Word Counter

What counts as a word

For languages written with spaces, a word is a run of characters between whitespace, so hyphenated compounds count as one and contractions are not split. Punctuation attached to a word does not create another.

Japanese, Chinese and Korean are the interesting case, because they are written without spaces between words. Splitting on whitespace would report a single word for an entire essay. Each CJK character therefore counts as one word, which is the convention both Microsoft Word and Google Docs follow, and which makes the number comparable to what an editor or a client is likely to be using. It is a convention rather than a linguistic truth - a two-character compound is one word by any reasonable definition - so for CJK text the character count is usually the figure you actually want, and it is shown alongside.

Mixed text is handled by applying both rules: Latin runs are counted as words, CJK characters individually.

Reading time, and how much to trust it

The estimate uses 238 words per minute for silent reading of English prose, the central figure from Brysbaert's 2019 meta-analysis of some two hundred studies. CJK text is measured in characters per minute instead, because words are not the unit anyone reads in.

Treat the result as an order of magnitude, not a measurement. That same meta-analysis found individual rates ranging from roughly 175 to 300 words per minute for ordinary prose, and material changes it more than the reader does: dense technical writing, code samples, tables and unfamiliar vocabulary all slow reading substantially, while a familiar news article speeds it up. A reading time on a blog post is a rough courtesy to the reader, not a promise.

The limit bars

Four ceilings are shown because they are the ones people routinely write against: 280 characters for a post on X, 160 for a meta description, 60 for a title tag, and 160 for a single SMS.

Each has a caveat worth knowing. X counts characters in Chinese, Japanese, Korean and a few other scripts as two, so a Japanese post reaches the limit at around 140 characters, and it measures links at a fixed length regardless of how long the URL actually is. Meta descriptions and title tags are not hard limits at all but truncation points, and search engines truncate by pixel width rather than character count, so a title full of capitals and wide letters is cut sooner than the number suggests. SMS fits 160 characters only in the GSM 7-bit alphabet; a single character outside it, an emoji or an accented letter included, switches the whole message to 16-bit encoding and the limit drops to 70.

Most used words

The frequency list exists to show what a piece of writing actually dwells on, which is why very common function words are filtered out - articles, prepositions, pronouns and auxiliaries would otherwise occupy every slot in the list on any text in any genre. Single letters are dropped for the same reason.

Words are lowercased before counting, so a term is not split between its capitalised and uncapitalised forms. There is no stemming: run and running are counted separately, because collapsing them requires guessing and the guesses are wrong often enough to be misleading. CJK text is counted per character here too.

Length conventions in publishing and academia

Word counts carry real weight in editorial work. A journalism commission is specified in words and filed to within a few per cent of it. Academic assignments usually allow ten per cent either side, and many institutions exclude the bibliography and appendices but include footnotes, which is the detail that catches people out.

Book-length categories are roughly settled by convention: flash fiction under a thousand words, short stories to seven thousand and change, novellas to about forty thousand, and most commercial novels between eighty and a hundred thousand. Publishers care because paper and printing costs scale with length, and because reader expectations for a genre are surprisingly narrow.

Online, the useful figures are smaller. A blog post of eight hundred to fifteen hundred words covers most subjects properly without padding, an email newsletter tends to lose readers past five hundred, and a landing page does its work in far fewer than either. Length is not quality, but it is a constraint worth knowing before you start rather than after.

Readability, and what the formulas measure

Readability scores such as Flesch-Kincaid are arithmetic over sentence length and syllable count. They correlate with difficulty well enough to be useful and they measure nothing about whether the writing is clear: a sentence of short familiar words in the wrong order scores well and communicates badly.

Where they genuinely help is as a warning light. A jump in average sentence length usually marks a paragraph where you lost the thread, and an unusually high score in a technical document often means jargon has been replaced with vagueness rather than explanation.

The counts here support the same kind of review without pretending to be a score. Sentence and paragraph counts show the rhythm of a piece; the repeated-word list shows what it actually dwells on, which is frequently not what the writer intended. Reading a paragraph aloud remains the best test anyone has found, and it costs nothing.

Where character counts decide what people see

Search results truncate a title at around 600 pixels, which is roughly 55 to 60 characters in a typical font - and because the cut is by width rather than by count, a title full of capitals or wide letters disappears sooner. Meta descriptions are cut around 155 to 160 characters, and search engines frequently ignore the description entirely and generate their own from the page.

Social previews have their own limits, and the pattern is the same: the first 50 to 60 characters of a title do the work, and everything after that is context rather than message. Email subject lines are cut by the client, often near 40 characters on a phone.

The practical approach is to front-load. Put the distinguishing words first so the sentence still means something when it is cut, and treat everything past the limit as a bonus for anyone who sees the full text. Checking the count before publishing is a two-second habit that prevents a title reading as an unfinished sentence.

Translation changes length, sometimes by a lot

Text does not keep its length across languages. English to German or Russian typically expands by twenty to thirty-five per cent; English to Finnish or Hungarian can expand further. Translating into Chinese, Japanese or Korean usually contracts the character count substantially, since a word often becomes one or two characters.

That is a design constraint as much as a writing one. A button labelled Save in English becomes Speichern in German and Enregistrer in French, so a fixed-width button designed around four characters breaks in both. The standard advice is to design for roughly forty per cent expansion on short strings, and to test with the longest target language rather than the shortest.

For anyone writing source text destined for translation, shorter sentences, fewer idioms and no puns save real money, because translation is priced per word and ambiguity is priced in questions. It also makes machine translation of the same text substantially better.

Counting when the text is code, markup or mixed

A word count over markup counts the markup. Pasting HTML gives a figure inflated by tags and attributes; pasting Markdown counts link URLs alongside their text. If the number is going to a client or an editor, strip to the rendered text first - copying from the rendered page rather than the source is the quickest way.

Mixed-language text is handled here by applying both rules at once: runs of Latin script are counted as words, and Japanese, Chinese and Korean characters individually. That keeps a bilingual document from reporting a misleading total in either direction.

For CJK text specifically, the character count is usually the figure that matters commercially, since translation and editorial rates in those languages are quoted per character rather than per word. Both numbers are shown together for exactly that reason.