How hard is this to read?
Six formulas answer.

The formulas, as published

Flesch Reading Ease (Rudolf Flesch, “A new readability yardstick”, Journal of Applied Psychology 32, 1948) 206.835 − 1.015 × (words ÷ sentences) − 84.6 × (syllables ÷ words)

Flesch-Kincaid Grade Level (Kincaid, Fishburne, Rogers and Chissom, US Naval Technical Training Command, Research Branch Report 8-75, 1975) 0.39 × (words ÷ sentences) + 11.8 × (syllables ÷ words) − 15.59

Gunning Fog (Robert Gunning, The Technique of Clear Writing, 1952) 0.4 × [ (words ÷ sentences) + 100 × (complex words ÷ words) ] A complex word has three or more syllables, and Gunning left out proper nouns, compounds of short words, and words that reach three syllables only by an ending. Here that is: a capitalised word that does not open its sentence; a hyphenated word none of whose parts has three syllables; and a three-syllable word ending in -ed, -es or -ing whose stem has two. “Familiar jargon”, which Gunning also left out, is a judgement no rule can make, and is not attempted.

SMOG (G. Harry McLaughlin, “SMOG grading — a new readability formula”, Journal of Reading 12, 1969) 1.0430 × √( words of 3+ syllables × 30 ÷ sentences ) + 3.1291 McLaughlin counted thirty sentences: ten from the start, ten from the middle, ten from the end. This page counts every sentence and scales to thirty, which is the same thing on a thirty-sentence text. With fewer than thirty the card says so, and the figure is left out of the range in the headline.

Coleman-Liau (Meri Coleman and T. L. Liau, Journal of Applied Psychology 60, 1975) 0.0588 × L − 0.296 × S − 15.8 L is letters per 100 words, S is sentences per 100 words. Letters only: digits and punctuation are not counted.

Automated Readability Index (E. A. Smith and R. J. Senter, Aerospace Medical Research Laboratories, AMRL-TR-66-220, 1967) 4.71 × (characters ÷ words) + 0.5 × (words ÷ sentences) − 21.43 Characters are letters and digits.

The constants were checked on 7 October 2026 against the formulas as Wikipedia’s articles on each index give them, and the arithmetic against six answers worked by hand. Formulas are facts and carry no licence; the papers are named so they can be found.

What counts as a word and a sentence

A word is a run of letters, with any apostrophes and hyphens inside it (so “don’t” and “well-known” are one word each); a numeral (“1,837”, “3.14”); or a dotted abbreviation (“U.S.”). A sentence ends at a full stop, question mark, exclamation mark or ellipsis that is followed by a space and a capital letter, a figure or an opening quotation mark — or at a blank line. Text typed without capitals is read by a second rule: when the first letter of the sentence being read is a small one, a full stop, question mark or exclamation mark followed by a space and a small letter ends it too (an ellipsis does not). That rule is the weaker of the two: it cannot tell an abbreviation it does not know from the end of a sentence. A full stop does not end a sentence after a title (Mr., Dr., Prof.), inside a number or an address (3.14, labs.llc), after an initial (J. R. R. Tolkien), or after “etc.” and “U.S.” unless a plain sentence-opening word follows. A closing quotation mark stays with its sentence, and “Stop!” he said is one sentence.

Measured on this page’s own set of 68 cases written for it (abbreviations, initials, decimals, quotations, ellipses, numbered lists, text without capitals), the splitter gets 62 right. The six it gets wrong are kept in the set on purpose: a unit with a full stop (“5 ft. 10 in.”, and “5 ft. tall” in a text without capitals, where nothing marks the abbreviation), a sentence that ends on a single letter or on “U.S.” before a name, a quoted question followed by a name and “said”, and an ellipsis that ends a sentence before “I” (“It was over... I left.”). No rule can settle those without understanding the sentence. A stop with a letter straight after it, no space between, is never read as an end.

The page’s reviewer then wrote a separate set of 53 cases: 50 came out right. One of the three misses was text typed without capitals, which was read as a single sentence until the second rule above was added on 7 October 2026; the cases for it in this page’s set were written with that rule, so they flatter it.

Syllables, and how good the count is

Syllables are counted by a counter written for this page: it counts the groups of vowels in a word, takes off the silent endings (-e, -es, -ed), adds the pairs of vowels that are said apart (cre-ate, ra-di-o, qui-et), and then defers to a hand-made list of 192 words the rules get wrong (people, business, recipe). No dictionary is downloaded and no library is used. Numerals are counted the way they are read aloud, as Flesch instructed: “1918” is nineteen-eighteen, four syllables.

It was measured against three lists of words written for it, each with the count an American dictionary gives, and then against a fourth written by its reviewer, on 7 October 2026:

• The tuning list, 630 words. 622 right (98.7%) before any correction; all 630 after the rules were mended on it. This is the list the counter was fitted to, so it flatters.
• A held-out list, 324 common words, written before the counter was run on it: 323 right (99.7%) on the first run. The one miss, “continuing”, showed a rule counting twice and was then fixed.
• A list of 75 loanwords and odd spellings (yacht, tsunami, entrepreneur), run once, blind: 70 right (93.3%). “Genre”, “guacamole” and “ukulele” are still wrong today.
• The reviewer’s list, 581 words, 317 of them in none of the three lists above: 571 right (98.3%), and 307 of the 317 new words (96.8%). Eight of the ten misses were one rule, a doubled r before -ed (“occurred”, “stirred”), which the reviewer went looking for; the other two were “gooseberry” and “reenter”. All ten were then mended and are tested with the page; the list itself was not kept, so this figure is the reviewer’s report and is not re-run.

Read that as: about right on everyday prose, worse on names, borrowed words and technical terms. The lists were compiled by hand for this page and are not a random sample of English, so they are not a general accuracy figure. Words that are honestly said two ways (fire, every, family, different) were left out of the lists; the counter gives them one of their two counts. A wrong syllable moves Reading Ease, Flesch-Kincaid, Fog and SMOG a little; Coleman-Liau and the Automated Readability Index count no syllables and are untouched.

Time to read

Words divided by a rate. 238 words a minute is the mean for silent reading of non-fiction, and 183 for reading aloud, in Marc Brysbaert’s review and meta-analysis (“How many words do we read per minute?”, Journal of Memory and Language 109, 2019). 150 words a minute is a round convention for a prepared talk, not a measurement. Your own pace will differ.

The samples

Three public-domain passages, quoted word for word and point for point; line wraps are joined and nothing else is changed. The opening of Beatrix Potter’s The Tale of Peter Rabbit (1902; the author died in 1943), from Project Gutenberg’s edition, without its picture markers and with the four names run on one line. Abraham Lincoln’s Gettysburg Address (1863) in the Bliss copy, the one he signed, from Wikisource’s transcription of the manuscript. The first two paragraphs of the Introduction to Charles Darwin’s On the Origin of Species, first edition (1859), from Project Gutenberg’s edition. All three were read from those sources on 7 October 2026 and are built into the page; pressing a sample fetches nothing.

What happens to your text

It stays in this page. The script that measures it makes no request of any kind, there is no form to submit, and the text is never put in the address bar. This was checked by recording every request the page attempted while text was typed, pasted, measured, kept and forgotten, with the site’s analytics switched off as it is on any host but labs.llc: after the page had loaded, nothing more was requested. On labs.llc the site’s analytics tag, the same on every page, counts the visit and afterwards reports that the tool was first used, that the page was open and in use for thirty seconds, and, when you leave, up to three page-speed measurements: how soon the page painted, how far it shifted, and how long a key or a tap waited for an answer. Each of those names the part of the page concerned, which can be the text box itself, and never what is in it. None of the five carries a word, a count or a score from the box; see Legal.

Nothing is stored unless you press “Keep in this browser”. Then the text is saved in this browser’s own storage on this device, under the name labs.readability.text, so it is there when you come back; press the key again and it is removed. It is readable by anyone who uses this browser profile, so do not keep anything private on a shared computer.

Speed, and what this page does not know

The scores follow each keystroke. Measured on 7 October 2026 in headless Chrome on the laptop this page was built on, with 20,284 words in the box: each keystroke was re-measured and the six scores redrawn in 5.3 ms at the median and 7.8 ms at the slowest of 40 keystrokes, and the next frame was painted 18.4 ms after the key at the median. The marked copy below is redrawn about a quarter of a second after the typing stops (12 ms for that text). The same text without a single capital letter, which this page first read as one sentence and took about a second a key over: 5.2 ms at the median and 6.1 ms at the slowest of 15 keystrokes. A slower phone will take longer; the page prints its own figure under the box.

It does not know what your text means, whether it is true, or who will read it. It does not check spelling or grammar. It reads English only and says nothing useful about any other language. It is a measuring tool, not a judge of writing. Questions and corrections: labs@labs.llc.