Readability scores: what they measure, and the one thing they are good at

Paste the text into PLAIN, which gives four scores with what they are worth, and then the part that helps: the sentences over your threshold listed longest first, the passive constructions, the verbs that became nouns, and the long words that have exact short equivalents. Nothing is uploaded, which matters for an unpublished draft.

What the four scores measure

All of them are arithmetic on two things: how long the sentences are, and how long the words are. They differ in how they measure word length and how they combine the two.

Flesch reading ease is the one most people have seen: above 60 is generally readable prose, below 30 is heavy going. The other three output something that looks like an education level, which gives them a false air of authority — they are not measuring comprehension, they are reporting a function of two averages in units of schooling.

Why the disagreement is informative

Because the scores weight sentence length and word length differently, a large gap between them tells you which problem you have.

• Fog much worse than Flesch–Kincaid: long words in short sentences. Typically jargon, or a document stuffed with nominalisations. The fix is vocabulary.
• Flesch–Kincaid much worse than fog: short words in long sentences. Typically clause after clause joined by commas. The fix is full stops.
• Both bad: official prose, and the fix is both.

That is genuinely more useful than any single number, and it is why showing four is better than showing one.

Why writing to a score makes prose worse

Because both inputs can be gamed without improving anything, and gaming them usually damages the writing.

Sentence length falls if you cut sentences in half regardless of whether the halves belong together. “The claim was refused. This was because the form was late.” scores better than “The claim was refused because the form was late” and reads worse, because the conjunction was carrying the logic and has been deleted.

Word length falls if you replace a precise long word with a vague short one. Replacing “anticoagulant” with “blood thinner” may be right for a patient leaflet and is plainly wrong in a prescribing note. The score cannot tell the difference because it cannot read.

And the deeper point: none of these measures meaning. A paragraph of fluent, grammatical nonsense scores beautifully. A dense, precise, carefully qualified paragraph of legal or clinical writing scores badly and may be exactly right. The score is blind to the only thing that matters.

The one thing they are reliably good at

Finding the sentence that is too long. Not the average — the specific one.

This works because it is not really a readability judgement at all. It is a count, and a forty-eight word sentence is a fact about the document rather than an opinion about it. What makes it valuable is that this particular fact is invisible from the inside: the writer knows what the sentence means, so it does not feel long to them. It feels long to everybody else, and it is where readers stop.

So the useful output of a readability check is not the number at the top. It is a list of sentences, longest first, that you can go and look at. Most of them will split naturally at a comma or a conjunction, and the document improves immediately without a single word being simplified.

Twenty-five words is a reasonable threshold for general prose. Technical and legal writing legitimately runs longer; set the threshold where your own writing sits and look at the tail.

The passive voice, properly understood

The passive is not an error and the blanket advice against it is wrong. It is the right choice when the actor is unknown, irrelevant, or obvious: “the samples were refrigerated overnight” is better than inventing somebody to have done it.

It is a problem in one specific class of document: anything that tells somebody what to do or records what somebody did. In a procedure, an instruction, or an account of an incident, the actor is the fact the reader needs, and the passive is precisely the construction that removes it. “The form is checked” leaves every reader assuming somebody else checks it.

So the right question is not “is this passive?” but “has this hidden somebody who matters?”

Nominalisation, which does more damage than either

A nominalisation is a verb turned into a noun: decide becomes decision, assess becomes assessment, implement becomes implementation.

The damage is structural rather than stylistic. Once the verb has become a noun, the sentence needs a new verb to carry it, and the available ones are empty: make, carry out, undertake, effect. So “we decided” becomes “a decision was made” — longer, vaguer, and with the actor now optional and usually omitted.

This is the mechanism behind most official prose. It is not that officials like long words; it is that nominalising every verb makes the actor grammatically unnecessary, and a sentence with no actor is a sentence nobody is accountable for. Reversing it — finding the nouns ending in -tion, -ment, -ance and turning them back into verbs — usually shortens a paragraph by a third and puts people back into it.

Words with exact short equivalents

Not a style rule, and worth separating from the general advice to use short words. A small set of long words have shorter ones that mean exactly the same thing in every context: utilise is use, commence is start, prior to is before, in order to is to, in the event that is if.

These are free improvements: nothing is lost, the sentence gets shorter, and the register gets less stiff. That is different from replacing a technical term with an approximation, which costs precision. Keeping the two categories apart is what stops plain-English advice turning into dumbing down.

How to use a readability check

1. Look at the longest sentences. Split the ones that split naturally.
2. Look at the passives in anything instructional and put the actor back.
3. Turn the nominalisations back into verbs.
4. Swap the free words.
5. Ignore the score. It will have improved anyway, and it was never the point.