Counting words, and what the other numbers tell you
Why two tools disagree about the same text, what the readability score actually measures, and the counts that matter for a given medium.
Last reviewed · 1,269 words
In short
- Word counts differ between tools because hyphenated words, numbers, URLs and em dashes are handled differently. A 2% disagreement is normal.
- Character count with spaces and without are both used, and it matters which one a limit refers to.
- Flesch Reading Ease measures sentence length and syllables per word. It says nothing about whether the writing is any good.
- Reading time at 200 to 250 words a minute is a rough guide; technical material is read considerably more slowly.
- Shorter sentences do more for readability than shorter words, and they are easier to fix.
Counting words looks unambiguous and is not, which is why two tools looking at the same paragraph can disagree.
Why the counts differ
A word counter has to decide what a word is, and reasonable implementations decide differently.
Hyphenated compounds. Is "state-of-the-art" one word or four? Most tools say one; some say four.
Numbers. Is "1,500" a word? Is "3.14"? Usually yes, and not always.
URLs and email addresses. Frequently split at the dots and slashes into several "words".
Em dashes without spaces. "word—word" reads as one token to a splitter that breaks on whitespace, and as two to one that breaks on punctuation.
Contractions and possessives. "don't" and "John's" are one word almost everywhere, which is the one point of general agreement.
A 1% to 2% difference between tools on ordinary prose is normal. Where a limit is hard — a submission cap, a contract, a competition — the count that matters is the one produced by whatever the recipient uses, and it is worth checking against that specific tool rather than assuming.
Characters, with and without spaces
Both are used and the difference is substantial: spaces are roughly 15% to 18% of a typical text.
With spaces is the usual meaning of a character limit on social platforms, SMS and most forms.
Without spaces is the convention in translation and copywriting pricing, and in some academic abstract limits.
A 1,000-character limit therefore means about 850 characters of actual text if spaces count, and 1,000 if they do not. Where a limit matters, the specification usually says which — and where it does not, assume spaces count.
Reading time
The standard conversion is 200 to 250 words a minute for adult silent reading of ordinary prose.
| Content | Typical rate |
|---|---|
| Light fiction | 250–300 wpm |
| General non-fiction | 200–250 wpm |
| Technical or academic | 100–150 wpm |
| Reading aloud | 130–150 wpm |
| Podcast or audiobook | 150–160 wpm |
A "5 minute read" label on an article is roughly 1,000 to 1,250 words. It is a useful signal for a reader deciding whether to start, and it should not be treated as precise — the same 1,000 words of dense technical writing takes twice as long.
For a speech, the number to use is 130 words a minute. A ten-minute talk is about 1,300 words, and most people write far more than that and then rush.
The readability scores
Flesch Reading Ease produces a score from 0 to 100, higher being easier:
206.835 − 1.015 × (words per sentence) − 84.6 × (syllables per word)
| Score | Level |
|---|---|
| 90–100 | Very easy, age 11 |
| 70–89 | Easy, age 12–13 |
| 60–69 | Standard, age 14–15 |
| 50–59 | Fairly difficult |
| 30–49 | Difficult, university level |
| 0–29 | Very difficult, graduate level |
Flesch-Kincaid Grade Level rearranges the same two inputs to give a US school grade instead.
Both measure exactly two things: how long your sentences are and how many syllables your words have. That is the whole of it.
Which means a passage of complete nonsense in short words and short sentences scores as highly readable, and a beautifully clear explanation using necessary technical terms scores as difficult. The score cannot see logic, structure, coherence or whether the argument works.
Used properly, it is a flag rather than a verdict. A score of 25 on writing intended for a general audience suggests the sentences have got away from you, and shortening them is usually the right response. Chasing a target score by replacing accurate words with shorter inaccurate ones makes the writing worse while improving the number.
Sentence length is the lever worth pulling. It carries roughly as much weight in the formula as syllables and is far easier to fix — splitting one 40-word sentence into three does more than rewording a paragraph.
Length conventions
| Medium | Typical length |
|---|---|
| Tweet | 280 characters |
| SMS segment | 160 characters |
| Meta description | 150–160 characters |
| Page title tag | 50–60 characters |
| LinkedIn post before truncation | ~140 characters |
| Blog post | 800–2,000 words |
| Long-form article | 2,000–5,000 words |
| Academic abstract | 150–300 words |
| Novel | 70,000–120,000 words |
| Novella | 20,000–50,000 words |
SMS is worth a note: 160 characters applies to plain GSM encoding. A single emoji, or any character outside the GSM set, switches the whole message to UCS-2 and drops the limit to 70 characters — which is how a message that looked short arrives as three.
Counting for a purpose
Academic work usually excludes the abstract, references, tables and appendices from the count, and the specification says which. Getting this wrong in either direction is a common and avoidable problem.
Legal and contractual limits are counted by whatever tool the receiving party uses, which is worth establishing in advance.
Payment by word — translation, freelance writing — needs the basis agreed: source or target text, with or without spaces, and how numbers and repeated text are handled.
SEO has no word count that ranks. What matters is whether the page answers the question fully, and length follows from the subject rather than from a target. Padding to reach a number is precisely what the helpful-content systems demote.
The other counts, and what they are for
Paragraph count is a rough structural check. Long stretches without a break are harder to read on a screen than on paper, and a piece averaging more than about 120 words per paragraph usually benefits from splitting.
Sentence count matters mostly as an input to average sentence length, which is the readability lever worth pulling. Under 20 words per sentence reads easily; above 30 the reader is working.
Unique word count and the ratio of unique to total words describe lexical variety. A low ratio in a long piece can indicate repetition, and it can equally indicate a technical subject with a small necessary vocabulary — so it is a prompt to look rather than a finding.
Longest sentence is more useful than the average, because prose is spoiled by its outliers. One 60-word sentence in an otherwise clear page is what a reader stumbles on, and the average will not show it.
Keyword frequency is worth checking for accidental repetition — the word you used four times in one paragraph without noticing — rather than for keyword density, which has not been a meaningful ranking factor for many years.
Editing to a length
Cutting to a word count is a craft skill and it has a reliable order.
Cut whole sections first. The largest reductions come from removing an idea, not from tightening sentences. If the piece is 30% over, something belongs somewhere else.
Then cut sentences that restate. Most drafts say the same thing twice, once abstractly and once concretely, and the concrete one is usually the keeper.
Then compress phrases. "In order to" becomes "to". "Due to the fact that" becomes "because". "At this point in time" becomes "now". These are worth 5% to 10% and no more.
Adverbs and intensifiers last. "Very", "really", "quite", "actually" — removing them tightens prose and rarely gets you to a target on its own.
Going the other way, padding to reach a minimum is visible to readers and, on the web, to ranking systems built specifically to detect it. If the subject is covered in 600 words, publish 600.
What this tool assumes
- Words are sequences separated by whitespace, with hyphenated compounds counted as one word.
- Sentences end at a full stop, question mark or exclamation mark, which slightly overcounts where abbreviations contain full stops.
- Syllable counting for the readability scores uses a heuristic, which is accurate to about 95% on English prose and worse on names and technical vocabulary.
- Reading time uses 225 words a minute, midway through the ordinary range.
- Everything is computed in your browser. Nothing you paste is sent anywhere.