Prompt reading level checker
Clear prompts get clearer answers. If your instructions are buried in long, clause-heavy sentences and rare vocabulary, a model is more likely to miss part of what you asked. This tool scores your prompt with the same readability formulas used for plain-language writing — Flesch-Kincaid, Gunning Fog, and Flesch Reading Ease — so you can spot and fix density before it costs you a bad response. Everything runs locally in your browser.
How it works
You paste your prompt text. The tool splits it into sentences and words, counts syllables with a vowel-group heuristic, and identifies complex (three-plus syllable) words. From those counts it computes the Flesch-Kincaid grade level, the Gunning Fog index, and the Flesch Reading Ease score, and reports your average words per sentence and complex-word count. If the estimated grade level rises above 12, it flags the prompt as likely too dense.
Understanding the three scores
Each formula captures a slightly different aspect of reading difficulty:
Flesch-Kincaid Grade Level maps your text to an approximate US school grade, so a score of 10 means the instruction reads at a 10th-grade level. For prompts, aim for grade 8–12. Below that is fine; above 12 suggests unnecessarily complex sentence structure.
Gunning Fog Index emphasises the role of complex words — those with three or more syllables. It tends to be more sensitive than Flesch-Kincaid to technical jargon, which makes it particularly useful for prompts that mix technical vocabulary with instructional language.
Flesch Reading Ease is scored 0–100, where higher means easier. A score of 60–70 corresponds roughly to plain English business writing. Below 50 indicates dense or technical prose; below 30 is academic or legal in complexity.
The three scores together give a richer picture than any one alone. A prompt can score deceptively well on Flesch-Kincaid (short sentences) but poorly on Gunning Fog if it is packed with three-syllable technical terms.
Why readability matters for LLMs specifically
LLMs are trained on vast corpora that include both dense and simple text, so they can technically parse complex instructions. The problem is not parsing — it is instruction following under ambiguity. When a sentence is long and clause-heavy, a model has more opportunity to weight one part of the instruction over another, to miss a conditional, or to anchor on an earlier phrase at the expense of a later constraint.
In practice, prompts with long sentences often produce outputs that are correct on the general request but miss a specific constraint that was buried mid-sentence. Splitting the constraint into its own short sentence, or moving it to a bullet point, is usually the fastest fix.
The two high-leverage edits
Sentence splitting is the single most effective move. A sentence of 35 words can almost always be split into two without changing meaning, and the grade level drops by several points immediately.
Vocabulary substitution is the second lever. Technical terms that are genuinely necessary are fine — do not replace “API endpoint” with something less precise. But when you use a long word by habit rather than necessity — “utilise” instead of “use,” “facilitate” instead of “help,” “demonstrate” instead of “show” — the substitution costs nothing and drops the Gunning Fog score.
Tips and notes
Aim for roughly grade 8 to 12 for instructions — readable without being simplistic. The syllable counter is a heuristic, so treat results as accurate to about a grade level rather than exact. Because nothing leaves your browser, you can check sensitive or proprietary prompts here freely, then paste the simplified version into your model and compare the quality of the responses.