A readability score is a description of how hard a text is to read, produced from two or three countable things: sentence length, word length, and how many long words there are. Every formula in this guide uses some subset of those, and none of them knows what the text means. That is the whole of the technique and also the whole of its limitation — the formulas are useful when a text has to reach a stated audience and is not reaching it, and misleading when a single number is treated as a verdict on the writing. This guide works through the five, then the structure measures underneath them. Everything runs in your browser.
Readability is not one number
The five established formulas routinely disagree by two or three grades on the same passage, and that is not a bug in any of them — they were derived from different populations for different purposes. Flesch Reading Ease and Flesch-Kincaid come from US Navy technical manuals; Gunning Fog from newspaper training; SMOG from comprehension testing in health information; the Automated Readability Index from keystroke counting on typewriters. SMOG ignores sentence length entirely and counts only polysyllables; Fog treats long sentences and long words as equally expensive; the Flesch family weights vocabulary about thirty to eighty times as heavily as sentence length. Every formula page in this category therefore runs all five on your text and prints them side by side. The disagreement is the finding; a lone number invites you to treat it as a measurement.
The two measurements everything is built on
Sentences and syllables. Sentence splitting sounds trivial and is not: a counter that tallies full stops is thrown off by “3.5 pages”, by “et al.”, by initials and by ellipses, and it entirely misses a final sentence that ends without a terminator — which is common in lists and headings. The shared splitter in this category masks decimal points, abbreviations and initials before it splits, then treats a run of terminators as a single end. Syllables are the harder half. Counting vowel groups with a rule for a silent final e and a rule for the consonant-plus-le ending gets most ordinary prose right, but it is a heuristic, and words like queue, business and every are off by a syllable. Every Flesch-family and Fog figure carries that approximation, which is why the bands are labelled estimates rather than measurements.
Why the Automated Readability Index is the reference
ARI counts letters and words instead of syllables, both of which are countable exactly, so its result is reproducible to the digit on the same text with no heuristic anywhere in it. That makes it the one formula in the set that can be used to sanity-check the others: when ARI and Flesch-Kincaid disagree by more than a grade or two, the words in the text are either longer than they are difficult or shorter than they are hard. The trade is that ARI knows nothing but length — a passage of long simple words scores as harder than it reads, and a short technical term scores as easy. It is also the only one of the five that counts letters rather than characters and words rather than non-space characters, matching the keystrokes the index was designed around.
SMOG's sample size is a real requirement
SMOG scales its polysyllable count to a 30-sentence sample, and that 30 is a requirement of the original method rather than a constant of the language. The consequence is practical: on a whole document the figure is stable, and on a paragraph it moves by a whole grade when a single long word appears. Paste in 600 to 700 words of prose and the number is the formula; paste in a paragraph and it is an approximation of the formula, which is what the tool says plainly when the input is short. If a standard specifies SMOG, it specifies the sample size too.
Sentence length is a distribution, not an average
Two texts can both average eighteen words a sentence while one is a steady eighteen and the other alternates between five and forty — and it is the second that reads badly. This is why the Sentence Length Analyzer reports a histogram, a median and the longest sentences as well as the mean. The useful habits are visible in the shape: too many short sentences in a row reads as choppy, too many long ones as impenetrable, and the single forty-word sentence carrying three ideas is usually the paragraph's actual problem. A handful of outliers drags the mean and leaves the median alone, which is exactly the case where the average misleads.
Vocabulary measures need their length dependence understood
Type-token ratio — distinct words over total words — is the standard measure of how varied a vocabulary is, and it falls as a text grows, because the words already used keep reappearing while the supply of new ones runs out. Comparing a 200-word paragraph with a 2,000-word essay on that number compares lengths rather than vocabularies. The Vocabulary Diversity Calculator reports the counts the ratio is made of and the most frequent words alongside it, which is where the number becomes actionable: a low ratio is usually a handful of words carrying the whole text. Lexical density has the same shape of caveat. It splits content words from function words using a stop-word list rather than grammatical tagging, so a listed word doing real work — “one” in a maths passage — counts as a function word. Academic writing runs at 40–50% and conversation nearer 30–40%; the tool lists both sides of the split so the heuristic can be inspected rather than trusted.
Turn the score into a plan
A readability grade is a diagnosis; the reading arithmetic is what follows from it. The Reading Speed Calculator turns a timed passage into words per minute and names the band it falls in, the Pages per Hour Calculator turns that into seconds per page — the figure people actually plan with — and the Book Completion Calculator combines the pace with the minutes a day you really have to give a calendar date. That last one usually produces the surprise: either a book is a fortnight of ten-minute sittings, or it needs an hour a day for a month and was never going to fit the plan it was assigned to. Both answers are more useful than the intention to read more.