Flesch-Kincaid Explained: What the Score Means and How to Use It

Paste a draft into any readability checker and the number that appears is almost certainly Flesch-Kincaid. It is in Word, in Hemingway, in Yoast, and in most state plain-language statutes.

It is also a formula with one exact half and one guessed half, and knowing which is which changes how much weight the number deserves.

The two formulas

Flesch-Kincaid is two formulas sharing an input.

Reading Ease, 0 to 100, higher is easier:

206.835 - 1.015 * (words / sentences) - 84.6 * (syllables / words)

Grade Level, mapped to US school years:

0.39 * (words / sentences) + 11.8 * (syllables / words) - 15.59

Both depend on exactly two ratios: average sentence length and average syllables per word. Longer sentences and longer words move both numbers the wrong way. There is nothing else in the model, which is worth remembering every time someone treats the output as a judgement about quality.

Our Readability Checker computes both, plus Gunning Fog and Coleman-Liau, in the page.

What the corpus looks like

Rather than repeat the benchmark table every article about this repeats, here is the same formula run over something you can go and read.

Thirty-one articles, from percentage arithmetic to block cipher modes, and every one of them lands between grade 5.7 and 9.1, with a median of 7.3. The article on AES-GCM nonce reuse is not meaningfully harder to read than the one on sale discounts, which tells you what the formula is measuring: sentence and word length, not subject difficulty.

That is the honest use of the number. It cannot tell you whether a reader will follow the argument.

Where the score comes from

Two people, twenty-seven years apart.

Rudolf Flesch published Reading Ease in the Journal of Applied Psychology in 1948¹, reacting to academic prose that had become genuinely unreadable. The formula was widely quoted before his book on the subject appeared the following year, and the paper is where the constants come from.

The grade-level half arrived in 1975, when J. Peter Kincaid and colleagues recast the formula for the US Navy, deriving it from comprehension tests on 531 enlisted personnel across four technical training schools². The Navy needed to know whether sailors could read their own training manuals, and "aim for grade 9" was an instruction a writing instructor could act on.

That dual origin explains the score's split personality. It is simultaneously a 1948 editor's rule of thumb and a 1975 procurement specification, which is why it shows up in both blogging plugins and contract law.

The half of the formula that is a guess

The arithmetic is exact. The syllable count is not, because English has no reliable syllable algorithm and never will. Standard heuristics count vowel groups, subtract a trailing silent e, and special-case a few suffixes. Run one over ordinary words and it gets science, being and poem all wrong in the same direction, and business wrong in the other.

The consequence is measurable. Running two defensible heuristics across our 31 articles moved the grade by 0.37 on average, and by 1.4 on the worst case, which is more than the distance between a newspaper and a magazine.

So a Flesch-Kincaid score is only comparable against another score from the same tool. Comparing Word's number to Yoast's number to ours is comparing three different syllable counters wearing the same formula.

What it is good for

• Catching a wall of long sentences. A 600-word section that comes in three grades above the rest of the piece almost always contains one enormous sentence, and splitting it is the whole fix.
• Comparing two drafts of the same text. Same writer, same tool, same subject: a two-grade move is real.
• Meeting a regulatory target. Several US states set Flesch minimums for consumer insurance contracts, and federal plain-language guidance leans on the same scores.
• Setting expectations. "This one is grade 12" is useful information for an editor, independent of whether it should be.

What it is not good for

• Comparing writers. The formula rewards short words. A specialist using the correct term once scores worse than someone circling it vaguely for a paragraph.
• Other languages. The constants were fitted to English. They do not transfer.
• Judging clarity. Nonsense scores well if the sentences are short. The formula cannot read.
• SEO. Google does not compute a reading level for your page and does not rank on one. Mueller has said so directly, and an analysis of more than 750,000 pieces of content across 30,000 queries found no correlation between reading level and position³. Readability still affects whether people stay and link, which is a real mechanism and a different one.

A workflow that respects the number without obeying it

1. Draft without thinking about the score.
2. Run the finished draft through a checker.
3. If the grade is more than two above your target, find the longest sentences and split them. That is nearly always the entire remedy.
4. If the grade is fine but the prose is dead, ignore the number and fix the prose.

Our Readability Checker shows four scores at once, which is more useful than one: when Flesch-Kincaid and Gunning Fog disagree, the gap usually points at a specific paragraph. For working paragraph by paragraph rather than re-pasting a whole document, the word counter handles the smaller unit.

The bottom line

Flesch-Kincaid answers one question well: are my sentences too long and my words too dense? It answers nothing else, and its input is an approximation that varies between tools by more than the differences people argue about.

Use it like a stopwatch. It confirms a hunch. It does not tell you where to run.