To improve a draft's readability: find its longest sentences and split each at a natural break, replace long words with shorter ones that mean the same thing, put the point of each paragraph in its first sentence, and break walls of text with subheadings and lists. Then measure — Flesch Reading Ease, Flesch–Kincaid Grade and Gunning Fog all estimate difficulty from sentence length and word length, and they are useful as a check that your edits changed the text in the direction you intended. They cannot judge whether the text is true, clear in its ideas, or right for its audience. Run the numbers on your own draft with the Content Analyzer, and treat every score below as a diagnostic, never a grade for the writing itself.

What the three scores measure — and what they do not

All three formulas use the same two ingredients: average sentence length (words ÷ sentences) and word length measured in syllables (Flesch family) or the share of complex words of three or more syllables (Fog).

  • Flesch Reading Ease = 206.835 − (1.015 × average sentence length) − (84.6 × average syllables per word). Higher is easier; most general web writing sits somewhere in the 50–80 range. Rudolf Flesch published the formula in 1948.
  • Flesch–Kincaid Grade = (0.39 × average sentence length) + (11.8 × average syllables per word) − 15.59. The result is expressed as a US school grade level.
  • Gunning Fog = 0.4 × (average sentence length + percentage of complex words). Also expressed as a grade level; it punishes long words more directly.

What none of them can see: whether a short word is the right word, whether the argument makes sense, whether a term is familiar to your specific readers, or whether a sentence is short but tangled. A page of three-word sentences of nonsense scores beautifully. That is why the workflow below edits first and measures second.

Tested example: one passage, edited and re-scored with the published formulas

We wrote a deliberately dense passage, scored it with a small script implementing the three formulas exactly as published above (syllables counted by vowel-group rules), edited it, and scored the revision. Both texts are quoted in full so you can reproduce the counts.

Before (64 words, 1 sentence):

"The implementation of a comprehensive image optimisation strategy necessitates the careful consideration of numerous interrelated factors, including, but not limited to, the selection of an appropriate file format, the utilisation of responsive delivery mechanisms, and the ongoing evaluation of compression parameters in order to ensure that visual quality is maintained whilst file sizes are simultaneously reduced to the greatest extent that is practically achievable."

After (56 words, 5 sentences):

"Good image optimisation rests on three choices. Pick the right format for each image. Serve a size close to the size on screen, not the full original. Then set compression by eye: lower the quality until you can just see a flaw, then raise it one step. Check the result at full size before you publish."

Measured scores (our script, formulas as published):

Measure Before After
Average sentence length 64.0 words 11.2 words
Average syllables per word 2.22 1.34
Complex words (3+ syllables) 23 of 64 4 of 56
Flesch Reading Ease −45.8 82.2
Flesch–Kincaid Grade 35.6 4.6
Gunning Fog 40.0 7.3

Yes, the "before" scores are off the familiar scales — a negative Reading Ease and a grade level of 35 are what the formulas honestly return for a 64-word sentence full of three-syllable words, and tools that clamp the display to 0–100 hide that. The edits that produced the change were mechanical and small: one sentence became five; "necessitates the careful consideration of" became nothing (the idea moved into a plain opening claim); "utilisation of responsive delivery mechanisms" became "serve a size close to the size on screen"; and the trailing purpose clause became an instruction the reader can act on. No idea was removed — the revision says the same three things (format, size, compression) in words a reader can hold in one pass. Note also what the scores still miss: they cannot tell you the revision's advice is correct. Accuracy needs a different check entirely — see How to Check Facts in AI-Assisted Writing.

An editing workflow that uses scores honestly

  1. Read the draft aloud, or have it read to you. Mark every place you run out of breath or have to restart a sentence. That list, not a formula, is your first edit queue.
  2. Split the longest sentences. Find sentences over about 25 words. Most contain two ideas joined by "and", a comma chain, or a trailing clause; split there. Keep some longer sentences — uniform length reads as robotic; variety is natural.
  3. Cut complex words with plain equivalents. "Utilise" → "use", "in order to" → "to", "a large number of" → "many". Keep technical terms your audience actually uses; the target is needless weight, not precision.
  4. Front-load paragraphs. The first sentence should carry the paragraph's point; detail and caveats follow. Readers skimming first sentences should still get the argument.
  5. Structure for scanning. Subheadings that say something ("Split the longest sentences", not "Tips"), lists for genuinely parallel items, and one idea per paragraph.
  6. Score the draft, then re-read instead of re-chasing. If the score improved and your aloud-read still stumbles somewhere, trust the stumble. If the score is low because the topic genuinely needs its terminology — a legal or medical text for specialists — that can be the right score for that audience.
  7. Check the facts separately. Readability passes and accuracy passes are different passes. A fluent, well-scored paragraph can carry a wrong date or an invented figure smoothly past a readability-only review.

Common mistakes

  • Writing to the score. Chopping every sentence to ten words and purging every long word produces text that scores well and reads badly. The score is a smoke detector, not the fire brigade.
  • Treating grade levels as reader ages. A "grade 8" result does not mean the text suits 13-year-olds in every way; the formulas know nothing about background knowledge, layout or motivation.
  • Comparing scores across tools without checking the counting. Syllable counters differ on words like "optimisation" and hyphenated compounds, so two tools can disagree by several points on the same text. Compare like with like — same tool, same counting rules — as we did above.
  • Editing the introduction only. Scores are computed over the whole text. One polished opening over dense body text is lipstick, and the average will say so.
  • Confusing readability with quality. Readability formulas estimate surface difficulty. They detect neither AI authorship nor errors, and our analyzer states that limit on its own page.

Frequently asked questions

What Flesch Reading Ease score should I aim for?

For a general web audience, somewhere around 60–80 is a reasonable zone to check against — but aim at the edits (shorter longest sentences, plainer words), not the number, and let audience and topic move the target. Specialist readers handle — and expect — denser text.

Why do two readability tools give different scores for my text?

Because the formulas are exact but the counting is not standardised: syllable estimation, sentence splitting around abbreviations and decimals, and treatment of headings and lists all vary by implementation. Small differences are normal; use one tool consistently for before/after comparisons.

Can a text be too easy?

Yes, for its audience. Oversimplified text can strip out distinctions that matter and read as patronising. The formulas reward shortness blindly; the editor's job is to stop where clarity stops improving.

Does readability affect search rankings directly?

Search engines do not publish a readability-score ranking input, and this guide makes no such promise. The defensible claim is simpler: text more people can finish and understand serves readers better, and readability work is worthwhile on those grounds alone.

How does readability relate to topic structure?

Readable pages still fail if readers cannot find the next thing they need. Pair sentence-level editing with site-level organisation — our guide to topic clusters and internal links covers the structure half of the problem.

References