Make sure your AI output actually reads at the right level
Language models frequently default to long, formal, multi-syllable prose that scores several grades above what a general audience comfortably reads. This tool runs your LLM output through four established readability formulas — Flesch Reading Ease, Flesch-Kincaid Grade Level, Gunning Fog, and SMOG — and tells you whether it matches your target audience, then flags the sentences dragging the score up.
How it works
Paste the text and the tool tokenizes it into sentences and words, estimates syllables with a vowel-group heuristic, and counts complex words (three or more syllables). From those counts it computes all four indices. Flesch Reading Ease returns a 0-100 score where higher is easier; the other three return a US grade level. You pick a target grade, and the tool shows how far each index is from it. Finally it ranks individual sentences by length and difficulty so you know exactly which ones to simplify.
What each formula measures
The four indices differ slightly in what they penalise and how heavily:
Flesch Reading Ease (0-100, higher is easier) measures sentence length and syllables per word. A score of 60-70 is appropriate for most general-audience writing. Below 50 is difficult. Most newspaper copy sits around 60-65.
Flesch-Kincaid Grade Level converts the same inputs into a US school grade number. Grade 8 means an average eighth-grader can read it. Consumer product copy typically targets grades 7-9. Legal and academic writing commonly scores grade 14+.
Gunning Fog Index adds a specific penalty for “complex words” — three or more syllables excluding common suffixes. It tends to score technical writing higher than the Flesch formula, making it useful for catching jargon-heavy passages that the simpler formula underweights.
SMOG (Simple Measure of Gobbledygook) uses only the count of polysyllabic words per sample. It is considered reliable for health communication materials, where readability standards are often cited as a patient safety concern.
Using all four together rather than just one gives a more stable picture: if three of four formulas flag the same paragraphs, those are the ones to rewrite.
Why LLM output often reads too complex
Models trained on academic papers, technical documentation, and formal writing inherit those registers. Without an explicit instruction about reading level, a model default-selects vocabulary and sentence structures that appear in its training data — which skews formal. Common patterns that inflate readability scores include long sentences with multiple subordinate clauses, nominalisations (“consideration” instead of “think about”, “utilisation” instead of “use”), passive constructions, and technical terms that could be replaced with common words. The Gunning Fog index is particularly good at catching nominalisations because they tend to be polysyllabic.
Tips and example
For consumer-facing copy, aim for a Flesch Reading Ease of 60+ and a grade level around 7-9. If your model output lands at grade 13, the fastest fix is to add a constraint to the prompt — “write at a 7th-grade reading level, short sentences, plain words” — and re-score. Watch the flagged sentences: a single 40-word sentence stuffed with jargon can pull the whole document up two grades, so breaking it into two or three plain sentences often does most of the work. The syllable estimate is heuristic but the indices are robust across a paragraph.