Text tools

Text tools look like the most trivial category on any converter site, and they are the one where the numbers most often disagree. This page explains why, so that when two tools give you different answers you know which one to trust for what you are doing.

Why every tool gives a different word count

There is no single definition of a word. Most software splits on whitespace, which means "state-of-the-art" is one word to some tools and three to others, "don't" may be one or two, and a numeral like "1,500" is sometimes a word and sometimes not.

The disagreements get much worse with Chinese, Japanese and Korean, because those scripts do not put spaces between words at all. A tool that splits on whitespace will report a 400-character Chinese paragraph as one single word. That is not a rounding error — it is a completely useless answer, and it is what a large share of English-first counters actually return.

The honest approach is to report more than one number: characters with spaces, characters without spaces, English words, and CJK characters separately. Then you can use whichever matches the rule you are actually being held to.

Which tool in here you actually want

Use the word counter when you need the headline "how long is this" figure: words, CJK characters, sentences, paragraphs and reading time in one view. Use the character counter when a limit is being enforced by software rather than by a person — it separates code points, UTF-16 units, visible characters and UTF-8 bytes, which are four different numbers as soon as your text contains an emoji, an accented letter or a single Chinese character.

Use the word frequency counter when the question is what you repeat rather than how much you wrote. It is the one where the difference between our approach and the usual implementation is most visible: splitting on whitespace turns an entire Chinese paragraph into one token, so we count characters or two- and three-character phrases instead and say plainly that phrase extraction is an estimate rather than dictionary segmentation.

Two neighbouring categories are worth knowing about. The keyword density checker and readability score live under SEO tools because that is who searches for them, and the Unicode inspector lives under developer tools — it is the one to reach for when a specific character is breaking a comparison, a database insert or a filename.

Which number the person judging you is using

Academic and publishing limits are usually word counts, and usually the count that Microsoft Word produces, because that is what the person checking will open the file in. Word counts CJK characters as words, which is why a Chinese document can look startlingly long in Word compared to a whitespace-splitting web tool.

Social platforms and SMS count characters, not words — and they count them in code units, which is why an emoji or an unusual symbol can consume more of your limit than it looks like it should. Form fields and databases also count characters, and often bytes rather than characters, which matters because one Chinese character is three bytes in UTF-8.

Search snippets and page titles are limited by rendered pixel width rather than character count, which is why two titles of the same length can truncate differently. Character count is only a rough proxy there.

Reading time, and why the estimates vary so much

Reading speed estimates for adults reading their own language on screen usually land somewhere between 200 and 250 words per minute for prose. Technical material is slower, and any figure quoted to the nearest minute is a rough guide rather than a measurement.

For Chinese the convention is characters rather than words, and the commonly used figures are considerably higher in raw character terms. A "five minute read" is therefore an editorial signal to the reader, not a fact about the reader — treat it as a way of setting expectations about length.

Why local matters for text specifically

People paste things into word counters that they would never upload as a file: draft contracts, cover letters, private messages, unpublished manuscripts, internal announcements, medical descriptions. Because it is "just text", the reflex to be careful does not fire.

A server-side counter receives all of it. This one cannot, because the counting happens in the same browser tab you typed into and there is no request that carries it. That is also why it keeps working with your network disconnected — which is a decent way to prove the point to yourself.