Word Frequency Counter

Any run of Unicode letters, marks, or digits, optionally joined by apostrophes or hyphens. So don’t is one word (not don + t), state-of-the-art is one hyphenated compound (not four words), and Hindi or other non-Latin scripts tokenize correctly. Pure numbers like 2026 are counted by default; untick “Count numbers as words” to exclude them. Counting is case-insensitive unless you enable the case-sensitive toggle — useful when Apple the company and apple the fruit must stay separate.

TTR = unique words ÷ total words, a standard measure of lexical variety. “The cat sat on the mat” has 5 unique words out of 6 total: TTR 0.833. “The cat sat, the cat slept, the cat ate” has 5 unique out of 9: TTR 0.556 — more repetitive. There is no universal “good” value because TTR falls as texts get longer (you inevitably reuse words), so only compare texts of similar length. For 1,000-word articles, roughly 0.35–0.45 reads varied; under 0.30 usually feels repetitive.

Single-word counts blur distinct topics together: “customer”, “service”, and “support” might each rank high separately. The bigram tab shows whether your text actually says “customer service” 14 times versus “customer support” twice — the phrasing your readers (and search engines) actually see. Phrases that start or end with a stopword are filtered out when the stopword filter is on, so you see “customer service” rather than “of the”.

Not as a dial you can turn. Google has said for years that there is no ideal keyword-density percentage, and stuffing a term harms rather than helps. Frequency analysis is still genuinely useful as a diagnostic: if your target phrase never appears in 1,500 words, the page probably is not about that topic in any recognisable way; if one phrase is 4–5% of all bigrams, it will read as spam to humans too. Use the table to check coverage and naturalness, not to chase a magic number.

Paste a full draft, keep the stopword filter on, and read the top 25 single words. Most writers have tics — “really”, “actually”, “just”, “very”, “quite” — that survive editing because each instance looks innocent. Seeing “just” with count 31 (0.8% of all words) makes the pattern undeniable. The 3-word tab catches phrase-level tics like “it is important” or “at the end”. Repeat after editing and watch the counts drop.

Live analysis runs up to about a million characters (roughly 200,000 words — a full-length novel) and recalculates as you type. Beyond that the tool switches to an explicit Run button so a single keystroke never re-tokenizes a giant text mid-edit. The counting itself is a single pass building three hash maps, so even novel-length inputs analyse in a couple of seconds.