Argument structure analyser

By Jack Stovell · published 2026-09-24

Argument structure is whether your writing leads with the conclusion, leads with the evidence, or unfolds as a narrative. ScriptGrain reads it as one of three labels: evidence-first, conclusion-first or narrative. A model judges it rather than code counting it, and it weighs 1.5 in the voice match score, the heaviest in the Rhetoric group.

Analyse your argument structure free

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

Try the six-signal check first

At a glance

What argument structure measures

Argument structure measures the order you put things in, not the quality of the argument itself. The app's own definition: "Whether arguments lead with evidence, lead with the conclusion, or unfold as a narrative." So it's a sequencing question. Do you say the answer first and defend it after? Do you build the case, brick by brick, and land on the conclusion at the end? Or do you tell it as a story, letting the point emerge from events rather than stating it upfront?

Here's the thing: different kinds of writing pull different ways. Academic papers often build evidence-first. News and sales copy tend to put the conclusion up top, because a reader may stop after the first lines. Novelists and some essayists go narrative, because the whole pleasure is in the unfolding. None of these is wrong. They're just different tools for different jobs, and a ScriptGrain profile records which one you actually reach for. It sits in the Rhetoric group alongside opening style and closing pattern, the two ends of whatever order you choose.

How argument structure is scored

Argument structure does move your voice match score, at a weight of 1.5. That's the heaviest weight in the Rhetoric group, though some features elsewhere in the profile weigh more. Here's how it works: Claude Haiku reads your draft and judges its structure label, that label gets compared in code against what's stored in your profile, and the comparison isn't strictly pass or fail.

An exact label match scores 1. If either side reads as "mixed" or "varied", you get 0.7. If your profile's stored value contains the draft's label, that's 0.9. Anything else scores 0.15, a steep drop. Labels like this are also the least stable part of a profile: ScriptGrain's study SGR-003 found that the 13 enumerated labels in profiles built from fewer pieces matched a 15-piece profile 72% to 87% of the time, depending on piece count (82% at one piece, 87% at eight). Argument structure is one of those labels, so treat any single reading as a judgement, not gospel.

For the maths behind how this weight combines with everything else, see the voice measurement framework and the voice match glossary entry.

How to analyse argument structure in your own writing

You can do this by hand, starting with a single piece of writing in front of you.

  1. Pull out three or four paragraphs from something you've written recently.
  2. Read the first sentence of each paragraph and ask: is this a claim, a fact, or a scene?
  3. If it's a claim ("this approach works better"), you're probably conclusion-first.
  4. If it's a fact or observation building towards something later, you're probably evidence-first.
  5. If it's setting a scene or telling what happened, that's narrative.
  6. Do this across five or six pieces, not just one, since structure often shifts by context (an email versus a report, say).
  7. Look for the pattern that shows up most, not the one that shows up once.

That's the manual version. ScriptGrain's free voice profile reads this attribute, among all 45, straight from a writing sample, no sorting by hand required.

Examples of argument structure in real writing

Example (conclusion-first): "Cut the meeting to fifteen minutes. Nobody needs an hour to agree on a colour scheme."

Example (evidence-first): "Three of the five reviewers flagged the same paragraph. Two more asked for the same edit independently. The paragraph needs to go."

Example (narrative): "The client rang at nine, furious about the invoice. By ten, we'd found the billing error was ours."

Notice the shape in each. The first states the verdict, then defends it in one sentence. The second stacks small observations until the conclusion falls out naturally. The third just tells you what happened and trusts you to draw the line yourself.

How to make AI lead with the conclusion

If you want ChatGPT or any assistant to argue conclusion-first, don't ask it to "explain" something; ask it to answer, then explain.

  1. Ask for the answer in the first sentence, then the reasons.
  2. Ask explicitly for the recommendation before the analysis, not the other way round.
  3. Tell it plainly to put the bottom line in the first line.

When ScriptGrain writes from your profile, `argument_structure` is sent as part of the voice profile JSON in the system prompt, treated as a hard constraint rather than a nudge, with a named rule attached: "Use the argument structure specified (e.g. conclusion-first)". The profile itself reaches ChatGPT, Claude and other tools through the API and MCP server, so they can read your structure instead of you retyping the same brief every session.

How to make AI build the argument from evidence

If evidence-first is what you're after, the trick is patience: ask for the case, not the verdict, up front.

  1. Ask for the observations or data first, and the conclusion last, explicitly in that order.
  2. Let each paragraph add exactly one piece of evidence, no jumping ahead.
  3. Tell the model to earn the claim before it states it, rather than asserting and backfilling.

Same mechanism here. If your samples read as evidence-first, that label goes into every ScriptGrain draft as a hard constraint, and ChatGPT, Claude or any other tool reading your profile through the API or MCP server gets the same label. You're not re-explaining your structure preference each time.

Questions

How do I structure an argument in writing?

Pick your order deliberately: conclusion-first if your reader is busy and wants the point fast, evidence-first if you're building trust through data, narrative if the story itself carries the argument. Busy-reader formats such as emails, reports and pitches usually suit conclusion-first. Academic and technical writing often earns more credibility evidence-first. There's no universal right answer, only the right fit for your reader.

Conclusion first vs evidence first, which is better?

Neither is objectively better; they solve different problems. Conclusion-first respects a reader's time and works well when your ask is simple. Evidence-first works when the conclusion would seem unearned without the buildup, like a technical recommendation or a controversial claim. Bottom line up front (BLUF) writing suits business and marketing contexts because readers skim.

How is argument structure scored in a voice profile?

It's judged, not counted: Claude Haiku reads your draft, assigns a label, and code compares that label against your stored profile value. Exact matches score 1, "mixed" or "varied" overlaps score 0.7, a profile value containing the draft's label scores 0.9, and a mismatch scores 0.15. It's weighted 1.5 in the overall voice match calculation, the heaviest of the rhetoric attributes.

How do I make AI lead with the conclusion?

Ask for the answer before the analysis, every time: the conclusion in the first sentence, then the reasoning. Tell the model directly to put the bottom line in the first line. If your ScriptGrain profile reads conclusion-first, tools connected through the API or MCP server can read that label instead of you re-specifying it each session.

How do I make ChatGPT structure arguments better?

Tell it which structure you want before it writes: conclusion-first, evidence-first or narrative. Left unspecified, a model picks its own order, which may not be yours. Name the order in the brief, paste one of your own paragraphs as an example of the move, and check the first sentence of each paragraph in the draft. A ScriptGrain profile holds your structure as a label that ChatGPT can read through the API or MCP server.

Related

Sources