: drop that one period and the engine reports the reader's 2. "a.m." adds no phantom, because its first period is followed by a letter and its second coincides with the real sentence end. The pattern generalises. An abbreviation adds a phantom sentence only through its final period, and only when that period is followed by whitespace in the middle of a sentence. "e.g.", "i.e.", and "U.S." therefore each add at most one, never one per dot, and an abbreviation that is the last word of a sentence adds nothing at all, since the phantom boundary and the real one are the same boundary.

Because a phantom boundary can only ever cut a sentence into more pieces, the error is one-directional: abbreviations push the count up and never down. It is also bounded — one per mid-sentence abbreviation — which makes it correctable by hand: count the abbreviations that sit mid-sentence and subtract that many. A short text dense with them shows the extreme case:

4 abbreviations, 4 phantom boundaries: the counter reports 6 sentences where a reader counts 2, and the average falls from 10 words per sentence to 3.3. The reader's figure is the engine's own count of the memo with the final period stripped from each abbreviation, which leaves its 20 words untouched, so the two averages divide the same word count. Subtracting one per abbreviation recovers the reader's count exactly. Two things make this the extreme rather than the norm: the text is short, so every phantom is a large share of the total, and an abbreviation appears in nearly every clause. Most prose has neither property.

How much the error matters at paragraph length

The same phantom weighs far less inside a full paragraph. This passage contains a single abbreviation, "Dr.":

The counter reports 6 sentences against a reader's 5: the raw count is 20% high, where the memo above was 200% high. The average moves from 12.8 to 10.7 words per sentence — a shift of 2.1 — because the 64 words are now divided by a number that is only one off. This is why the average is the more robust figure for judging style. Each phantom boundary is a fixed error of one sentence, while the average divides the total word count by the sentence count, so as the text grows the phantom's share of the denominator shrinks toward nothing. On anything longer than a paragraph an abbreviation or two barely registers in the average, though it still shows in the raw count.

Reading the average

The average is the word count divided by the sentence count, shown to at most one decimal place. It shows a dash until the text contains at least one sentence — one segment with a letter or digit in it — and, as the trail example shows, its words are the engine's word-tokens: runs of letters, digits, apostrophes, and hyphens, so "state-of-the-art" is one word, "it's" is one word, and "3.5" is two. Both inputs follow rules stated on the methodology page, and the average inherits their conventions along with the abbreviation over-count described above.

Read it as a rhythm gauge, not a grade. This page does not publish a target number, because a single figure would hide what matters more: variation. A passage whose sentences are all the same length reads as a drone whatever that length is, and a passage that alternates short statements with longer, qualified ones can carry a high average and still move. The productive use of the number is comparative. Paste a draft and note the average, then paste the paragraph you suspect is the heavy one; if it runs well above the rest, that is where the three-clause sentences are hiding. Compare like with like — an earlier draft of the same piece, or other writing of the same kind — rather than against an abstract standard.

The average also matters because it is one of the two inputs to the Flesch readability formulas, alongside syllables per word: longer sentences lower the reading-ease score and raise the grade level, other things equal. This page reports the average only. The full scores, with the syllable heuristic they depend on, are on our sister site WordCounterPro.ai, which runs the same counting engine byte for byte.

Headings, fragments, and paragraphs

The counter has no notion of a heading. A line without an ending mark is not cut off from what follows it, so a heading folds into the first sentence of the paragraph beneath it, and its words are credited to that sentence:

Relative to a count that treats the heading as a unit of its own, the figure is low by one; relative to a reader who does not count headings as sentences, it is right. Either way, "Quarterly summary" has been absorbed into the sentence that follows it. The whole passage has 9 words and an average of 4.5; the body on its own has 7 words, the same 2 sentences, and an average of 3.5. A heading-heavy outline therefore pulls the count down, or holds it level, while pushing the average up — the reverse of what abbreviations do — and the remedy is the same as for any outline-heavy text: paste body prose rather than headings and bullets when the average is what you are after.

Paragraphs, the fourth figure on the panel, follow a different convention entirely: blocks separated by one or more blank lines, with a single line break leaving a block whole and a line containing only spaces treated as blank. The heading example is 2 paragraphs because of the blank line between the two parts, not because the heading is short. Word processors count paragraph marks instead, so their figure can differ in both directions; the paragraph counter works through those cases, and the sentences, paragraphs, and lines guide sets the three rules beside each other, including the line convention this page does not use.

Empty text, limits, and privacy

Empty or whitespace-only text returns 0 sentences, 0 words, and a dash (—) for the average. The engine accepts up to two million characters per analysis and reports an explicit message rather than a silent wrong answer beyond that. There is no sampling at any size: every keystroke recounts the full text. Nothing is transmitted, either — the counting is local JavaScript on a static page, which you can confirm from your browser's network tab, and a loaded page keeps working offline.

Frequently asked questions

Counting happens locally in your browser — nothing you type is transmitted or stored. See the methodology page for the exact sentence-splitting rules.