Rhetorical analyser

By Jack Stovell · published 2026-09-24

A rhetorical analyser reads how a piece is built to persuade: how it opens and closes, its metaphor and repetition, how it moves between ideas and whether the argument leads with evidence, with the conclusion or as a narrative. ScriptGrain's Rhetoric group is 6 of 45 voice attributes, all judged by a model; 4 move the voice match score.

Analyse your rhetoric attributes free

Your free voice profile measures every rhetoric 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 rhetoric group measures

The rhetoric group captures how a piece is built to persuade: how it opens and closes, whether metaphor and repetition do any work, how ideas transition, and what order the argument unfolds in. Six of the 45 voice attributes, all judged by the extraction model, none counted in code and none pulled from a catalogued word list.

Opening style asks how pieces typically start: a hook, context first, a direct point, an anecdote, or a question thrown at the reader. It's judged, and it carries a voice-match weight of 0.75, so it does move your score.

Closing pattern is the mirror image: how pieces tend to end, captured as a short description of up to 15 words. Judged, but not scored: it never moves the match percentage, though the profile still sends it to every draft ScriptGrain writes.

Metaphor usage estimates how often metaphors turn up, expressed as a rate per 1,000 words. Judged, weight 0.75. Someone who reaches for a metaphor every other paragraph reads very differently from someone who never does, and the profile records that difference.

Repetition as emphasis is a plain yes or no: does this writer repeat a phrase or structure on purpose, to land a point? Judged, weight 0.5. Smaller weight, still real.

Transition style describes how the writing moves from one idea to the next, again as a short description capped at 15 words. Judged, not scored; like closing pattern, it guides drafts without touching the match percentage.

Argument structure is the big one here: evidence-first, conclusion-first, or narrative. Judged, weight 1.5, the heaviest weight of the six. This single attribute separates writing that sounds like a report from writing by someone who's already made up their mind and is now explaining why.

How rhetoric features are scored

Rhetoric features are judged by a model at both ends, because you can't count your way to knowing whether an argument is conclusion-first or evidence-first. Claude Sonnet 5 reads all 6 from your samples into the profile; when a draft is scored, Claude Haiku reads the 4 scored ones from the draft and code compares the two. None of the 6 is counted in code or taken from a word list.

That's different from the rest of the profile. For 14 of the other 39, such as sentence length and contraction frequency, code counts the feature when it scores a draft or runs the free tools. Others are catalogued from word lists.

Four of the six move your voice match score: opening style at weight 0.75, metaphor usage at 0.75, repetition as emphasis at 0.5 and argument structure at 1.5. Closing pattern and transition style never move it. Across the full profile, 28 of the 45 attributes can move the score, and the voice measurement framework publishes the weights.

Every value you see in a profile, scored or not, is the extraction model's reading of your writing (Claude Sonnet 5), built across two passes: one that reads each sample and pulls every attribute, and a second that synthesises the lot into a single profile with a narrative and a confidence score from 0 to 1. Nothing here is guessed by a human. It's a model, reading, twice.

How to analyse the rhetoric of your own writing

You can do this by hand if you've got a spare afternoon and a handful of your own drafts. Here's how:

  1. Pull five or six pieces you've written and read only the first two sentences of each. Note whether you open with a hook, a question, straight context, an anecdote, or the direct point itself. Look for the pattern, not the exception.
  2. Do the same for the last two sentences. Write a one-line description of how you tend to close, in your own words, capped at around 15 words.
  3. Count metaphors across a few thousand words and work out a rough rate per 1,000 words. Be honest: a stray "at the end of the day" doesn't count, an actual comparison does.
  4. Check for deliberate repetition: do you repeat a phrase, a sentence shape, or a word on purpose to land a point? Yes or no, no hedging.
  5. Read the joins between paragraphs. Are you using short bridging fragments, or long connecting clauses? Describe your transition habit in a sentence.
  6. Look at three or four of your strongest pieces of persuasive writing and ask where the conclusion sits: at the top, at the bottom, or woven through as a story. That's your argument structure.

ScriptGrain's free voice profile reads all six of these straight from a writing sample, so you don't have to do the manual version.

How to make AI writing match your rhetoric

Making an AI argue like you means giving it your opening move, closing habit, metaphor rate and argument structure instead of a vague instruction to "sound more like me".

  1. Gather real samples, ideally 3,000-plus words across a handful of pieces (the app takes up to 20 samples). More samples, steadier reading.
  2. Get a reading of all six from several pieces, by hand or from a voice profile, rather than asking a chatbot to eyeball your last email.
  3. Feed the model your argument structure explicitly. If you're conclusion-first, say so, and show it. Argument structure carries the heaviest voice-match weight of the six (1.5), so getting it wrong costs a draft more than any other rhetoric miss.
  4. Give it your metaphor rate and your repetition habit as instructions, not vibes. State the rate as a number from your own samples (for example, "about two metaphors per 1,000 words"); "be more colourful" gives it nothing to aim at.
  5. Check the output against your closing pattern and transition style too, even though they're not scored. They're still what makes a piece sound like you rather than like a competent stranger.

A ScriptGrain profile sends all six into every draft it writes, as hard constraints, and carries them into ChatGPT, Claude and other tools through the API and the MCP server: `GET /v1/profiles/{id}` and the MCP tool `get_profile` both return all 45 attributes, rhetoric included, as JSON, on every plan. That's the difference between typing "sound like me" into a chat window and handing the tool your rhetoric as data it can hold onto.

Attributes in this group

Questions

What is a rhetorical analyser?

A rhetorical analyser reads how a piece of writing is built to persuade: how it opens and closes, whether it leans on metaphor or repetition, how it moves between ideas, and whether it leads with evidence, with the conclusion, or as a narrative. ScriptGrain treats this as one group of 6 attributes out of 45, all judged by an extraction model rather than counted in code.

How do I analyse the rhetoric in my own writing?

Read your openings and closings across several pieces and describe the pattern in your own words. Count metaphors per 1,000 words. Check whether repetition is deliberate. Note how you transition between ideas, and where your conclusion sits: top, bottom, or threaded through as a story. That last one is argument structure, the rhetoric attribute with the heaviest weight in ScriptGrain's voice match score.

What is a rhetoric score, and does it affect my voice match?

ScriptGrain's voice match gives rhetoric no separate score. Instead, 4 of the group's 6 attributes feed the single voice match score with their own weights: opening style (0.75), metaphor usage (0.75), repetition as emphasis (0.5) and argument structure (1.5). Closing pattern and transition style are judged but never scored. Across the full 45-attribute profile, 28 attributes can move the score.

How can I make AI argue like me?

Give the model your real argument structure (evidence-first, conclusion-first or narrative) rather than a vague style instruction, since it carries the heaviest rhetoric weight in the voice match score (1.5). Add your actual metaphor rate, your opening habit and whether you repeat phrases for emphasis. A ScriptGrain profile stores all of this and carries it into ChatGPT, Claude and other tools through the API and MCP server, so the tool has the pattern, not a guess.

How do I make AI structure an argument like me?

Tell it explicitly where your conclusion lives: first, last, or carried by a story. That's argument structure (evidence-first, conclusion-first or narrative), which the extraction model judges and the voice match score weights at 1.5, the heaviest of the six rhetoric attributes. Pair that with your opening style and transition habits so the shape of the argument, not just its tone, matches. A voice profile holds this as structured data and passes it through the API or MCP server into whichever tool you're drafting in.

Related

Sources