Power sentence position

By Jack Stovell · published 2026-09-24

Power sentence position (power_sentence_position) is where the strongest sentence in a paragraph tends to land: start, end, both or varied. ScriptGrain's extraction model judges it by reading your samples; it is never counted in code. It sits in the Quirks and cadence group and never moves the voice match score.

Analyse your power sentence position free

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

Try the six-signal check first

At a glance

What power sentence position measures

Power sentence position measures the habit, not any single paragraph. Some writers lead with the claim and spend the rest of the paragraph backing it up. Others build, clause by clause, towards a line that lands at the close. Some do both, opening strong and closing stronger. Others vary it paragraph to paragraph, and that variety is itself the pattern. ScriptGrain's own definition puts it plainly: "Where the strongest sentence tends to land in a paragraph." Note the scale: start, end, both, or varied. Four values, no in-between grades, no percentages attached to any of them.

Unlike sentence length, which code counts when a draft is scored, this one has no counter. Nobody can point to a line of code that scans your paragraphs and tags the "strongest" sentence, because strength is a judgement call. Which sentence carries the most weight? A model has to read the paragraph and decide, the same way a human editor would. The extraction model's reading is stored as written, without being snapped to one of the four labels, and like other judged labels it can vary slightly from run to run: ScriptGrain's SGR-003 study measured label agreement at 72% to 87%.

How power sentence position is scored

Power sentence position never moves your voice match score. The voice match engine compares a draft with your profile feature by feature: code counts sentence, punctuation, pronoun, article and word-length features, and Claude Haiku judges formality, humour, structure and rhythm. For power sentence position there's no comparator at all. So it sits in your profile and shapes how ScriptGrain writes for you, but it plays no part in the 0 to 1 score a draft receives, and polish can't target it. Of the 45 attributes, 28 can move the score. If you want the mechanics behind what does move the score, the voice measurement framework walks through it.

How to analyse power sentence position in your own writing

You can do this by hand with a few paragraphs pulled from your own work.

  1. Pick five or six paragraphs you're happy with, ideally from different pieces.
  2. In each one, find the sentence that carries the actual point, the one you'd keep if you had to cut everything else.
  3. Note where it sits: opening line, closing line, both, or somewhere in the middle.
  4. Tally the results across your sample. A clear majority in one spot is your default. An even spread across positions means you vary it.
  5. Check whether that default matches how you think you write.

ScriptGrain's free voice profile reads this straight from a writing sample as one of all 45 attributes, no manual tallying required.

Examples of power sentence position in real writing

Here's what three of the four values look like on the page.

Example (start): "This plan won't work. The budget's wrong, and the timeline is tighter than anyone admits." The claim opens the paragraph; everything after it is support.

Example (end): "The budget's wrong, and the timeline is tighter than anyone admits. This plan won't work." Same facts, reordered so the verdict lands last.

Example (both): "This plan won't work. The budget's wrong, and nobody has asked the obvious question. It won't work." Bookended: the claim opens and closes, doing double duty.

No typical value applies here; there's no reference corpus for this attribute, so there's nothing to benchmark your paragraph against.

How to make AI lead with the strongest sentence

Getting an AI assistant to open with the point takes a handful of direct instructions, not a vague "be punchier" note.

  1. Ask for the key sentence first in every paragraph, explicitly, in the prompt.
  2. Move the conclusion of each paragraph to its opening line and let the reasoning trail behind it.
  3. Check that the first lines alone, read in sequence, tell the whole story without the rest of the paragraph.

When a ScriptGrain profile has power_sentence_position set to "start" (or "both"), that value is sent in the voice profile JSON of the system prompt, under "Treat every attribute as a hard constraint, not a suggestion." There's no separate rule bolted on for this one; the value itself does the work. That profile reaches ChatGPT, Claude and other tools through the API and the MCP server.

How to make AI end paragraphs on the strongest sentence

Getting a paragraph to build towards its best line is mostly a matter of cutting what comes after it.

  1. Ask each paragraph to build to its key sentence rather than open with it.
  2. Cut anything written after the line that actually lands; a strong closer followed by a weak add-on undoes itself.
  3. End sections on the most concrete claim you've got, not a hedge or a summary restatement.

Generation works the same way here as it does for "start": the profile sends "end" (or "both") inside the voice profile JSON as a hard constraint, and it reaches ChatGPT, Claude and other tools through the API and MCP server. Rhythm pattern and structural signatures, in the same Quirks and cadence group, describe the rest of a paragraph's shape.

Questions

Where should I put the strongest sentence in a paragraph?

There's no single right answer; it depends on effect. Leading with it gives readers the point immediately, useful for skimmers or dense arguments. Ending with it builds tension and rewards the read. Doing both bookends the paragraph for emphasis. Varying it keeps longer pieces from feeling mechanical. ScriptGrain records whichever pattern you already favour rather than telling you which one to use.

How is power sentence position measured?

It's judged, not counted. Claude Sonnet 5 reads your writing samples across two passes, one per sample and one synthesis pass, and decides which value (start, end, both, or varied) best fits your habits. There's no code scanning for it and no numeric score attached. It's the extraction model's reading of your own samples, nothing more precise than that.

Does power sentence position affect my voice match score?

No. The voice match engine has no comparator for this attribute, so it never moves your score up or down, and polish can't target it. It still shapes how ScriptGrain writes on your behalf; it just sits outside the scoring mechanics entirely. Of the 45 attributes, 28 can move the score, and this is one of the 17 that cannot.

How do I get ChatGPT to lead with the key point?

Tell it directly: key sentence first, every paragraph, no exceptions. Ask it to move each paragraph's conclusion to the opening line, then check that the first sentences alone, read straight through, tell the full story. When ScriptGrain writes from a profile with power_sentence_position set to "start", that value goes into its system prompt as a hard constraint.

How do I stop AI writing from trailing off after the good line?

Ask for paragraphs that build towards their strongest sentence, then cut whatever follows it. A closing line often gets undercut by a weaker sentence tacked on out of habit. Ending sections on the most concrete claim available, rather than a soft summary, helps too. It's the same discipline good editors apply by hand.

Related

Sources