Word Counter & Text Analyzer

A word counter is almost entirely DEFINITIONS, and the ones most tools pick are wrong in ways nobody notices until the number matters.

Characters, for instance. This tool reports three numbers and names all three, because they are genuinely different: on the text "naïve café" followed by a thumbs up and a flag of Japan, a reader counts 14 characters, there are 15 Unicode code points, and String.length gives 18. The flag alone is one character made of two code points, each of which is a surrogate pair. Most counters report the last number, and a form with a character limit is usually counting one of the other two, so the tool shows you which is which rather than picking one silently.

Words are a choice too, offered as three rules with the difference shown. The default is a run of letters, digits and combining marks, with an apostrophe or a hyphen counted as part of the word when it sits between two of them, so O'Brien's and co-op are one word each and a stray dash is none. The second is your browser's own Unicode word segmenter (UAX #29), which splits on the hyphen and gives five where the first gives four. The third is anything between spaces, which is what most counters do. Whichever you pick, the other counts are shown beside it.

Sentences are where the arithmetic goes wrong quietly, because every readability formula divides by the sentence count. Splitting on every full stop reads "Dr. Smith paid $3.14." as three sentences; your browser's Unicode segmenter reads it as two, because UAX #29 has no notion of an abbreviation; a reader says one. This tool says one: a sentence ends at a full stop, exclamation mark, question mark or ellipsis followed by whitespace or the end of the text, unless the word before it is one of 64 listed abbreviations, or a single initial, or the stop sits between two digits. A line break inside a paragraph does not end a sentence; a blank line does. The abbreviation list is on the page, because a list is not a standard and you should be able to see what is in it.

The six readability scores are each computed from their own publication and shown with it: Flesch (1948), Flesch-Kincaid (1975), Gunning Fog (1952), SMOG (1969), Coleman-Liau (1975) and the Automated Readability Index (1967). Nothing is clamped. Flesch Reading Ease is not a 0 to 100 scale: a deliberately dense academic sentence measures around -176 here, and a tool that clips it to 0 shows the same number for that as for merely difficult prose. Nothing is suppressed either: a nine word text gets real scores rather than a row of zeros. SMOG is shown with a note when the text has fewer than the thirty sentences it is defined on, because the number means nothing below that. Syllables, which four of the six depend on, are an estimate with a measured error rate: the rule is stated in full and is right on 108 of 128 words counted by hand, wrong on 12 (business, Wednesday and colonel among them) and defensibly different on 8 more. Coleman-Liau and the ARI deliberately avoid syllables, which is why they exist and why they are worth reading when the others disagree.

The SEO and keyword density tabs work off the same counts, so an accented or non-Latin word appears in the frequency table as itself. The Text Statistics Analyzer reads the same module, so the two tools cannot give different answers to the same text.