Four numbers called “characters”
A family emoji is one visible character, seven code points, eleven UTF-16 code units and 25 UTF-8 bytes. A counter that reports only one of those is right about one question and wrong about the others. The counter on this site shows graphemes (what a reader sees), code points (what Unicode calls characters) and UTF-8 bytes (what storage and transport count), so the disagreement is visible instead of mysterious.
| Measure | “👨👩👧👦” | Where it matters |
|---|---|---|
| Grapheme clusters | 1 | Word-processor counts, cursor movement |
| Code points | 7 | Unicode APIs, this counter's “characters” |
| UTF-16 code units | 11 | JavaScript .length |
| UTF-8 bytes | 25 | Column limits, headers, bandwidth |
What counts as a word
The counter's word rule is a run of letters or digits, optionally continuing with letters, digits, apostrophes or hyphens. That treats “don't” as one word and “well-known” as one, and it does not count a standalone punctuation mark. A different tool may split on whitespace only, count CJK characters individually, or exclude numbers — all defensible, all different. If a word limit is a hard requirement, use the counter that matches the checker, or leave a margin.
Word-frequency ranking adds a second definition: case and normalisation. The frequency tool can fold case, so “The” and “the” are one entry; leaving case sensitive treats them as two. Neither is wrong, but the CSV export should be read with the setting in mind. Very common words dominate any raw ranking, which is why the table shows each word's share rather than only its count.
Reading time and speech
Reading time is an estimate, usually words ÷ 200 per minute for prose. It is a planning figure, not a measurement: code, tables and dense technical text take longer, and skimming takes less. If a page quotes a reading time, it should say which rate it used; the counter uses 200 words per minute and reports the raw word count alongside, so the estimate can be recomputed with a different rate.
Counting for limits
A limit is only meaningful with its unit. “280 characters” on a social platform usually means graphemes or code points; a database column measured in bytes will reject a string that fits the visible count; a form that counts UTF-16 units disagrees with both for emoji. Before trimming text to a limit, find out which number the destination enforces and leave a margin — the counter shows all three so the difference is visible before the submission fails.
Repetition is a generator, not a measurement
The text repeater takes the whole input as one block and repeats it with a separator — a newline, a space, nothing or your own string. It is for producing repeated content (a divider, a test string, a pattern), not for counting. The count is capped so the page stays responsive; for very large outputs, generate a shorter block and combine it in code rather than in a textarea.
WPM and accuracy
A typing test's WPM uses the five-character word: WPM = characters ÷ 5 ÷ minutes. Net WPM
counts only characters that match the target, so errors reduce it twice — they consume time and lose
credit. Accuracy is correct ÷ typed. A 60-second attempt with 250 correct characters is 50 net WPM;
ten errors in the same 250 keystrokes make it 48 net WPM at 96% accuracy.
The clock convention matters as much as the formula. This test starts on the first keystroke and stops at the end of the passage; another site may start on page load or subtract an error penalty. Compare your own trend on one site, not a score across sites, unless both state the same rules.