Counting and measuring a piece of writing

“How many characters is this” has three answers

They are all correct and they disagree, which is why the number in one tool does not match the number in the form you are pasting into.

What a reader sees. The unit is what Unicode calls a grapheme cluster: the thing that looks like one character. A flag emoji is one. An é is one whether it was stored as one code point or as an e with a combining accent.

Code points. The unit Unicode assigns numbers to. That é is one or two depending on how it was written. A flag emoji is two. A family emoji can be seven.

Bytes. What actually travels. In UTF-8, a Latin letter is one byte, an accented letter is two, most CJK characters are three, and an emoji is four or more.

For plain English these three agree and nobody notices. They diverge exactly where a limit matters most: a bio with an emoji in it, a name with an accent, anything not written in the Latin alphabet. A field that rejects your text as too long while your editor says it fits is usually counting bytes while you count what you can see.

The counter shows the ones that matter together, because the useful answer to “is this short enough” depends on which one the destination is enforcing.

Word counts are simpler and still not obvious

A word is a run of characters between separators, which works for English and stops working immediately after. Text in Chinese, Japanese or Thai has no spaces between words, so a space-based count returns something close to one, and a character count is the number that means anything there.

Hyphenation is the other place two tools disagree. Whether state-of-the-art is one word or four is a choice, not a fact, and reasonable tools make it differently.

Readability scores measure length, not clarity

The well-known formulas, Flesch and Flesch-Kincaid and the family that followed, combine two things: average words per sentence, and average syllables per word. That is all they can see.

This makes them genuinely useful in one direction. A document scoring badly almost certainly has long sentences and long words, and shortening both almost always helps. Government plain-language guidance in both the US and UK says roughly this, at greater length.

It makes them useless in the other. A formula cannot tell whether a short word is jargon to your reader: “the claim was denied per policy” scores beautifully and helps nobody. It cannot tell whether your ideas are in a sensible order. It cannot tell whether the sentence is true.

So the readability check is worth running as a warning and not as an approval. If it complains, it is right. If it does not, you have learned that your sentences are short.

What the word frequency actually tells you

Counting how often each word appears finds two things worth finding.

The first is repetition you cannot hear. Every writer has a word they lean on in a given week, and it is invisible while writing and obvious in a list sorted by count.

The second is subject drift. If a page about refunds mentions “refund” four times and “policy” nineteen, the page is about policy. That is a useful thing to know before publishing it, and it is a better use of word frequency than keyword counting, which stopped being how search works a long time ago.

Diffing is how you check a tool did what it said

Comparing two versions line by line is the practical way to verify a change rather than trust it. Clean a document and diff it against the original: every difference should be one you meant, and the ones you did not mean are the interesting output.

This is worth doing after any bulk operation. A find-and-replace that matched more than you expected shows up here as thirty changed lines where you were expecting three.

And placeholder text, which is not a joke

Lorem ipsum exists because real copy is distracting when what you are judging is a layout. You cannot see line length, spacing or hierarchy while your eye is busy reading the words, and nonsense in the right language and at the right length shows the shape without the reading.

The failure mode is famous and worth naming: it ships. Placeholder text has reached printed brochures, live sites and at least one national newspaper. If you use it, put something in your final checks that greps for it.

None of this needs an upload

Every tool here runs in your browser. Text you are measuring is often text you are still deciding whether to publish - a draft, a résumé, an internal note - and none of it needs to reach a server to be counted.