What these formulas actually measure
Every score on this page is a straight line drawn through two inputs: how long your sentences are, and how long your words are. Flesch and Flesch-Kincaid measure word length in syllables. Gunning fog counts the share of words with three or more syllables. SMOG counts the same polysyllables but takes their square root. Coleman-Liau and the Automated Readability Index skip syllables and use letters per word instead. That is the whole mechanism. There is no step in which the formula considers whether a sentence is ambiguous, whether the argument follows, whether the terminology is defined, or whether the reader has any reason to care.
Understanding that ceiling is what makes the scores usable. A passage of confident nonsense, cut into eight-word sentences with mostly one-syllable words, scores as easy reading. A precise, well-organised paragraph about pension accrual scores as hard, because "accrual" has three syllables no matter how clearly it is explained. The formulas are useful as a rough alarm on sentence length and vocabulary weight. They are not useful as a verdict.
Where each formula came from, and what it was for
| Formula | Origin | Designed for |
|---|---|---|
| Flesch Reading Ease | Rudolf Flesch, 1948 | General adult prose, magazines and newspapers |
| Flesch-Kincaid grade | US Navy, 1975 | Technical manuals for enlisted personnel |
| Gunning fog | Robert Gunning, 1952 | Business writing and newspaper copy |
| SMOG | G. Harry McLaughlin, 1969 | Health and safety material where comprehension is near-total |
| Coleman-Liau | Coleman and Liau, 1975 | Machine scoring without syllable data |
| ARI | US Air Force, 1967 | Teletype-era automated scoring |
Those dates matter. These were calibrated decades ago, on printed material, against reading tests of the era, for populations that are not your audience. The Navy formula was tuned so that a manual scoring at grade 9 could be read by a sailor who had finished ninth grade in 1970s America. Applying that same constant to a 2020s product page for a specialist audience is an act of extrapolation nobody has validated. Treat a reported grade level as a comparative number — this draft versus the last one — rather than an absolute claim about who can read it.
Syllable counting is the weak point, and here is exactly how it is done
English spelling does not encode syllable boundaries reliably, so any counter without a pronunciation dictionary is guessing. This one guesses in six documented steps: words of three letters or fewer count as one; a final -ed is treated as silent unless it follows t or d, so walked is one syllable but wanted is two; a final -es is treated as silent unless it follows s, x, z, ch or sh, and never when it follows a consonant plus l; a final silent e is dropped unless a consonant-plus-l precedes it, which is what keeps table at two while reducing whale to one; a leading y is not treated as a vowel; and what is left has its runs of a, e, i, o, u and y counted, with a floor of one.
It gets beautiful right at three and readability right at five. It gets science wrong, calling it one syllable instead of two, because the ie reads as a single vowel run. It gets business wrong in the other direction, calling it three where a speaker says two. Created comes out at two rather than three. Names, loanwords and anything with a diaeresis are unreliable.
Here is a check you can repeat. Flesch used the sentence The Australian platypus is seemingly a hybrid of a mammal and reptilian creature as a worked example, counting 13 words and 26 syllables, which gives a Reading Ease of 24.4. This page reports 13 words and 24 syllables for the same sentence, and therefore 37.5. The formula is identical to the decimal; the entire 13-point gap is two syllables, both of them in Australian and reptilian, where the heuristic reads a vowel run as one beat where a speaker gives it two. That is the size of error a syllable heuristic can produce on a short sample, and it is why the syllable count sits on the page next to the score. This is why the syllable total is printed on the page rather than hidden inside the score: if your text is full of a word the heuristic mishandles, you can see the total drift and discount the result accordingly. A tool that reports a Flesch score to two decimal places while concealing its syllable count is claiming a precision it does not have.
Sentence splitting, and the abbreviations it knows about
Sentence count is the other half of every formula, and a splitter that breaks on every period inflates the count and flatters the score. A period here ends a sentence only when it is followed by whitespace or the end of the text, and only when the next word does not begin with a lowercase letter. On top of that, a lone period is ignored when the letters before it are a single character — which covers initials, e.g., i.e., a.m., p.m. and U.S. — or when those letters match a known abbreviation.
The abbreviation list covers titles (Mr, Mrs, Ms, Dr, Prof, Sr, Jr, Rev, Gen, Col, Capt, Lt, Sgt, Gov, Sen, Rep, Pres), organisational forms (Inc, Ltd, Co, Corp, Dept, Univ), reference terms (Fig, Vol, vs, etc, al, cf, ca, eq, pp, ed, eds, trans, approx), street types (St, Mt, Apt, Ave, Blvd, Rd) and the abbreviated months and weekdays. Decimals are handled structurally rather than by a list: in 3.5 no whitespace follows the point, so it never qualifies as a terminator. Ellipses break only when the following word is capitalised, which keeps a trailing-off mid-sentence pause intact. Blank lines and line breaks always end a sentence, so bullet lists and headings are counted as separate units.
Two known failures: a sentence ending in a single-letter word (He was awarded an A. Then he left.) does not split, and an abbreviation genuinely ending a sentence (Bring rope, tape, etc.) merges into the next one. Both are rare and both are visible in the sentence count, which is printed at the top of the counts panel for exactly this reason.
Reading the six numbers together
Notice that the page reports a spread rather than one grade. The five grade-scale scores usually land within two or three years of each other, and when they do not, the disagreement is the information. If Coleman-Liau and ARI — the two that never look at syllables — come out much lower than fog and SMOG, your words are short but a handful of long ones are concentrated somewhere. If SMOG runs far above Flesch-Kincaid, polysyllables are clustered rather than spread. And if the whole spread is wide on a short passage, the sample is simply too small: none of these formulas was designed for a paragraph, and a single 40-word sentence can move a grade level by two years on a 200-word sample.
For the counts on their own without the formulas, the text statistics tool covers words, characters and bytes. For the distribution behind that average sentence length, the sentence length analyzer shows the histogram this page reduces to a single mean.
Questions people ask
Which score should I actually pay attention to?
Watch the spread rather than picking a favourite. If all five grade scores agree within a couple of years, the text is homogeneous and any one of them summarises it fine. When they disagree, that is telling you something the average would hide: fog and SMOG rising above the syllable-free scores means long words are concentrated in a few places, and a wide spread on a short passage usually means the sample is too small for any of them. If a specific formula is required by a house style or a regulator, use that one and ignore the rest.
My draft scores at grade 12. Is that bad?
It is a fact about your sentence and word lengths, not a judgement. Grade 12 is normal for a technical explanation, a legal summary or an argument that has to hold several conditions in view at once, and rewriting it to grade 6 by chopping every sentence in half can easily make it harder to follow, because the connective tissue between clauses is where the reasoning lives. A low score is worth chasing when the audience genuinely includes people who will struggle, or when the material is safety-critical. Otherwise treat a rising score across drafts as a prompt to check the longest sentences, which the page lists for you.
Why does another tool give me a different Flesch score for the same text?
Almost always the syllable counter or the sentence splitter, not the formula. The Flesch constants are fixed and public, so any two tools that agree on words, sentences and syllables will agree on the score to the decimal. They rarely agree on syllables, because every counter uses a different heuristic or dictionary, and they rarely agree on sentences, because splitters differ on abbreviations, ellipses, headings and bullet points. This page prints all three counts so you can see which input is responsible for the gap.
Does it work on text that is not English?
No, and it will still print numbers, which is the trap. Every constant in these formulas was fitted to English, and the syllable heuristic is built entirely from English spelling rules. Run German or Spanish through it and the syllable counter will systematically misfire; run a language without alphabetic syllable structure through it and the result is meaningless. There are language-specific adaptations of Flesch with different constants, and none of them are implemented here.
Is my text uploaded anywhere?
No. The whole calculation runs in the page on your own machine, in JavaScript that ships with the page, and there is no network request at any point. That matters for the obvious case of pasting something confidential, and also for the boring one: it keeps working with the connection off.