A score is a question, not a grade
You paste a memo you are rather proud of and the panel answers with six numbers. One of them says 14.2. Fourteen is a school grade, you are not writing for fourteen-year-olds, and the document suddenly looks like a problem.
That reaction is the reason this page exists. The six numbers do not measure how good your writing is. They measure how long your words and sentences are, on scales invented for other purposes. Used carelessly, they will send you rewriting a clear sentence into a choppy one to move a decimal point. Used well, they tell you which sentences deserve a second look.
For the history and interpretation of the scores themselves, what readability scores actually mean covers the formulas in depth. This page is about the panel in front of you: what it computes, where the arithmetic wobbles, and how to get value out of it in five minutes.
What the six formulas weigh
All six reduce your text to one or two measurements and do arithmetic on them. None of them reads your argument. The table shows what each one actually counts.
| Formula | What it weighs | Which way is easier |
|---|---|---|
| Flesch reading ease | syllables per word and words per sentence | higher is easier, roughly 0 to 100 |
| Flesch–Kincaid grade | the same two inputs, reported as a school grade | lower is easier |
| Gunning fog grade | sentence length and the share of words with three or more syllables | lower is easier |
| Coleman–Liau grade | letters per 100 words and sentences per 100 words, with no syllable counting at all | lower is easier |
| Automated readability grade | letters per word and words per sentence, also without syllables | lower is easier |
| SMOG grade | the count of long words against the number of sentences | lower is easier |
Two of the six never touch syllables. That difference is the source of most of the disagreement between the six scores, and it is worth pausing on.
Why the six numbers disagree with each other
The panel estimates syllables rather than looking them up, because a browser-side tool has no pronunciation dictionary to consult. The estimate is a set of sensible shortcuts: a word of three letters or fewer counts as one syllable, a trailing "ed" is stripped, so are "es" and a silent "e" when they follow a consonant, and whatever remains is counted by vowel groups.
The shortcuts are right most of the time and spectacularly wrong on the words you would least expect.
| Word | Syllables counted | Syllables in a dictionary |
|---|---|---|
| created | 1 | 3 |
| rhythm | 1 | 2 |
| queue | 2 | 1 |
| beautiful | 4 | 3 |
| business | 3 | 2 |
Every one of those errors feeds into the three syllable-based formulas, which is why Gunning fog and Flesch reading ease can disagree with Coleman–Liau on the same paragraph. Since Gunning fog and SMOG both treat anything the estimator gives three syllables or more as a complex word, a single misjudged word moves two scores at once.
The numbers also refuse to stay inside their advertised ranges. Paste a very short, simple passage and Flesch reading ease reports something above 100. Paste dense management prose and it goes negative, with grade estimates climbing past 30. The formulas are linear, and linear formulas have no ceiling; the familiar 0-to-100 band is a convention, not a constraint.
The sentence list under the scores is the real output
Below the six scores sits a list labeled "Sentences worth a second look": every sentence in your text longer than 25 words, up to fifteen of them, each shown with its word count.
That list is the actionable part of the page, and most people scroll past it on the way to the grade. The cutoff is strict: a sentence of exactly 25 words does not appear, and one of 26 does. If your text is all short, punchy sentences the list will be empty, which means only that nothing crossed the threshold.
The list is capped at fifteen, so a long, tangled document shows you the worst offenders rather than all of them.
Using a score honestly
The useful comparison is always two versions of the same text: before and after a revision, or your draft against the published piece you are trying to match. Two different pieces of writing produce numbers that reflect their content more than their clarity.
Lower is not automatically better. A consent form and a technical specification both serve readers who need exact terms, and dragging either to a fifth-grade score would be a failure. The scores do not know who your reader is, and no formula can. That judgement is the subject of plain language editing, which is about clarity rather than arithmetic.
Where the scores are reliable is on a document you have already revised. Run the check on the old version, then on the rewrite. If the numbers moved down while the meaning held, you probably did something right.
Plain-language editors use that workflow on forms and letters, where the audience is fixed and the reading level is a policy target. Technical writers watch the sentence list for documentation that has accumulated qualifications. Teachers check whether a passage sits near the level of the class.
Limitations, stated plainly
The formulas see words and sentences, and they are blind to everything else: paragraph structure, headings, whether the examples land, whether the argument is broken, whether the reader already knows the jargon. A text can score as easy and still be impossible to follow.
The panel is English-only. It estimates syllables heuristically, as shown above, so the numbers wobble more on technical vocabulary and proper nouns than on ordinary prose. Short samples wobble most, and SMOG was designed around a sample of about 30 sentences, so its figure on a three-sentence paragraph is a rough guess.
And the tool cannot tell you what your reader will understand. It can hand you a list of long sentences and six numbers that moved in a direction. What you do with them is an editorial decision.
Frequently Asked Questions
Why did my Flesch reading ease score come out above 100?
Because the formula has no upper limit. The familiar 0-to-100 band is a convention, and on a short or unusually simple sample the raw arithmetic runs past it. A score of 118 is not better than a score of 95; it signals a very short or very plain sample.
How much text should I paste before the scores mean anything?
Longer is better, and 200 words is a reasonable floor. Short samples produce unstable numbers because one long word or one long sentence carries a large share of the total. For SMOG, anything shorter than about 30 sentences is an estimate with a wide margin.
Why do the six scores disagree so much on the same paragraph?
Because they weigh different things. Gunning fog, Flesch, and SMOG count syllables, Coleman–Liau and the automated readability grade count letters instead, and each applies its own arithmetic. When the syllable estimate misjudges a word like "created" or "beautiful", the syllable-based scores move and the letter-based scores do not.
Which of the six should I trust?
None of them alone, and the disagreement between them is a useful signal in itself. A wide spread usually means the text mixes long technical words with short sentences, which is exactly the situation where a single averaged score stops being informative. Read the sentence list instead.
Can these scores prove a document is accessible?
No. Accessibility depends on structure, headings, contrast, screen-reader behavior, and plain syntax, none of which these formulas measure. A high Flesch reading ease score is not evidence that a screen reader will make sense of your page.
Why did the score change when I only swapped two words?
Because syllable estimation is sensitive to individual words, and one extra syllable in a short sample moves the average noticeably. To see a meaningful movement, revise a whole paragraph rather than tweaking words and watching the decimal.
Split one sentence and look again
Here is the habit that turns this panel into something useful. Ignore the six numbers on the first pass and read the sentence list instead. Pick the longest sentence, split it where a new subject appears, and paste the revision back in. Then look at how far the scores moved. That loop teaches you what these formulas respond to faster than reading about them.