Count words, characters and reading time

What this counts

Words, characters with and without spaces, sentences, paragraphs, how many different words were used, the average sentence length, and how long the text takes to read silently and aloud. Below the figures is a table of the twenty-five most-used words, which downloads as a spreadsheet.

The interesting question is not what to count but where one word stops and the next begins. Splitting text on spaces is the obvious answer and it is wrong for a large part of the world: Japanese, Chinese, Thai, Lao, Khmer and Burmese are written without them, so a space-based counter reports a whole paragraph as one word and nobody notices until somebody who writes in one of those languages tries it.

The method

Every count comes from Intl.Segmenter, which is the browser’s own copy of the Unicode segmentation algorithm described in UAX #29. It is not an approximation of a text-analysis library — it is the reference implementation, shipped with the browser, and for the scripts that need one it carries a dictionary.

Words are the segments the algorithm marks as word-like, which excludes spaces, punctuation and symbols but includes numbers. Characters are grapheme clusters, so an accent written as a separate combining mark, or an emoji carrying a skin-tone modifier, counts once rather than twice or four times. That is a different number from the one JavaScript’s own length gives, and it is the one a person would arrive at by counting.

Sentences need one addition to the standard rules. UAX #29 cannot tell that the full stop in “Dr. Smith” is part of a title rather than the end of a sentence — it sees a period, a space and a capital letter, which is the definition of a boundary. ICU solves this with a per-locale suppression list, but the browser API does not expose it, so a short list of twenty abbreviations plus any single initial is applied here instead. Abbreviations that really can end a sentence are left off it deliberately, because merging two real sentences is a worse error than splitting one.

Paragraphs are blocks with a blank line on either side. Single line breaks do not divide one, which matches how prose is written and how every other counter behaves.

Reading times are simple division, and the rates are editable because averages are only averages. Both defaults come from a 2019 meta-analysis of 190 studies: 238 words a minute reading English non-fiction silently, 183 reading aloud.

Before you read on

A thumbs-up emoji carrying a skin-tone modifier. How many characters does this tool count it as?

• AFour — that is what JavaScript reports for it.

  It is, and it is why so many counters disagree with the person using them. Four is the number of UTF-16 code units.

• BOne.

  Yes. One grapheme cluster, which is one thing a reader would point at.

• CTwo — the emoji and the modifier.

  Two code points, but a modifier is not something anybody would count separately.

One. Characters here are grapheme clusters as defined by UAX #29 — so an accent written as a separate combining mark counts once rather than twice, and an emoji with a skin-tone modifier counts once rather than four times. That is a different number from the one JavaScript's own length property gives, and it is the one a person would arrive at by counting. Every figure on this page comes from Intl.Segmenter, which is not an approximation of the Unicode algorithm but the browser's own copy of it.

• Different words4980.3%
• Repeats1219.7%

A worked example

The text the page loads with is two paragraphs and 321 characters:

The quick brown fox jumps over the lazy dog. That one sentence contains
every letter of the alphabet, which is why typographers have used it for
well over a century to show a typeface at work.

Paste your own text over this. The counting happens in this browser tab,
on your machine, and there is no server to send it to.

At the default rates of 238 and 183 words a minute:

Words	61
Characters	321
Characters without spaces	260
Sentences	4
Paragraphs	2
Different words	49
Average sentence length	15.3
Reading time	15 seconds
Reading aloud	20 seconds

61 words at 238 a minute is 15.4 seconds, and at 183 a minute is 20.0. The average sentence length is 61 divided by 4. Forty-nine different words out of sixty-one means only twelve are repeats, and the table shows which:

Word	Times used	Share
the	4	6.56%
over	3	4.92%
to	3	4.92%
a	2	3.28%
is	2	3.28%

These are the same numbers asserted in this tool’s test file, so if the counting ever changes without this page changing with it, the build fails.

What it does not do

It does not read a file, and it stops at 2,000 characters — about one side of A4 — because the text travels in the page link and a link has to survive being pasted somewhere. It does not know what a heading, a list item or a table cell is, so a pasted document with structure in it is counted as the plain text it becomes. It does not do readability grades: those formulas count syllables, and syllable counting is language-specific guesswork of a kind this deliberately avoids. And the reading times are arithmetic, not a prediction — they assume a reader who never pauses, which is not a reader.

It counts what is there and changes none of it. To alter the text itself — title case, sentence case, snake_case — that is a conversion rather than a count, and a different tool.