Word Counter
Words, characters, sentences, reading time and the phrases you repeat most.
- Words
- 0
- Characters
- 0
- Without spaces
- 0
- Sentences
- 0
- Paragraphs
- 0
- Reading time
- —
The arithmetic
Inputs shown, not hiddenReading time: —
Flesch ease: —
The reading speed is fixed at 225 words per minute. Adults reading silently from a screen run somewhere between 200 and 250, technical material with tables and code runs slower, and skimming runs two to three times faster.
Counting rules
- Words. Trim the text, then split on runs of whitespace. Two spaces between two words is one separator, not an empty word.
- Sentences. A run of . ! ? followed by a space or the end of the text. A closing fragment with no terminator counts as a sentence of its own.
- Paragraphs. Blocks separated by a blank line. A line holding nothing but spaces counts as blank.
- Characters. Counted twice, so the difference between the two figures is the whitespace you have used.
Repeated words
Density is a word's share of the whole text: 41 mentions inside 2,140 words is 1.92%. The divisor never changes when stopwords are hidden, so the percentages still add up against the full length.
The syllable count is an approximation. There is no pronunciation dictionary in a browser, so the counter reads vowel groups: it scores business at 3 syllables where a dictionary says 2, and colonel at 3 where the answer is 2 again. Expect a few percent of error on ordinary prose and more on names. Flesch is built from that count, so it inherits the error — a point or two, not ten.
Reading time is a floor, not an estimate
The figure is built from one assumption — a silent adult reading prose from a screen — and it holds well for exactly that. Prose reads at the stated rate. Almost nothing else does. A reader moving through a code block is parsing syntax rather than recognising words, and one line of it can cost as much as a paragraph. A table has to be entered, held and compared row by row. A numbered list is slow for a different reason: the reader keeps checking the ordinal to remember where they are. Footnotes and parentheticals break the run and cost a re-entry every time.
What that means in practice is that the estimate is a good relative measure and a poor absolute one. Two drafts of the same piece compare fairly, because the same kinds of thing are in both. A technical document with a third of its words inside code and tables will take noticeably longer than the number suggests, and treating the figure as a floor rather than a promise is the honest way to use it — especially when the number is being quoted to somebody deciding whether to read the piece at all.
Rank your own drafts against each other rather than against a scale. The readability figure is a number derived from two counts, and the counts are approximations. It is at its best when it answers a question about your writing — which paragraphs got away from you, whether a revision helped — and at its worst when it is read as a grade level that something external awards. Nothing in it knows whether what you wrote made sense, or whether the sentences that came out short are short because they are clear or because they were cut badly.
The readability score measures two ratios
Every formula of this kind is built from average sentence length and average syllables per word, and almost nothing else. Once that is clear, its behaviour stops being mysterious:
Flesch Reading Ease
206.835
- 1.015 x (words / sentences)
- 84.6 x (syllables / words)
Two texts with the same average sentence length and the
same average syllable count score identically — one of them
lucid, the other meaningless. The formula has no term for
whether the sentences connect, whether the words are the
right ones, or whether the reader already knows the subject.The sentence-length term is doing most of the work in ordinary writing, which is why the score responds so sharply to breaking a long sentence in two. That is not a trick; long sentences really are harder to hold, and the reader who has to keep the subject in mind until the verb arrives is spending attention on structure instead of meaning. But it is also why the number can be gamed by chopping prose into fragments, and why a passage of very short punchy sentences can score well while reading as a list of assertions with nothing joining them.
The syllable term is the weaker half. It is standing in for how familiar the vocabulary is, on the assumption that longer words are rarer ones — true on average, and wrong often enough to matter in any text with technical vocabulary in it. A writer who uses the right term for the thing rather than a plain word that means approximately the same thing is penalised for being precise, which is the opposite of what a reader of a technical document wants.
The phrases you repeat are where the editing is
A list of the phrases a text uses most is not a list of errors. Most of what it shows is your own vocabulary coming back, and a written voice depends on some of that. What it does show is the habits you cannot hear. A phrase that turns up five times in a page is almost never deliberate, and it is invisible while you are writing because each instance was written on its own and read on its own.
Two patterns are worth looking for specifically. The first is a repeated phrase at the start of paragraphs and sentences — a sign that the transitions are all doing the same job with the same words rather than doing the job the paragraph needs. The second is a pair: a pair of words that keep arriving together in the same order, which is the fingerprint of a construction you have adopted rather than chosen. Both are easy to fix once seen, and neither is visible at all from inside a single paragraph.
Cutting is a structural job
Trimming to a word count by tightening sentences is the slowest way to lose words and the one that costs the most. Most of the removable material in a draft is not inside the sentences; it is the material that should not be there at all. A section that repeats a point made earlier, a paragraph of context the reader will not need until later, an explanation of something the audience already knows — those come out whole, and each one is worth more than a dozen small edits.
The order that works is to cut at the largest unit first. Take out a section, then a paragraph, then a sentence, and reach for individual words last. Doing it the other way round produces a text that has been tightened everywhere and is still too long, and by then every remaining sentence has been polished so hard that removing it feels like waste. Where the length is a hard limit, it is worth measuring the pieces rather than the whole: paste a section on its own to see what it is costing, and the one that is out of proportion tends to be obvious.
Reference
What counts as a word, and who decides
| Token | Counted as | Why counters disagree |
|---|---|---|
| mother-in-law | One word | A counter that treats the hyphen as a boundary reports three, and nothing in Unicode or in any style guide says which answer is the correct one. |
| word—word | One word here | A split on whitespace gives one token; a counter that also breaks on punctuation gives two. On prose with long dashes in it, this row alone moves the total by about a percent. |
| don't, it's | One word each | A counter that breaks on the apostrophe reports two per contraction, which inflates any draft that leans on them. |
| 2026 | Yes | A bare number is counted by nearly everything. The exceptions are tools set to count alphabetic tokens only, which is rare. |
| £4.50 | One word, usually | Some counters separate a currency symbol from its digits and report two; a few count the symbol on its own as well. |
| a lone # or % | No | A token with no letter and no digit is skipped by most counters. The ones that count it are counting tokens rather than words, which is a different job. |
| and/or | One word | Break on the slash and it is two; break on the conjunction as well and it is three. No rule settles it, so the same six characters come out three different ways. |
| a run of whitespace | A separator, not a word | The baseline nearly every counter uses: a word is any run of non-space characters, and the gap between two runs is never a token. Tools that also split on punctuation are the ones that part company with it. |
| Chinese text | One word per paragraph | Chinese does not put spaces between words, so a whitespace rule counts a whole paragraph as a single word. Telling the words apart needs a dictionary, which is why counts for CJK text vary so much between tools. |
Read by a browser, an office suite and a submission form, the same paragraph comes out differently before anyone has edited a word. The rules in the right-hand column are why, and a counter that follows a different rule is not broken.
No standard defines this. Unicode decides where a character boundary falls and where a line may break; it does not decide what a word is, and no style guide fills the gap. Every counter therefore makes its own choices about the hyphen, the apostrophe, the dash and the slash, and those choices are the whole reason counters reading the same page return different totals. A gap of a few percent is not a fault in either of them.
When the gap is large, punctuation is rarely the cause. Large gaps come from text the counter cannot see at all: a text box, a comment, a footnote, a tracked change, a header that sits outside the body. Those can move a total by hundreds of words. If your figure and a colleague's are far apart, find out what one of you did not copy before you argue about hyphens.
One habit is worth keeping. A form that asks for characters nearly always means characters with spaces, and in ordinary English prose that figure runs about a sixth higher than the without-spaces one. Check which is wanted before you cut a paragraph to fit.
Questions
Counting, answered plainly
Why does my word processor disagree with this count?
Different rules produce different totals. Most editors split on whitespace too, but they treat a lone dash, a bullet character and a stray line number as words in their own right, and some count hyphenated compounds as two. Expect the two figures to differ by a handful on a long document, and by more on a document full of lists and tables. If a publisher has given you a limit, ask which rule they use before trimming to hit a number.
What does the reading time assume?
225 words per minute, silence, and a reader who is paying attention. That is the middle of the range quoted for adults reading from a screen. A 2,000-word article comes out at 8 min 53 s. Nothing here knows about tables, footnotes or code blocks, all of which are much slower per word, so treat a technical document's figure as a floor rather than an estimate.
Can I trust the Flesch score to two decimal places?
No, and the tool does not offer them. The syllable count is a heuristic, and the sentence count is a heuristic, and the score is built entirely from those two. It is useful for ranking your own drafts against each other, and for spotting the paragraph where the sentences got away from you. It is not a grade level, and it does not know whether what you wrote made sense.
Does it count the words in headings, captions and lists?
Everything you paste is counted as one run of text. A bullet point, a caption and a table cell are words like any other, because the tool has no idea which is which. That is usually what you want for a length limit that covers the whole piece, and wrong for a limit that applies to body copy alone — in that case, paste the body copy alone.
Do the numbers change if I paste something confidential?
The text stays in the box you pasted it into. There is no account and no upload step, and the page has nothing to send and nowhere to send it — open your browser's network panel while you type and watch it stay empty. Paste a contract, a medical letter or an unpublished manuscript and it is counted as quietly as anything else.
Will this match a character counter for a social post?
Not exactly. Here, a character is a UTF-16 code unit, which is what JavaScript's length reports: an accented letter is one, most emoji are two, and a combining accent adds another. Social platforms weight their own ranges instead — CJK characters and emoji usually cost two against a limit — so if you are writing to a hard cap, leave a margin rather than landing on the boundary.
Do you count characters with or without spaces?
Both are shown, as Characters and Without spaces, and the difference between the two figures is the whitespace you have used. A form that asks for characters almost always means the first one, with spaces. In ordinary English prose the gap comes to roughly a sixth of the total, so a 6,000-character limit leaves you about 5,000 characters of actual writing once the spaces between words are paid for.
Is there a limit on how much I can paste?
Nothing in the page enforces one — the text area takes whatever you give it, and the count runs over the whole thing. What gives out first is the browser rather than the counter: past a few hundred thousand characters the tally starts to lag behind your typing and the page itself feels slow. At that size, count the document in sections and add the totals together.
Can I count the words in a PDF?
Not directly. A PDF stores drawing instructions rather than a clean run of text, so you have to select the words in a viewer and copy them out, or put the file through something that can read its text layer. A scanned PDF has no text layer at all: the page is a picture of words, and until optical character recognition has run over it there is nothing for a counter to read.