Burstiness: the sentence rhythm that gives writers away
Burstiness is the variation in sentence length and structure across a passage. Human writers write in bursts: a long, winding sentence, then a short one. Then another. Language models tend to produce sentences of similar length, one after the other, at the same medium pace. Humanize 360 measures this directly as sentence-length variance, the first of the eleven signals in the free AI detector, and it is one of the strongest indicators in the set.
Here is the idea in one paragraph. "The market opens at six. By seven the fish is gone, most of it bought by the same four restaurant owners who have been coming since before the road was paved, and the rest of the morning belongs to the tomato sellers." Five words, then thirty-six. A model writing the same scene would usually hand you two sentences of about eighteen words each. Both versions are grammatical. Only one has a pulse.
How the signal is calculated
The detector splits your text into sentences, counts the words in each and looks at the spread around the average. A passage where every sentence lands between fourteen and twenty words has low variance. A passage with a four-word sentence beside a forty-word one has high variance. We also look at sentence shape: whether each one starts with subject then verb, or whether some open with a clause, a question or a fragment.
Why an even rhythm reads as machine
A model chooses each word by likelihood, and the likeliest place to end a sentence is roughly where the last one ended. The result is a steady, tidy cadence. Human rhythm follows meaning instead. You use a short sentence to land a point and a long one to walk through a reason. You break a rule for effect. None of that is random, but none of it is even, either.
What the humanizer does about it
- Standard and Flowing modes merge some sentences and split others so the lengths stop clustering around one number.
- Academic mode keeps the formal register and still varies length; formal does not have to mean uniform.
- A custom style profile records your own average sentence length and variance, so the engine aims for your rhythm, not a generic one.
- Selective rephrase on Standard plans and above lets you rewrite one flat sentence without touching its neighbours.
Where people get burstiness wrong
- It is not about chopping everything short. Ten short sentences in a row are as uniform as ten long ones.
- It is not perplexity. Burstiness is about sentence rhythm; perplexity is about word choice. Detectors report them together because they usually fail together.
- Bullet lists distort it, because every item is a "sentence" of similar length. Check the prose parts of a document separately.
- Very short passages, under about a hundred words, do not contain enough sentences for the measurement to mean much.
Common questions
Is burstiness the same as varied sentence length?
Mostly, yes. The formal version also considers sentence shape and where the variation occurs, but if you remember it as "mix short and long sentences on purpose", you have the practical part.
Can I raise burstiness by hand?
Yes. Read the passage aloud and mark where you would naturally pause or speed up. Cut one sentence to five words. Let another run. The detector breakdown shows whether the variance moved.
Why does my formal report score low on this signal?
Formal writing often falls into a steady eighteen-to-twenty-two word cadence. That is a habit, not a requirement. Academic mode keeps the register and breaks the cadence.
Does Humanize 360 send my text to another AI to measure this?
No. The analyser is our own code on our own servers. Anonymous checks are processed in memory and discarded when the result is returned.