Writing rhythm analyser

By Jack Stovell · published 2026-09-24

Rhythm pattern (rhythm_pattern) is the overall cadence of your writing, labelled punchy, flowing or mixed. In a ScriptGrain voice profile it is a label the extraction model judges from your samples, not something a script counts. It sits in the Quirks and cadence group and moves the voice match score at weight 1.5.

Analyse your rhythm pattern free

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

Try the six-signal check first

At a glance

What rhythm pattern measures

Rhythm pattern measures the overall cadence of your writing: punchy, flowing, or mixed. That's the app's own definition, and it's blunt. Punchy means short sentences, fragments and hard stops. Flowing means longer sentences joined by connectives, the kind of prose that carries a reader through several clauses before it lets them breathe. Mixed means both. An average sentence length can hide the difference: in ScriptGrain's SGR-001 study, both arms of an opinion brief averaged about 15 words a sentence, yet the share of short sentences went from 16% to 36%. Rhythm pattern is one of five attributes in the Quirks and cadence group.

No code counts sentence lengths and assigns this label by threshold. The extraction model, Claude Sonnet 5, reads your samples in two passes: pass one goes sample by sample, pass two synthesises a single profile. The label comes out of that judgement and is snapped to punchy, flowing or mixed when the profile is saved. Think of an editor who reads a paragraph and says "yeah, that's punchy" without running a word count first.

The closest counted signals are burstiness variation (the standard deviation of sentence length divided by the mean) and the share of sentences under 8 words. They are proxies: code can count them, but neither sets the label. Burstiness covers the first in more depth.

How rhythm pattern is scored

Rhythm pattern moves your voice match score at a weight of 1.5, one of 28 of the 45 attributes that count towards it at all. When a draft is scored, Claude Haiku reads it and judges its rhythm label. Code then compares that label with your profile's. An exact match scores 1. "Mixed" or "varied" on either side scores 0.7. A profile value that contains the draft's label scores 0.9. Anything else scores 0.15.

So this one is a category to match, with no number to hit. Write punchy when your profile says punchy. Fair enough, that sounds obvious, but it's easy to drift into flowing sentences without noticing. The full scoring mechanics, including how the other 27 attributes weigh in, live on the voice measurement framework page.

How to analyse rhythm pattern in your own writing

You can do a rough version of this yourself, no software required. Here's how:

  1. Pull a sample of your writing, at least a few paragraphs.
  2. Count the words in every sentence and jot them down in order.
  3. Work out the average sentence length and then the standard deviation.
  4. Divide standard deviation by the average. That's your burstiness variation.
  5. Count what percentage of your sentences fall under 8 words.
  6. Read the sample aloud. Does it feel like short jabs or one long unbroken pull?

That last step matters more than the maths. Numbers give you a proxy; your ear gives you the actual judgement, which is closer to what the extraction model is doing anyway. ScriptGrain's free voice profile reads rhythm pattern from a writing sample as one of all 45 attributes.

Typical rhythm pattern in published writing

ScriptGrain's reference corpus has no figures for the rhythm label itself, only for its two counted proxies. The corpus is 299 public pieces from 47 sources (835,655 words, built 2026-09-20). Sentence variation, the standard deviation of sentence length divided by the mean, runs p10 0.5, p25 0.54, median 0.61, p75 0.695, p90 0.78. Short sentence share, the percentage of sentences under 8 words, runs p10 8, p25 13, median 20, p75 30, p90 40.

Translate that: at the median, one sentence in five runs under 8 words, and the standard deviation of sentence length is 0.61 of the mean. The corpus was measured once and discarded, no piece stored, just the numbers. One caution on the label itself: in ScriptGrain's SGR-003 study, the 13 enumerated labels read from fewer pieces matched the 15-piece profile 72% to 87% of the time depending on piece count (82% at one piece, 87% at eight). Labels, rhythm pattern included, are the least stable part of a profile. Treat it as a strong signal, not gospel.

How to make AI writing punchier

If you're prompting ChatGPT or another assistant and the output reads too smooth, here's what to change:

  1. Allow fragments and short sentences, especially near the ends of paragraphs.
  2. Cut filler words and throat-clearing before the main clause arrives.
  3. End paragraphs on their shortest sentence, not their longest.

When ScriptGrain writes, rhythm pattern travels in the profile block of the system prompt under "Treat every attribute as a hard constraint, not a suggestion", and fragments are allowed where the profile's rhythm pattern permits. A burstiness block also asks for at least one sentence under 8 words in every paragraph of three or more sentences. The profile reaches ChatGPT, Claude and other tools through the API and MCP server.

How to make AI writing flow better

If the output reads choppy and you want it to carry further, do the opposite:

  1. Link sentences with "and", "so" and "which" instead of full stops.
  2. Ask for longer sentences that carry a qualification inside them.
  3. Keep a few short sentences in reserve, purely for contrast.

Keep step 3: even flowing writing needs the odd short line, and paragraph length and power sentence position cover where it lands. When ScriptGrain writes, a flowing label goes in as a hard constraint like any other value, and the burstiness block still asks for one sentence under 8 words in every paragraph of three or more sentences. The profile reaches ChatGPT, Claude and other tools through the API and MCP server.

Questions

What is a writing rhythm analyser?

It's a way of judging whether your prose reads punchy, flowing or mixed. Some tools count proxies such as sentence length variation and the share of short sentences. In a ScriptGrain voice profile, the extraction model judges the rhythm pattern label from your samples; no word-count script sets it, and the proxies are only a way to check the reading by hand.

Punchy vs flowing writing, what's the difference?

Punchy writing uses short sentences and fragments, often landing hard at the end of a paragraph. Flowing writing links longer clauses with connectives like "and" or "which", carrying the reader through without a stop. ScriptGrain's reference corpus has no counts for the labels, but in its median piece one sentence in five runs under 8 words.

How do I improve rhythm in my writing?

Vary your sentence lengths sharply rather than evenly. Put a short sentence next to a long one, especially at paragraph ends. Read your draft aloud and listen for whether it jabs or glides. Check what percentage of your sentences run under 8 words; the median piece in ScriptGrain's 299-piece reference corpus has 20%, a rough benchmark rather than a target.

How is writing rhythm scored?

Rhythm pattern feeds voice match at weight 1.5. Claude Haiku judges your draft's rhythm label, then code compares it against your profile's stored label: an exact match scores 1, "mixed" on either side 0.7, a profile value that contains the draft's label 0.9 and anything else 0.15. The stored label itself is the extraction model's judgement, not a count.

How do I make ChatGPT writing flow better?

Ask it explicitly to link sentences with "and", "so" or "which" rather than stopping at every clause. Request longer sentences that carry a qualification inside them. Keep a handful of short sentences for contrast, since even flowing prose needs a break. Through the API or MCP server, ChatGPT can also read your ScriptGrain profile, rhythm pattern included, as JSON.

Related

Sources