Sentence length variance analyser

By Jack Stovell · published 2026-09-24

Sentence length variance (sentence_length_variance) measures how much your sentence lengths swing between short and long; higher means more variation. When ScriptGrain scores a draft, code measures the variance of its sentence lengths in words squared; across 299 published pieces measured that way, the median is 102. A profile stores the extraction model's own reading.

Analyse your sentence length variance free

Your free voice profile measures sentence length variance among all 45 attributes, from a sample of your own writing. No card needed.

Try the six-signal check first

At a glance

What sentence length variance measures

Sentence length variance measures the spread of your sentence lengths, not their average. For example, two writers can both average 15 words a sentence. One writes every sentence at 15 words. The other swings between 5-word fragments and 25-word build-ups that land on a punch. Same average, wildly different feel. That difference is variance.

In ScriptGrain's system it sits in the Syntax group alongside things like average sentence length and sentence complexity; you can see the whole family on the syntax group page. The app calls the scale unitless: the number goes up as your sentences vary more and down as they flatten out. When a draft is scored, code takes the population variance (the mean squared deviation from the average sentence length), so that number is in words squared. The extraction prompt asks for a bare number with no unit, so a profile's value is the model's reading and need not sit on the same scale.

Related but separate is burstiness, which ScriptGrain measures as two numbers: variation, the standard deviation divided by the mean (a normalised version of variance), and short sentence share, the percentage of sentences under 8 words. Both concepts have their own glossary entries: sentence length variance and burstiness.

How sentence length variance is scored

Sentence length variance moves your voice match score at a weight of 1.5. When a draft is checked against your profile, code measures the draft's variance, takes the gap between it and your profile's stored value, then converts that gap to a similarity score: 1 minus the gap divided by a tolerance, floored at zero so it never goes negative. The tolerance is the larger of 25 or 1.2 times your profile's own value.

It is one of 28 attributes (out of 45) that can move the score. In the full method, code counts sentence, punctuation, pronoun, article and word-length features, and Claude Haiku judges formality, humour, structure and rhythm; the voice measurement framework sets it out, and the voice match glossary entry has the short version. The app calls 90%+ Excellent and under 60% Low.

How to analyse sentence length variance in your own writing

Sentence length variance can be measured by hand with nothing more than a word count and a calculator. Here's how:

  1. Take a sample of your writing, at least a few paragraphs, and split it into individual sentences.
  2. Count the words in each sentence.
  3. Add up all the word counts and divide by the number of sentences. That's your mean.
  4. For each sentence, subtract the mean from its word count, then square the result.
  5. Add up all those squared differences and divide by the number of sentences. That's your population variance.
  6. Optionally, take the square root of that number to get the standard deviation, which is easier to read against your mean.

Try it on a paragraph you're proud of, then one you dashed off in a hurry. ScriptGrain's free writing style analysis shows the standard deviation from step 6 under the label "Sentence length variance", and its AI cliché checker flags paragraphs where every sentence sits between 20 and 40 words. ScriptGrain's free voice profile reads the attribute from a writing sample among all 45 attributes.

Typical sentence length variance in published writing

Typical sentence length variance in published writing has a median of 102 words squared. That comes from ScriptGrain's reference corpus: 299 recent public pieces from 47 sources (blogs, newsletters and essay sites), built on 20 September 2026, each measured with the product's own code and discarded. The percentiles: p10 is 47, p25 is 69, the median 102, p75 146, and p90 260. These are code measurements of the variance of sentence lengths; a profile's stored value is the extraction model's reading, so don't read one against these numbers.

The related burstiness measure (standard deviation divided by mean) runs from 0.5 at p10 to 0.78 at p90, median 0.61. Short sentence share, the percentage under 8 words, runs from 8% at p10 to 40% at p90, with a median of 20%. In the median piece, one sentence in five is under 8 words.

How to make AI vary its sentence length

AI can be made to vary its sentence length by giving it explicit rules about short sentences, not just a vague instruction to "sound more natural". Try these:

  1. Require at least one short sentence, under 8 words, in every paragraph of three or more sentences.
  2. Ask for a short sentence immediately after a long one, right where the point lands.
  3. Reject any paragraph where every sentence sits in the same length band, say all between 20 and 40 words.

When ScriptGrain writes from a profile, this attribute goes into the system prompt as a hard constraint, with a named rule to match sentence rhythm, length variance and complexity exactly. A burstiness block sits alongside it on every draft: at least one sentence under 8 words in every paragraph of three or more sentences, and adjacent sentences differing in length by 3x at least once. The profile reaches ChatGPT, Claude and other tools through the API and the MCP server.

How to make AI keep an even sentence rhythm

AI can be told to keep sentences more even when the context calls for it, such as reference documentation or step-by-step instructions. The moves:

  1. Ask for sentences of similar length in reference and instructional copy specifically.
  2. Merge short fragments into the sentence beside them instead of leaving them standalone.
  3. Keep one idea per sentence; length tends to steady out naturally when you stop cramming.

A ScriptGrain profile sends a low stored variance as a hard constraint, but its burstiness block still asks for a sentence under 8 words in every paragraph of three or more sentences, so ScriptGrain never asks its own drafts for a fully even rhythm. The profile reaches ChatGPT, Claude and other tools through the API and MCP server.

Questions

What is burstiness in writing?

In ScriptGrain's measure, burstiness is a normalised way of describing sentence length variation: standard deviation divided by mean. In the reference corpus it runs from 0.5 at p10 to 0.78 at p90, median 0.61. High burstiness means you swing hard between short punchy sentences and long build-ups; low burstiness means your sentences cluster near the same length. It's related to variance but scaled by your average, so it's easier to compare across writers with different average sentence lengths.

How is sentence length variance scored?

It's scored as part of voice match, with a weight of 1.5. Code measures your draft's variance in words squared and takes the gap to your profile's stored value; similarity is 1 minus that gap divided by a tolerance (the larger of 25 or 1.2 times your profile's value), floored at 0. It's one of 28 attributes, out of 45 total, that can move your score.

How do I score writing for sentence variety?

Calculate the mean sentence length in words, then the population variance (average of the squared differences from the mean). Optionally take the square root for standard deviation, and divide by the mean to get a burstiness ratio. ScriptGrain's free analyser shows the standard deviation under the label "Sentence length variance" and reads "High relative to your average" when it is more than half the mean.

How do I make ChatGPT writing less monotonous?

Give it explicit rules rather than vague style requests. Require a short sentence, under 8 words, in every paragraph of three or more. Ask for a short sentence right after a long one. Reject any paragraph where every sentence lands in the same length band. ScriptGrain's generation prompt carries rules like these on every draft it writes from a profile, including drafts requested through the API or MCP server.

What's a sentence length variance analyser?

It's a tool that measures how much your sentence lengths swing, usually as the variance or standard deviation across a sample of your writing. ScriptGrain reads it three ways: its free analyser shows the standard deviation, draft scoring uses the variance in words squared, and the free voice profile stores the extraction model's reading among all 45 attributes. Read any of them beside the average sentence length, since the same spread means more around a short average.

Related

Sources