Readability scores explained: what they measure and what they miss

A readability score looks like a verdict on your writing. It is really a small piece of arithmetic. Paste a paragraph into the readability checker and it counts words, sentences, and either syllables or letters. Then it feeds those counts into a formula someone fitted decades ago. Knowing what goes into that formula tells you when to trust the number and when to shrug.

What the formulas actually count

Nearly every common formula rests on the same two measurements:

• Sentence length. Average words per sentence. Long sentences ask the reader to hold more in mind before the point arrives.
• Word length. Average syllables per word (Flesch, Flesch–Kincaid), the share of words with three or more syllables (Gunning Fog, SMOG), or average letters per word (Coleman–Liau, the Automated Readability Index).

Neither count is what makes text hard. They are proxies. Long words tend to be rarer, more abstract, and more technical. Long sentences tend to pack in more clauses and more ideas. The proxies correlate with difficulty well enough to be useful across large samples, and that is all they were ever meant to do.

Why they were invented

The formulas grew up in the mid-twentieth century, when schools, publishers, newspapers, and the armed forces needed a quick way to match text to readers. Rudolf Flesch published his Reading Ease formula in 1948. Robert Gunning’s Fog index appeared in 1952. The US Navy commissioned the Flesch–Kincaid grade formula in 1975 to check training manuals for enlisted personnel.

The method was the same each time. Researchers took passages whose difficulty they already knew, often from reading tests, and measured them. Then they found the weights for sentence length and word length that best predicted that difficulty. A formula is a fitted line, not a law of language.

What the formulas miss

Because they see only counts, the formulas are blind to almost everything an editor cares about.

Meaning. Shuffle the words in each sentence and the score stays the same. A formula cannot tell sense from nonsense.

Order. A paragraph that buries its main point at the end scores the same as one that leads with it. Readers notice the difference at once. (See putting the main point first.)

Familiarity. “Television” has four syllables and every reader knows it. “Wont” has one, and plenty of readers will stumble. Short jargon such as “PII,” “CTA,” or “KPI” scores as easy and often isn’t.

The reader’s knowledge. A cardiologist reads a dense clinical note faster than a plain summary written for patients. The formulas assume one generic reader.

Layout. Headings, lists, white space, and tables make long documents easier to use. The formulas usually ignore them, and some tools count a bulleted list as one giant sentence.

Here is a pair that shows the blind spot. Both sentences have the same short words.

Before: The dog bit the man who fed the cat that the boy saw.

13 words · 1 sentence · grade 1.3

After: The boy saw a cat. A man fed it. The dog bit the man.

14 words · 3 sentences · grade -2.0

The first is a legal, short-worded sentence that most readers have to reread. The second splits it into three plain statements. The formula does reward the split, but only because the sentences got shorter. It has no idea the first version was a tangle.

Why different tools give different numbers

Paste the same text into three checkers and you may get three grades. That is normal. The differences come from the counting, not the formulas.

• Syllables. No two programs count syllables quite alike. “Fire,” “every,” and “business” each have more than one defensible count. Counting syllables walks through the hard cases.
• Sentences. Does “Dr. Lee arrived at 3 p.m. on Jan. 4.” contain one sentence or four? A naive splitter sees periods and says four. Headings, bullet points, and lines without punctuation cause similar trouble.
• Words. Tools disagree on whether “e-mail,” “3.5,” or “don’t” is one word or two. Our own rules are in how words are counted.
• Cleanup. Some tools strip URLs, code, and numbers before scoring. Others leave them in.

Even with identical counts, two formulas can disagree, because they weight the two measures differently. Flesch–Kincaid leans harder on sentence length than Flesch Reading Ease does. A text made of long sentences and short words can come out “easier” on one and “harder” on the other.

Tip: Compare scores only when they come from the same tool and the same formula. A drop from grade 12 to grade 9 in one checker means something. Grade 9 in one tool against grade 11 in another means very little.

Our methodology page explains how Op-Ed Insight counts sentences, words, and syllables.

How to use a score sensibly

Treat a readability score as a smoke alarm. When it goes off, look for the fire. Don’t rip out the alarm, and don’t assume a quiet alarm means the house is fine.

1. Write first, score second. Draft for your reader. Check the number after.
2. When the score is high, look at the cause. Is it a few very long sentences? The sentence length checker will find them. Is it long words? Check whether they are jargon you can replace (our plain words list helps) or terms your reader needs.
3. Fix the writing, not the number. Splitting a sentence at a natural break helps the reader. Chopping it in half mid-thought, or swapping a precise word for a vague short one, helps only the score.
4. Use the trend. A score is most useful across drafts of the same document, or across documents in the same series.
5. Test with real readers when it matters. For forms, instructions, and health or legal information, watching a few people use the text beats any formula.

For targets by audience, and why aiming at an exact number tends to backfire, see what readability score to aim for.