A client once sent back a contract summary with a single comment in the margin: "Flesch says this is grade 17. Fix it." The document was three paragraphs of unavoidable legal precision, and no score in the world was going to make jurisdiction clauses breezy. But that number stuck in her head like a bad grade, and she assumed the writing had failed.
It hadn't. The measurement had.
Readability scores are among the most widely used and most widely misunderstood numbers in writing. Every content management system, word processor, and style guide seems to have an opinion about your Flesch score. Before you let a formula edit your prose, it helps to know what these formulas are actually doing under the hood, because the answer shapes when to trust them and when to ignore them politely.
The formulas all count the same two things
Nearly every classic readability formula, and there are dozens, boils down to a ratio between word length and sentence length. That is not a simplification of my own. It is literally what they measure.
Rudolph Flesch published his Reading Ease formula in 1948, aimed at a market flooded with claims about which newspapers were easiest to read. His insight was that you can estimate how hard a text is without asking a single human reader: count syllables per word, count words per sentence, do some arithmetic. The Flesch-Kincaid Grade Level version, developed later for the US Navy, rescales the same inputs into a school grade number. Gunning Fog adds the share of words with three or more syllables. SMOG was built specifically for health education materials. Coleman-Liau counts letters instead of syllables because letters are cheaper for a machine to count.
| Formula | Output | What it weighs |
|---|---|---|
| Flesch Reading Ease | 0–100, higher is easier | Syllables per word, words per sentence |
| Flesch-Kincaid Grade | US school grade | Same inputs, rescaled |
| Gunning Fog | US school grade | Long words (3+ syllables), sentence length |
| SMOG | US school grade | Polysyllabic words, sample of 30 sentences |
| Coleman-Liau | US school grade | Letters per word, sentences per 100 words |
| ARI | US school grade | Characters per word, words per sentence |
Notice what never appears in that table: meaning. None of these formulas know whether your argument is clear, whether your examples land, or whether your reader cares. They measure the texture of the language, not the quality of the thinking.
What the numbers are actually good at
Dismiss them entirely and you lose something real. These formulas are excellent tripwires.
A sudden jump in average sentence length usually means you have started stuffing clauses on top of clauses, the habit that creeps in when a writer is tired or is padding toward a word count. A spike in syllables per word often catches jargon inflation, that drift from "use" to "utilize" that nobody decides on but everybody commits. A grade level that sits far above your audience is a genuine warning sign worth investigating.
For high-volume work, the triage value is enormous. Government agencies have used SMOG to screen patient information leaflets for decades, on the reasonable principle that a health warning nobody can parse is worse than no warning. If you edit a content library with five hundred articles, running every one through a readability checker and sorting by score will surface your problem documents in seconds.
Where the scores mislead you
Here is the part most listicles skip. The formulas have systematic blind spots, and they fail in predictable directions.
Short words can still be opaque. "The asset class allocation reflects a duration mismatch against the liability profile" scores beautifully. Every word is short. The sentences are trim. And almost nobody outside finance knows what it means. Vocabulary familiarity, not vocabulary length, drives comprehension, and syllable counting cannot see the difference.
Long sentences are not automatically bad. A well-constructed sentence of forty words, with parallel structure and clear signposting, can carry a reader further than three choppy fragments that refuse to connect their ideas. The formulas punish the good long sentence exactly as hard as the bad one.
Proper nouns and technical terms inflate everything. A medical report about "cholecystokinin" will always score as graduate-level prose no matter how plainly it is written. A word counter with keyword analysis will show you these outliers faster than a score will.
They estimate syllables, they don't know them. Most implementations, including browser-based ones, guess syllables with heuristics like counting vowel groups. "Queue" has four vowel letters and one syllable. "Barely" fools many counters. This is why two checkers can disagree by a full grade level on identical text, and why you should compare a document against itself over time rather than trusting absolute values.
The audience question beats the formula question
What grade level should you write at? The honest answer: lower than you think, for almost every audience, but for a reason the formulas obscure.
The conventional wisdom says general audiences sit around grade 8, and plain-language advocates have pushed that target for decades. But readers do not have one fixed reading level. They have a level for reading novels on a Sunday and a much lower tolerance on a phone, in a hurry, mid-task, with a toddler in the background. Your mortgage application instructions are being read by someone at their worst, not their best. Writing short sentences and common words is not condescension. It is meeting the actual conditions your text will be consumed in.
That reframing also dissolves the "dumbing down" worry. You are not removing ideas. You are removing friction between the reader and the ideas. When in doubt, keep the sophisticated concept and simplify the sentence carrying it.
A practical workflow that uses scores without obeying them
Treat readability numbers the way a pilot treats an altimeter: essential input, zero authority over decisions.
- Draft freely. Do not check anything mid-draft. Formulas in your head while writing produce anxious, stunted prose.
- Measure once the draft is done. Note the headline scores, average sentence length, and the longest sentences by name.
- Read the flagged sentences, not the score. Good checkers list your longest and most complex sentences directly. Almost every real fix happens there.
- Cut or split only where the sentence genuinely fights you. Usually two or three sentences per thousand words are the actual culprits.
- Re-measure and stop. If the number moved toward your baseline, you are done. Chasing single digits is procrastination with a dashboard.
Frequently Asked Questions
What is a good Flesch Reading Ease score?
For general web content, somewhere between 60 and 70 is the usual target, which corresponds to plain English readable by a early-teen student. Marketing copy often aims higher, 70 to 80. Academic and legal writing legitimately sits lower. The right question is not "what is good" but "what fits my readers and my genre."
Why do two tools give different scores for the same text?
Syllable counting. Most tools estimate syllables with different heuristics, and hyphenated words, proper nouns, and numbers are counted inconsistently across implementations. Pick one tool and use it consistently so your comparisons over time stay meaningful.
Do readability scores affect SEO directly?
No evidence supports readability as a direct ranking factor. It affects SEO indirectly and substantially: readable pages earn longer visits, more shares, and more links. Google's own guidance on helpful content emphasizes writing for people, which is the same advice the formulas approximate badly.
Should I rewrite old content to a lower grade level?
Only where the content underperforms and the sentences are genuinely tangled. Blanket simplification of a whole archive destroys voice and burns time. Use scores to find the worst ten documents, then fix those by hand.
Can these formulas detect bad AI-generated text?
Sometimes, by accident. Machine-written prose tends to hover in a narrow band of sentence lengths, which shows up as a suspiciously uniform distribution rather than a bad score. The flatness is the tell, not the number itself.
The one number to watch
If you take a single metric from any checker, make it the count of very long sentences, those over 30 or 35 words. Almost every readability failure a formula catches traces back to a sentence that tried to do four jobs at once. Find those sentences, decide which job each should actually do, and let the score take care of itself. The formulas were built in an era before machines could read meaning. Use them as smoke detectors, not as judges.