Flesch formulas estimate how hard text is to read using two things only: average sentence length and average syllables per word. They know nothing about meaning. They’re useful as a quick check for long sentences and heavy vocabulary, but a high score doesn’t make writing good, and a low one doesn’t make it bad. Google has indicated it doesn’t rank pages by such formulas, so write for your readers and don’t tune a number.
The two formulas
Both use the same inputs:
- ASL, average sentence length: words ÷ sentences
- ASW, average syllables per word: syllables ÷ words
Microsoft’s documentation for Word gives them as:
Flesch Reading Ease = 206.835 − (1.015 × ASL) − (84.6 × ASW)
Flesch-Kincaid Grade = (0.39 × ASL) + (11.8 × ASW) − 15.59Rudolf Flesch published the Reading Ease formula in 1948, and the Grade Level version came from work for the US Navy in the 1970s. Look at the weights: one extra syllable per word costs 84.6 points, whereas one extra word per sentence costs about 1. Vocabulary dominates the score.
Reading Ease goes up when text gets easier, and Grade Level goes up when it gets harder. They’re not on the same scale, so a page can’t be compared across them.
A worked calculation
Take this text:
The cat sat on the mat. It was a very sunny day outside.
Count by hand. There are 13 words and 2 sentences. Syllables: The, cat, sat, on, the, mat, It, was, a are one each (9). ve-ry, sun-ny and out-side are two each (6). day is one. That totals 16.
Get the averages.
ASL = 13 ÷ 2 = 6.5
ASW = 16 ÷ 13 = 1.2308Apply Reading Ease.
206.835 − (1.015 × 6.5) − (84.6 × 1.2308)
= 206.835 − 6.5975 − 104.123
= 96.1Apply Grade Level.
(0.39 × 6.5) + (11.8 × 1.2308) − 15.59
= 2.535 + 14.523 − 15.59
= 1.5So: very easy, around second grade. Our Readability Scorer gave the same results, 96.1 and 1.5.
How to read the scale
Flesch’s own bands are commonly described like this:
| Reading Ease | Description | Rough school level |
|---|---|---|
| 90 to 100 | Very easy | 5th grade |
| 80 to 90 | Easy | 6th grade |
| 70 to 80 | Fairly easy | 7th grade |
| 60 to 70 | Standard | 8th and 9th grade |
| 50 to 60 | Fairly difficult | 10th to 12th grade |
| 30 to 50 | Difficult | College |
| 0 to 30 | Very difficult | College graduate |
The scale isn’t truly bounded, despite how it’s usually shown. Scores can exceed 100 or go below 0, as the examples below show. Some tools clamp them to 0 to 100, and that’s one reason two tools can report different numbers for the same text.
Proof that it ignores meaning
We ran four short samples through our scorer. They make the limits clear.
| Text | Reading Ease | Grade |
|---|---|---|
| The cat sat on the mat. It was a very sunny day outside. | 96.1 | 1.5 |
| Mat the on sat cat the. Outside day sunny very a was it. | 96.1 | 1.5 |
| Set the TTL on the DNS zone to one day. Then flush the CDN edge and the WAF cache. | 112.6 | 0 |
| The utilization of comprehensive documentation facilitates organizational understanding. | −118.5 | 31.8 |
Row 2 is the same words scrambled. It’s nonsense, and it gets an identical score, because the formula has no notion of order or sense.
Row 3 scores as easier than a children’s book, though most readers would have no idea what it means. Short acronyms count as one syllable each. Jargon in short words is invisible to the formula, and that’s a big problem for technical content.
Row 4 scores far below zero for one wordy sentence with 8 words and 30 syllables. It’s hard to read, and the formula gets that right. Its plain version, “Good documents help the company understand”, conveys much the same thing.
Other limits
Syllables are estimated. No tool has a pronunciation for every word, so most use spelling rules, and different rules mean different scores. That’s why Word and various websites differ slightly on the same text. Treat a difference of a point or two as noise.
Sentence detection is imperfect. Dr., e.g., decimals and lists all confuse it. A bulleted list without periods can look like one enormous sentence, dragging the score down for no good reason. Headings cause similar trouble.
They were built for continuous English prose. The coefficients come from English reading tests. Using them on other languages gives a number that means little, even when the alphabet is the same. A German sentence we tried scored 103.2 in our tool, which measures only that its syllable estimator was misled. Other languages have their own adapted formulas.
Audience knowledge isn’t measured. The word “anticoagulant” is hard for most people and routine for a nurse. A doctor reading a medical article written at a sixth-grade level may find it less useful, not more.
Short samples are unstable. One unusual word moves a 50-word sample by several points. Score at least a few hundred words before you draw conclusions.
Optimizing can make writing worse. You can raise a score by chopping sentences into fragments, and that reads poorly. Cutting a precise long word for a vague short one can also hurt clarity. Over-short sentences can harm flow, and the formula can’t see that.
What Google has said
Google doesn’t list readability scores among its documented ranking systems. The most direct public comment we could find is from a 2018 Webmaster Central office-hours session, as reported by Search Engine Journal. Asked about reading level, a Google Search representative said that as far as he knew, Google doesn’t have basic algorithms that count words to work out reading level. He added: “From an SEO point of view, it’s probably not something that you need to focus on… But it is something that you should figure out for your audience.”
Note the hedged wording. It’s a spoken comment from some years ago and not formal documentation, and it’s not a promise about every system Google runs. We could only find it in reports, not an original transcript. It does fit Google’s documentation, whose advice is to create helpful, people-first content and which does not mention any readability score.
So: there’s no target Flesch score for SEO. A plugin that shows a red light for a score of 55 is not reporting a Google rule. If clearer writing helps visitors understand and stay, that’s a benefit worth having, but it isn’t the same as a ranking signal, and we won’t claim a measurable effect we haven’t got.
What our tool does
Our Readability Scorer computes Flesch Reading Ease, Flesch-Kincaid, Gunning Fog, SMOG and Coleman-Liau in your browser, along with sentence length distribution, likely passive voice, and difficult words. It estimates syllables by spelling rules, so expect small differences from Word.
It won’t produce English scores for text that’s mostly outside the Latin alphabet, but it will score other Latin-script languages, and those numbers aren’t meaningful. Remove navigation, code and lists before you paste, as they can distort the sentence counts. If headings are mixed in with your text, check their structure separately with the Headings Checker.
How to use a score properly
Use readability scores as a smoke alarm, not a target.
Compare your text with itself. Score your draft, edit, and score again. A move from 35 to 55 on a general-audience article reflects something real, even if the absolute number is arguable.
Set a range for the audience. For a general audience, the band Flesch called standard, 60 to 70, is a reasonable starting point. A legal or scientific audience can accept far less.
Look at the causes, not the score. Two things move it: long sentences and long words. Find your longest sentences and split the ones that have two ideas. Replace long words only when a short one means the same thing: “use” for “utilize” works, and “use” for “anticoagulant” doesn’t.
Then test with a real person. Ask someone from your audience to read it and tell you what it says. That beats any formula, and it finds problems like the acronym sentence above, which a formula will rate as easy.