Claim density in writing

By Jack Stovell · published 2026-09-24

Claim density is the average number of claims a writer makes per paragraph: the points stated as true, not the sentences around them. A ScriptGrain voice profile stores it as claim_density, the extraction model's estimate from your samples rather than a code count, and it never moves the voice match score.

Analyse your claim density free

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

Try the six-signal check first

At a glance

What claim density measures

Claim density measures how many actual assertions show up in a paragraph, on average, rather than how many words fill it. A "claim" here means a stated point, a position, a fact offered as true; not scene-setting, not throat-clearing, not a sentence that just gestures at the topic before getting to the point. So a paragraph with four sentences might carry one claim (heavily padded) or four (tightly packed). ScriptGrain's own definition puts it plainly: "the average number of claims made in a paragraph." That's the whole idea. High density means the writing keeps landing points. Low density means it's circling them.

This attribute sits in the content patterns group alongside three others. It pairs naturally with specificity level, which judges how concrete the writing is, and anecdote usage, which estimates how often a story carries the point.

How claim density is scored

Claim density is not scored: a profile stores a value for it, but it never moves your voice match score. The stored value is the extraction model's estimate, and no code counts claims. The voice-match engine has no comparator built for this attribute, so however dense or sparse your paragraphs are, polish can't target it and a draft won't be marked up or down for it. Twenty-eight of ScriptGrain's 45 voice attributes can shift a voice match score. This isn't one of them. If you want the full mechanics of how scoring works elsewhere, the voice measurement framework and the voice match glossary entry cover it properly.

How to analyse claim density in your own writing

You can do this by hand in a few minutes, no software required.

  1. Pick five paragraphs from a recent piece of writing, ideally ones you'd call representative.
  2. Read each paragraph and underline every sentence that makes an actual claim (a stated point, not a setup or a transition).
  3. Count the underlines per paragraph.
  4. Average the counts across your five paragraphs. That's your rough claim density.
  5. Reread any paragraph that scored below one. Ask what it's actually doing if it isn't making a point.

That fifth step is the useful one. It's where you find the paragraphs that exist purely to ease you into the next paragraph, which is a habit worth breaking.

ScriptGrain's free voice profile reads claim density from a writing sample, alongside the other 44 attributes.

Examples of claim density in real writing

Example: "Remote teams ship faster, and they document more." Two claims, one sentence, no padding.

Example: "The market shifted in March; prices haven't recovered since." Two claims again, joined by a semicolon rather than spread across a paragraph.

Example: "It was a Tuesday, and the weather had been unremarkable for weeks, and everyone had more or less settled into a routine by then." One long sentence of scene-setting, and not yet a claim the reader can use. That's low density. Not wrong, necessarily; sometimes you want the slow build. But if every paragraph reads like that, you've got a fluff problem.

How to make AI writing say more per paragraph

This is the fix for AI writing that reads full but says little.

  1. Ask each paragraph to make a point instead of describing the topic. There's a real difference between "productivity apps are popular" and "this app cuts admin time by half"; only one of those is a claim you can act on.
  2. Cut setup sentences that make no claim. If a sentence exists purely to lead into the next one, it's dead weight.
  3. Put the claim in the paragraph's first sentence. Everything after it should support that claim, not delay it.

When a ScriptGrain profile writes for you, claim density travels inside the voice profile JSON block as a hard constraint, same as every other attribute; there's no separate rule bolted on for it. The profile reaches ChatGPT, Claude and other tools through the API and MCP server, so they can read the same value.

How to make AI make one point at a time

This is the fix for AI writing that tries to argue three things at once and ends up arguing none of them well.

  1. Limit each paragraph to one claim, plus its support. Not two claims loosely related. One.
  2. Split paragraphs that argue two things. If you catch yourself using "also" or "additionally" to bolt on a second point, that's your signal to break the paragraph.
  3. Add the evidence or example before moving to the next claim. Support the point you just made, then move on; don't stack unsupported claims on top of each other.

Fair enough if that sounds mechanical. It is, a bit. A ScriptGrain profile with a low claim density sends that value the same way, as a hard constraint in every draft it writes. Its sibling analogy preference records whether analogies do the explaining.

Questions

How is claim density measured?

It's expressed as claims per paragraph, stored as a number. There's no code counting sentences against a rulebook; a model (Claude Sonnet 5, in ScriptGrain's case) reads each sample and judges how many claims a typical paragraph makes. Two passes happen: one per sample, one to synthesise the final reading. It's an estimate, not a literal tally.

How to make writing more persuasive?

Raise your claim density and back each claim with something concrete. Persuasive writing tends to state a position early, then support it, rather than easing in with throat-clearing. Pair that with tighter specificity, so claims land as facts rather than vague gestures, and a well-placed anecdote can carry a claim further than another paragraph of explanation ever will.

How to make AI writing more substantive?

Ask for one claim per paragraph, stated first, supported after. Cut any sentence that doesn't assert something. AI models left unconstrained tend to pad with scene-setting and soft transitions; naming the pattern (claim density, specifically) and asking the tool to raise it per paragraph is a direct, testable fix rather than a vague request for "better" writing.

How to make AI writing less fluffy?

Fluff is usually just low claim density wearing a lot of words. Strip setup sentences, put the point first, and cut anything that describes the topic instead of making an assertion about it. The steps above (point first, cut setup, one claim per paragraph) work as a fluff test as much as a density test; they're the same problem from two angles.

How to make ChatGPT get to the point?

Tell it explicitly: one claim per paragraph, claim in the first sentence, no setup sentences. ChatGPT often drifts toward hedged, padded prose, so plain instructions beat vague ones. A ScriptGrain voice profile carries your claim density into every draft it writes as a standing constraint rather than a one-off prompt, which matters if you're generating more than a single draft and don't want to repeat yourself every time.

Related

Sources