Writing cadence analyser

By Jack Stovell · published 2026-09-24

Cadence is the rhythm underneath your words, from punchy or flowing sentences to where the strongest line in a paragraph lands. ScriptGrain reads it through its Quirks and cadence group, 5 of the 45 attributes in a voice profile. The extraction model judges four and catalogues one, and only rhythm pattern moves the voice match score.

Analyse your quirks and cadence attributes free

Your free voice profile measures every quirks and cadence attribute among all 45, from a sample of your own writing. No card needed.

Try the six-signal check first

At a glance

What the quirks and cadence group measures

Quirks and cadence covers the personal habits that make your writing recognisably yours: recurring spellings, punchy or flowing rhythm, paragraph length, where the strongest sentence lands and repeated structural shapes. It is one of the eight groups that make up the 45 voice attributes.

Consistent misspellings catches the idiosyncratic spellings you repeat across samples, too regular to be typos. The extraction model lists up to three per analysis, quoted verbatim. It has no voice match comparator, so it never touches your score, but it stays in your profile as a fingerprint detail.

Rhythm pattern is the big one in this group. The extraction model labels your overall cadence punchy, flowing or mixed, and it's the only attribute here that moves your voice match score (weight 1.5). Short fragments and hard stops read punchy. Longer, connected clauses read flowing. Writing that switches between the two is what "mixed" is for.

Paragraph length records whether you typically write short, medium, long, or varied paragraphs. It's judged, not counted: the model reads the pattern across your samples rather than averaging word counts per paragraph, and the label does not move the voice match score.

Power sentence position tracks where your strongest line tends to sit: start, end, both or varied. A writer who saves the point for the last line reads differently from one who opens with the punch and explains after. Neither is wrong. The model judges it, and it does not move the score.

Structural signatures is the catch-all: a short description, 15 words at most, that the extraction model writes about the recurring shapes in how you build a piece, such as a fragment after every long sentence or numbered steps before the payoff. It does not move the score either.

How quirks and cadence features are scored

Only rhythm pattern moves the voice match score. When a draft is scored, Claude Haiku judges its rhythm label and code compares that with your profile's at weight 1.5: an exact match scores 1, "mixed" on either side 0.7 and a straight mismatch 0.15. None of the five is counted in code; the split is 4 judged and 1 catalogued. You can't tally "punchy" the way you tally commas.

The extraction model (Claude Sonnet 5) reads your samples and judges rhythm pattern, paragraph length, power sentence position and structural signatures directly, the same way it judges formality or humour register elsewhere in a profile. Consistent misspellings gets a different treatment again: it's a word list, catalogued rather than judged, capped at three entries. Judged labels vary slightly from run to run: ScriptGrain's SGR-003 study measured label agreement at 72% to 87%.

The other four are not decorative: ScriptGrain sends every attribute to the model when it writes a draft. A score simply cannot reward them and polish cannot target them; across all 45 attributes, 28 can move the score. Together the five are part of your stylometric fingerprint, the set of habits that mark writing as yours.

Voice match itself works on a wider mechanism than this group alone: code counts sentence, punctuation, pronoun, article and word-length features, Claude Haiku judges formality, humour, structure and rhythm, and a weighted comparison runs through a fixed calibration to produce a score from 0 to 1. The app labels that score: 90%+ Excellent, 75 to 89% Good, 60 to 74% Fair, under 60% Low. You can read the full mechanics in the voice measurement framework, and see where cadence sits among all 45 attributes in the reference list.

How to analyse the quirks and cadence of your own writing

You can analyse your own cadence by hand in five steps, using nothing more than a highlighter and a bit of patience.

  1. Pull three to five samples of your own writing, at least a few hundred words each, ideally from different pieces rather than one long document.
  2. Mark every sentence under eight words. If you mark a lot of them, you lean punchy; if they're rare, you likely lean flowing.
  3. Look at paragraph breaks. Are they short and frequent, or do you let paragraphs run long before you break? Note whether it's consistent or varied piece to piece.
  4. Find the strongest sentence in each paragraph, the one that would survive if you cut everything else, and note where it sits: opening, closing, or scattered.
  5. List any spelling habits that repeat across samples, and any structural tic you use more than once (numbered steps, a direct question to the reader, a fragment used as a bridge). That's your version of a structural signature.

Uneven sentence length has a technical name, burstiness, and step 2 is a rough check of it.

ScriptGrain's free voice profile reads all five of these attributes from a writing sample, alongside the other 40.

How to make AI writing match your quirks and cadence

You make AI writing match your cadence by giving the model an explicit, structured description of your rhythm rather than trusting it to infer one from a few pasted paragraphs.

  1. Gather a real sample of your own writing, at least a few hundred words, ideally over 3,000 across multiple pieces if you want the fuller picture.
  2. Feed it to a system that extracts cadence explicitly: rhythm pattern, paragraph length, power sentence position, structural signatures, consistent misspellings, rather than one that just "reads the vibe".
  3. Check the output states rhythm as punchy, flowing, or mixed, states a paragraph length preference, and names where the strongest line tends to land. Vague summaries ("engaging and dynamic") aren't cadence data.
  4. Give that structured profile to your AI tool as the instruction set, not just as an example to imitate.
  5. Regenerate and compare rhythm specifically: is the AI output still producing short fragments where you'd naturally cut hard, or has it smoothed everything into uniform, medium-length sentences? That smoothing is the tell that cadence got lost.

When ScriptGrain writes, all five travel in the profile block of its system prompt under "Treat every attribute as a hard constraint, not a suggestion", and rhythm pattern decides where sentence fragments are allowed. The profile also reaches ChatGPT, Claude and other tools through the API and MCP server: `GET /v1/profiles/{id}` and the `get_profile` MCP tool return all 45 attributes as JSON on every plan.

Attributes in this group

Questions

What is a writing cadence analyser?

A writing cadence analyser reads your rhythm, paragraph habits and structural patterns and turns them into explicit values rather than a vague impression. In a ScriptGrain voice profile that means five attributes: rhythm pattern, paragraph length, power sentence position, structural signatures and consistent misspellings, all read by the extraction model (Claude Sonnet 5) from your writing samples and stored alongside 40 other attributes.

What counts as cadence in writing?

Cadence is the rhythm of your sentences and paragraphs: whether they're punchy or flowing, how long your paragraphs typically run, and where your strongest line tends to land. It's separate from vocabulary or tone. Two writers using the same words can read completely differently depending on cadence, and ScriptGrain gives it a group of its own rather than folding it into formality or word choice.

What are writing quirks and why do they matter?

Writing quirks are the small, repeatable habits that make a piece recognisably yours: consistent misspellings, recurring structural shapes, a habit of ending paragraphs on a short line. They matter because they are specific to you, so a draft that drops them sounds like someone else. ScriptGrain's extraction model lists up to three recurring spellings per analysis and describes your structural signatures in 15 words or fewer.

How do I analyse the rhythm and cadence of my writing?

Mark sentences under eight words, check whether your paragraphs run short or long, and find where the strongest sentence in each paragraph sits. Do this across several samples and patterns emerge fast. That's a manual version of the judgement ScriptGrain's extraction model makes when it reads rhythm pattern, paragraph length, power sentence position and structural signatures from your samples as part of a 45-attribute profile.

How do I make AI keep my writing quirks when it generates text?

Give the AI structured cadence data, not just sample text to imitate. A profile that states your rhythm as punchy, flowing or mixed, names your typical paragraph length and locates where your power sentences land gives the model rules to follow instead of a style to guess at. ScriptGrain stores that data across 45 attributes and makes it reachable by ChatGPT, Claude and other tools through the API and MCP server.

Related

Sources