What a readability score can and cannot see

A readability score puts a number on prose, and a number invites a target. Before you set one, it is worth knowing exactly what is being counted, because the formula is far simpler than its reputation and it is measuring two things only.

What is actually being counted

The Flesch Reading Ease formula takes two ratios: the average number of words per sentence, and the average number of syllables per word. It multiplies each by a fixed constant, subtracts both from another fixed constant, and reports the result on a scale where a higher number means easier. The companion grade level uses the same two ratios with different constants and reports something that looks like a US school year.

That is the whole thing. Sentence length and syllable counts, used as a proxy for difficulty. The constants were fitted decades ago against reading tests, and they are frozen. Nothing in the formula looks at vocabulary, at whether a word is common or obscure, at grammar, at the order of ideas, or at whether any sentence means anything.

Which makes the failure mode obvious

Two ratios can be satisfied by text that no reader could follow. Short sentences of short words score well whether or not they are connected: a list of disconnected assertions with the joining words stripped out scores better than the coherent paragraph it was cut from, because the connectives that made it coherent were long sentences.

It runs the other way too. A perfectly clear sentence with a long technical noun in it — the name of a medication, a legal term, a place name — is penalised for the syllables, even though the whole point of the sentence is that specific word. Rewriting to avoid it makes the score better and the sentence worse.

And the score cannot see structure. It has no idea whether you defined a term before you used it, whether the paragraphs are in a sensible order, whether the example arrives before or after the thing it illustrates, or whether the reader has been told why any of this matters. Those are most of what makes writing readable, and the formula is blind to all of them.

How the counting is approximated

Worth knowing, because it is where small surprises come from. Sentences are counted by splitting on full stops, question marks and exclamation marks, so an abbreviation with a full stop in it reads as a sentence boundary, and a run of bullet points with no punctuation reads as one enormous sentence. Words are the whitespace-separated pieces that contain at least one letter.

Syllables are the hard part, because English does not spell them consistently, and every implementation guesses. The Readability Score tool guesses like this: any word of three letters or fewer counts as one syllable; otherwise it counts groups of adjacent vowels, treating y as a vowel, and then subtracts one for a silent trailing e if that would not take the count to zero.

That rule handles ordinary words well and is wrong at the edges by design. "Area" has its vowels adjacent and is counted short. "Business" counts a group that nobody pronounces. No syllable heuristic gets these right without a pronunciation dictionary, and the error is small across a few hundred words and large across one sentence. Which is the first practical rule: score a body of text, never a line.

Where it is genuinely useful

As a flag on your own draft, run after you have written it and before you have read it back. A grade level far above what you intended usually means one specific, findable thing: a few sentences that ran away from you, with three clauses where one would do. The number does not tell you which sentences, but it tells you to go and look, and the reported average words per sentence tells you whether that is the ratio at fault.

It is also useful comparatively. The same text before and after an edit, or two drafts of the same page, measured the same way. The absolute value carries little meaning; the direction of the change carries some.

Where it is useless

As a target. The moment a score becomes the goal, the cheapest ways to move it are the ones that damage the writing: chopping sentences at their connectives, replacing a precise word with a vaguer shorter one, deleting the qualification that made a claim honest. All three improve the number. All three make the page worse, and none of them is visible to the thing you are optimising.

It is also not a measure of quality, an audience test, or anything a search engine is documented to read. A page can be dull, wrong and unhelpful at any score you like.

The one-line version

Use it the way you would use a smoke alarm: something that tells you to go and look, not something you tune until it stops making noise. The only real test of whether a sentence is clear is whether somebody who did not write it understands it the first time, and no formula performs that test.