What each statistic actually counts
Words are counted by splitting on whitespace, which is what most online counters do and what makes “state-of-the-art” one word and “don’t” one word. When at least a fifth of the text is Chinese, Japanese or Korean there is nothing to split on, so the counter switches to Intl.Segmenter word segmentation and shows a CJK badge.
Characters are counted as graphemes — what a reader would call one character. An emoji built from several code points counts as 1, “é” typed as e plus a combining accent counts as 1, and a CRLF line break counts as 1. The “no spaces” figure is the same count minus every whitespace character.
Sentences use Intl.Segmenter with an abbreviation guard, so “Dr. Smith paid $3.14” stays one sentence instead of three; about thirty abbreviations (Dr, Mr, Prof, vs, etc, e.g, i.e, a.m, p.m and others) are merged back. Paragraphs are blocks separated by at least one blank line, and lines are newline characters plus one — with empty text counted as 0 lines, not 1.
Reading time and speaking time
Reading time defaults to 238 words per minute. That figure is the median adult silent reading rate for English non-fiction from Marc Brysbaert’s 2019 meta-analysis of 190 studies, published in the Journal of Memory and Language — not a round number someone picked. Speaking time defaults to 150 wpm, a common unhurried presentation pace.
A 1,200-word article is 1,200 ÷ 238 = 5 minutes 2 seconds of reading. A ten-minute talk at 150 wpm needs about 1,500 words, and a 45-second video script about 110. Both rates are editable — 50 to 1,000 wpm for reading, 50 to 500 for speaking — because a dense technical paper and a light blog post are nowhere near the same speed.
Treat the result as a planning figure, not a measurement. Reading speed varies with difficulty, familiarity and language, and comprehension drops sharply above roughly 400 wpm. For a script, rehearsing once and timing yourself beats any formula — then set the wpm field to what you actually did.
Lengths worth knowing
- 5-minute talk at 150 wpm: ~750 words · 10 minutes: ~1,500 · 20 minutes: ~3,000
- Journal abstract: 150–250 words · Conference abstract: often 300
- IELTS Writing: Task 1 at least 150 words, Task 2 at least 250
- Common Application personal essay: 650 words maximum
- Typical long-form blog post: 1,000–2,000 words · Newsletter: 500–800
- Meta description: 155 characters · Title tag: 60 characters (character limits, not words)
Why your word count differs from Word or Google Docs
Every counter draws its own boundaries. Hyphenated compounds, em-dash-joined phrases, numbers with thousands separators, URLs and inline code all get split differently. A whitespace split — what this tool uses for Latin text — counts “state-of-the-art” as one word, while a tokeniser that treats hyphens as separators counts four.
The gap widens in CJK text. Microsoft Word reports Korean, Chinese and Japanese as characters, so a 400-character Korean paragraph shows as 400 “words” there and as roughly 130 segmented words here. Neither is wrong; they answer different questions, which is why this page shows characters and words side by side.
If a submission has a hard limit, count the way the recipient counts. University and journal guidelines usually say whether footnotes, references, captions and headings are included — that decision moves the total far more than any tokenising difference. Hover any statistic on this page to see the rule it applied.
Reading the keyword density table
Density is a term’s count divided by the total number of tokens, as a percentage. Phrase length switches between single words, two-word and three-word phrases, and “Exclude common words” filters using a list of about 180 English function words plus around 60 Korean particles and dependent nouns. An n-gram is dropped only when every word in it is a stopword, so “end of the road” survives.
Percentages are calculated against all tokens, before stopword filtering, so the numbers stay comparable to the text as a whole rather than to a filtered subset. Ties keep first-appearance order, and punctuation and symbols are stripped before tokenising, so “SEO,” and “SEO” are the same term.
There is no target percentage to hit. Google has never published a density threshold, and naturally written prose usually lands somewhere between 0.5% and 2.5% for its main term. Use the table to spot accidental repetition and to check that the words you care about actually appear — not to tune a number upward, which reads as stuffing to readers long before it does to a search engine.
Limits, privacy and troubleshooting
Everything runs in this tab. The text is kept in this browser’s localStorage so a reload does not lose a draft, and “Clear” removes it. There is no account, no upload and no server copy — which also means the draft is gone if you clear site data or switch browsers.
Above 100 KB the grapheme walk is replaced by code-point counting and an “approximate” badge appears; the difference only shows up with emoji and combining marks. Above 250 KB the keyword density panel starts collapsed, with a button to compute it anyway.
Pasted text often carries invisible passengers: non-breaking spaces from a web page, soft hyphens from a PDF, or double spaces after full stops. They change character counts and can merge or split words, so if a total looks wrong, paste into a plain-text editor first and copy back.