Free text entropy / predictability calculator. Computes Shannon entropy at the character and word level, n-gram entropy, and a compression-ratio estimate. Compare your text against typical human and AI-generated ranges.
Built and maintained by Paul Clark, Redmoon Software
Use when you suspect a text is repetitive but can't spot why. Word entropy catches phrase-level repetition; character entropy catches alphabet-level oddities.
Lexical diversity measures unique-word *count*; entropy measures unique-word *distribution*. Two texts can have identical TTR but very different entropy.
Shannon entropy is computed as H = −Σ p(x) · log₂ p(x) over the chosen symbol set (characters, words, or word bigrams).
Predictability is 1 − H/Hₘₐₓ — higher means more repetitive and easier to compress.
Compare modes: AI-generated text often shows lower word entropy and higher word-bigram repetition than human writing of similar length.
A measurement of unpredictability. Higher entropy means each symbol (word or character) carries more information; lower means the text is more repetitive.
~9–10 bits per word for typical prose; ~4 bits per character. Highly templated text falls below.
Different views surface different patterns. Word-bigram entropy catches phrase-level repetition; character entropy catches alphabet quirks.
Input
The cat sat on the mat. The cat sat on the mat. The cat sat on the mat.
Output
Word entropy: 2.32 bits — Very predictable (predictability 51%).
Only 6 unique words across 18 — the same six repeated three times. Maximum possible entropy is log₂(6) ≈ 2.58 bits, so 2.32 is close to maximum-given-vocab but predictability vs uniform is low.
Entropy measures how unpredictable the next unit of text is, in bits. Low entropy means highly predictable — heavy repetition, a small vocabulary, formulaic structure. High entropy means varied and less predictable.
It is a genuinely different lens from every other measure here, because it makes no assumptions about English at all. It does not know what a syllable is, or a stopword, or a sentence. It just measures distribution, which makes it robust where the linguistic heuristics are weakest.
Very low entropy usually means the text is repetitive or templated. Very high entropy in ordinary prose often means something is wrong in the other direction — inconsistent terminology, elegant variation, or text assembled from mismatched sources.
The honest use is comparative: run it on a document and on something known-good from the same genre. An entropy figure with nothing to compare it against is close to meaningless, because the natural range differs enormously between a legal contract and a short story.
% content words vs. function words.
Text analysisVocabulary variety — type‑token ratio.
Text analysisTotal + per‑word syllable counts.
Text analysisHighlight passive sentences in your writing.