Word & Character Counter

about this tool

What is counted

Paste or type and you get characters with and without spaces, words, sentences, paragraphs, reading time and speaking time, updating as you type. Nothing is uploaded, which matters more than it sounds: the drafts people count are frequently unpublished, confidential, or somebody's coursework.

Characters, and why the number may differ from elsewhere

Characters are counted as graphemes — what a reader would call a character — not as the code units a programming language reports. This matters for anything outside plain English. A family emoji is one grapheme but several code points and more code units; é may be one code point or two depending on how the text was typed. So the count here matches what a human sees.

Be aware that a platform enforcing a limit may count differently. Twitter counts most characters as one and some scripts as two; a database VARCHAR counts bytes, and one emoji can cost four of them. If you are up against a hard limit, check which of the three numbers it applies to.

Words and sentences

A word is a run of characters between whitespace, so hyphenated compounds count as one and an em dash between words does not create one.

Sentence detection splits on full stops, question marks and exclamation marks, and knows about common abbreviations, so "Dr. Smith arrived." counts as one sentence rather than two. It is a heuristic, not a parser — decimal numbers and unusual abbreviations can still fool it, which is true of every sentence counter.

Reading and speaking time

Reading time assumes 220 words a minute, an average for silent reading of ordinary prose, and speaking time assumes 130, roughly the pace of a rehearsed presentation. Both are adjustable. Technical material with code in it reads considerably slower than the average, so treat the figure as a floor when sizing a talk.

Keyword density

The frequency table shows which words you have leaned on, with common words such as "the" and "and" filtered out by default. It is useful for spotting an unconscious tic in a draft. It is not an SEO target: writing to hit a keyword percentage has not worked for many years, and reads badly to the person you were trying to reach.

To cut an overused word everywhere at once, find & replace will do it with a count of what changed. For how hard the writing is to read rather than how much of it there is, readability scores gives Flesch, Fog and SMOG, and lists the sentences pulling them down.

A model counts text differently again: for a prompt, the LLM token counter gives the exact number of tokens GPT-5, GPT-4o and GPT-4 see, and shows where each one breaks.

questions

Why does the character count differ from the limit on another site?

Because platforms count differently. This tool counts graphemes — what a reader would call a character — while some platforms count code units and a database column counts bytes, where a single emoji can cost four. Check which number your limit applies to.

Is my text sent anywhere to be counted?

No. Counting happens in your browser as you type and nothing is uploaded, which matters given that most drafts being counted are unpublished, confidential or somebody's coursework.

How accurate is the sentence count?

It splits on full stops, question marks and exclamation marks and knows common abbreviations, so 'Dr. Smith arrived.' counts as one sentence. It is a heuristic, so decimal numbers and unusual abbreviations can still fool it.

What reading speed is used?

220 words a minute for reading and 130 for speaking, both adjustable. Technical writing with code in it reads much slower than average, so treat the estimate as a floor when sizing a talk.

Should I write to a target keyword density?

No. Keyword density has not been a useful ranking target for many years, and text written to hit a percentage reads badly to the person you were trying to reach. The frequency table is better used to spot words you have overused.