What burstiness means
Here, burstiness is the standard deviation of sentence length divided by the average. People tend to mix short and long sentences. Drafts that come straight from a model often keep every sentence at about the same length. As a rough guide, below 0.3 reads flat, 0.3 to 0.5 has some variation, and above 0.5 reads varied. Those cut-offs are ours, chosen for editing, not taken from any detector.
What perplexity means
Perplexity describes how predictable a sequence of words is to a language model. Familiar phrases and common word combinations are easier to predict and score lower. Computing it needs a language model, which would mean sending your text somewhere or downloading a large model, so this page measures rhythm only.
Why detectors use them, and why that fails
Some AI detectors combine the two: low perplexity (predictable wording) plus low burstiness (even rhythm) is treated as a sign of machine text. Both signals are weak on their own. A careful human writer can produce predictable, even prose. An edited AI draft can vary a lot. Topic, genre, translation, quotations, technical terms and the writer's language level all move both numbers. That's why neither can prove who wrote a text, and why a rhythm score should guide editing, not judge authorship.
How to vary rhythm without forcing it
Look for a highlighted run. Merge two short sentences that share a subject, or break up a long one where the idea turns. Put a hard point in its own short sentence. Don't insert rare words or random fragments to move a number; read the paragraph aloud instead. Rhythm should follow meaning.
A better review question
Instead of asking whether the score looks human, ask whether the reader can follow the argument, check the evidence, and hear the writer's judgment.
Questions
What does burstiness mean in writing?
How much sentence length and shape vary across a text. Writing that mixes short and long sentences has high burstiness; writing where every sentence is about the same length has low burstiness.
What is perplexity?
A measure of how predictable a text is to a language model. Common phrases and familiar word sequences have low perplexity. It needs a language model to compute, so this page doesn't calculate it.
Do perplexity and burstiness prove a text was written by AI?
No. A careful human writer can produce predictable, even prose, and an edited AI draft can vary a lot. Topic, genre and language level all move both numbers.
How is the burstiness score calculated here?
We split the text into sentences, count the words in each, and divide the standard deviation of those lengths by the average (the coefficient of variation). The rhythm score is the same one the Naturable editor uses.
Related: Signs of AI writing · Em dash checker · Full editor