Characters and whitespace
The text is normalized to Unicode NFC and CRLF line endings become LF. In supported browsers, character totals use grapheme segmentation: a base letter and combining accent remain one visible unit. Explicit line breaks are excluded from character totals. Tabs count as one space. The result without spaces excludes whitespace. Bytes, code points and a document’s file size are different measurements.
Words, sentences and paragraphs
Word tokens contain letters, marks or numbers, with internal apostrophes, hyphens and numeric separators retained under the declared rule. Standalone punctuation is not a word. Unique words ignore case and normalize apostrophe styles. Sentences use the selected language’s browser segmenter with a short abbreviation list; unusual abbreviations and lists can still need inspection. Paragraphs are non-empty blocks separated by a blank line; lines count explicit returns, including a final empty line.
Frequency and exact phrases
One-, two- and three-word sequences are counted inside sentences. Each percentage uses all eligible sequences of that length as its denominator. The common-word option only hides all-common-word expressions from the displayed list; it does not remove tokens from that denominator. An exact phrase search compares whole normalized tokens, not an arbitrary substring inside another word.
Conversions and duplicate lines
Case transformations use the selected language locale. They can expand a Unicode mapping, so count the final result again. Duplicate removal uses a separate comparison key, optionally ignoring case or edge whitespace. The first original line is kept in its original order. NFC-equivalent keys compare alike, while the retained spelling is preserved. This does not identify duplicate people or parse CSV records.
Imports and planning estimates
TXT is read as UTF-8. DOCX extraction takes body paragraphs and tables and excludes headers, footers, notes, text boxes, comments and deleted revisions. PDF text extraction must be checked for order and separators; OCR is not performed. Files are limited to 10 MB, PDF imports to 200 pages, and extracted text to 500,000 UTF-16 units. Reading and speech estimates divide words by an editable speed. The initial 200 and 130 words/min are assumptions, not measurements of every reader.
Compare a small sample with a different counter
Start with the exact same string, not two separately retyped versions. Café! gives five graphemes, four letters and six UTF-8 bytes. A clear message. gives 16 characters with spaces, 14 without spaces and three words under this method. If two tools disagree, check their treatment of line breaks, punctuation, hyphens, notes and selected text. Do not add a guessed adjustment to a whole manuscript: identify the rule on a controlled sample, then apply the destination’s required scope.
Read a percentage together with its base
In the sentence red blue red, there are three one-word positions and two two-word positions. The word red occurs twice, so its one-word share is 2/3. The phrase red blue occupies one of the two eligible pairs, so its two-word share is 1/2. A sentence boundary stops a sequence; it is not bridged merely because the displayed text is adjacent. Hiding common words changes the visible list, not the underlying number of eligible positions.
Check extraction before trusting a total
After import, inspect the beginning, ending and a passage containing a table or note. A PDF can store text in an order unlike its visual layout, and a scanned page needs OCR elsewhere before this tool can count its words. DOCX scope is deliberately limited to the supported body content; a thesis requirement may demand additional parts. Keep any required omissions in a separate inventory. After corrections, count the final editor text and record the unit and settings with the report.