Word Counter

Why splitting on spaces is the wrong way to count words

Nearly every word counter on the web works the same way: trim the text, split it on runs of whitespace, count what falls out. For English prose it is right often enough that nobody checks. For everything else it quietly fails.

Chinese, Japanese and Thai are written without spaces between words. A four-hundred-word Chinese article contains no spaces at all, so a whitespace split reports it as one word. That is not a rounding error, it is the tool being unusable for a large fraction of the world's writing. In the other direction, a whitespace split gets English right partly by luck: it happens to keep "don't" together because there is no space in it, and it happens to keep "1,200" together for the same reason, but it also counts an em dash flanked by spaces as a word and counts "word." and "word" as different tokens if you ever go on to compare them.

The Unicode Consortium publishes an algorithm for this, and browsers implement it as Intl.Segmenter. Ask it to segment by word and it returns every piece of the string with a flag saying whether the piece is word-like, which is exactly the question being asked. It knows that an apostrophe inside a word does not break it, that a comma between digits does not either, and where the boundaries fall in scripts that do not use spaces. That is what this counter uses.

The reading speed everyone gets wrong

Reading-time estimates on the web are almost universally based on 200 to 300 words a minute, usually 250, and the figure is usually inherited from another site rather than from any measurement. It matters because a reading time that is systematically wrong is worse than none: it teaches people to ignore it.

The best available number comes from Marc Brysbaert's 2019 meta-analysis in the Journal of Memory and Language, which pooled 190 studies covering more than 18,000 participants. For silent reading of English non-fiction the mean is 238 words per minute, with a range across studies of roughly 175 to 300. For reading aloud it is 183. Fiction is read faster than non-fiction, and technical material slower than both.

So this tool shows 238 for silent reading and 183 for reading aloud, and says which is which. If you are timing a talk or a voiceover, the aloud figure is the one you want, and the difference is not small: a thousand-word script is about four minutes read silently and five and a half spoken.

Counting sentences is a heuristic, and saying so

There is no reliable way to count sentences. Splitting on full stops, exclamation marks and question marks is the obvious approach and it overcounts immediately: "It cost 3.50" becomes two sentences, "Dr. Smith arrived" becomes two, and an ellipsis becomes three.

The common cases are worth handling and are handled here. A full stop directly between two digits is a decimal point. A run of terminators is one ending, so "Really?!" and "Wait..." break once rather than two or three times. A stop after a single capital letter is an initial, which keeps "J. R. R. Tolkien" in one piece. A stop after a word on a list of common abbreviations, Dr, Mr, approx, etc, vs and about twenty others, is not an ending. And a terminator not followed by whitespace is not an ending either, which catches most decimals the first rule misses.

That gets it right on ordinary prose and it will still be wrong sometimes, on an unusual abbreviation or a language whose conventions differ. The honest thing is to present it as close rather than exact, which is what the page does.

The limits worth checking against

Most people counting characters are counting against a specific ceiling, so the common ones are shown as bars rather than left for you to remember. A meta description is rendered up to roughly 160 characters on desktop and less on mobile, though the real limit is pixel width rather than character count and Google rewrites the description on most queries anyway. A page title gets about 60. A post on X gets 280, though X counts some characters as two and links as a fixed length regardless of their real one. A single SMS is 160 characters in the GSM alphabet and drops to 70 the moment you include one character outside it, which is why a message with one emoji in it can suddenly cost twice as much to send.

Characters here are counted by code point, so an emoji counts once. That is the intuitive answer and it disagrees with several platforms, which count the two UTF-16 units an emoji occupies, or in the case of some emoji built from several joined code points, considerably more. Where a limit really matters, check against the platform itself.