Clause order analyser

By Jack Stovell · published 2026-09-24

Clause ordering (clause_ordering) records whether your sentences lead with the point or work up to it, as one of three labels: front-loaded, build-to-point or mixed. No code counts it; a model judges it from your samples. In ScriptGrain's SGR-003 study, labels like this matched a 15-piece profile 72% to 87% of the time.

Analyse your clause ordering free

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

Try the six-signal check first

At a glance

What clause ordering measures

Clause ordering measures where the main idea sits in a sentence, relative to the conditions around it. Front-loaded writing puts the conclusion first and the "because", "if" and "when" clauses after. Build-to-point writing does the opposite: context first, verdict last. Mixed writing does both, depending on what the sentence needs. This sits in the Syntax group of a voice profile alongside average sentence length, sentence length variance and sentence complexity. Clause order isn't about vocabulary or tone. It's structural. It's about which bit of the sentence you meet first.

News writing is the familiar front-loaded case: the inverted pyramid puts the most important fact first. Neither order is right or wrong. It's a pattern, and patterns are what a voice profile is built to catch.

How clause ordering is scored

Clause ordering moves your voice match score at weight 1, and a model judges it. When ScriptGrain checks a draft against your profile, Claude Haiku reads the draft and assigns it a label: front-loaded, build-to-point or mixed. That label gets compared to your profile's stored value in code. An exact match scores 1. If either side says "mixed" or "varied", you get 0.7. If your profile's value contains the draft's label, that's 0.9. Anything else scores 0.15.

So this isn't a sliding scale measuring how front-loaded a sentence is. It's a label match. That matters because labels are blunter than continuous scores, and less stable: in ScriptGrain's SGR-003 study, the 13 enumerated labels matched the author's 15-piece profile 72% to 87% of the time depending on piece count (82% at one piece, 87% at eight). Clause ordering is one of those labels. The full method is on the voice measurement framework page, and voice match has the short version.

How to analyse clause ordering in your own writing

You can do this by hand in about ten minutes with a page of your own writing.

  1. Pull ten to fifteen sentences from something you've actually published, not a draft you're still fiddling with.
  2. For each sentence, find the main clause: the bit that would survive if you deleted every "because", "if", "when" and "although" clause attached to it.
  3. Check where that main clause sits. Front? Back? Split across the sentence?
  4. Tally your results. Mostly front means front-loaded. Mostly back means build-to-point. Roughly even means mixed.
  5. Repeat across a few pieces if you want confidence rather than a snapshot; one piece can mislead you.

That's the manual version. ScriptGrain's free voice profile reads clause ordering from a writing sample among all 45 attributes, no manual tallying required.

Examples of clause ordering in real writing

Clause order shows most clearly when the same facts are arranged both ways.

Example: "The launch is delayed until March, because the supplier missed two shipments." Front-loaded. The verdict lands first; the reason trails behind it.

Example: "Because the supplier missed two shipments, the launch is delayed until March." Build-to-point. Same facts, reversed order, and it reads like a slower reveal.

Example: "If the numbers hold, we ship in March." Build-to-point again, this time with "if" doing the leading. Swap it: "We ship in March, if the numbers hold" and you've front-loaded the same sentence.

How to make AI put the point first

Front-loading is a request you can make directly, as long as you're specific.

  1. Ask for the main clause first and the conditions after it.
  2. Tell the model to open each paragraph with its conclusion, not its scene-setting.
  3. Move "because" and "if" clauses to the end of the sentence, after the main point.

A ScriptGrain profile sends your label inside the voice profile JSON in the system prompt, under "Treat every attribute as a hard constraint, not a suggestion": if your profile reads front-loaded, every ScriptGrain draft starts from that. The profile reaches ChatGPT, Claude and other tools through the API and the MCP server.

How to make AI build up to the point

Build-to-point writing asks the model to hold its verdict until the end.

  1. Ask for context and conditions first, with the main point arriving at the end of the sentence.
  2. Ask it to end paragraphs on the claim the paragraph has earned, rather than opening with it.
  3. Let "if", "when" and "although" clauses lead the sentence instead of trailing it.

In ScriptGrain, a build-to-point label travels in the voice profile JSON as a hard constraint on every draft, so you never retype it as a one-off prompt line. The profile reaches ChatGPT, Claude and other tools through the API and MCP server.

Questions

What is a front-loaded sentence?

A front-loaded sentence puts its main point first and its conditions, reasons or caveats after. "The launch is delayed, because two shipments were missed" is front-loaded. Reverse the order and you get build-to-point: same information, different sequence. Front-loading tends to read as direct, and news writing uses it to put the key fact first.

How is clause ordering scored?

It's scored as part of voice match at weight 1. Claude Haiku labels a draft as front-loaded, build-to-point or mixed, and that label is compared to your profile's stored value in code. Exact match scores 1, "mixed" on either side scores 0.7, a profile value containing the draft's label scores 0.9, and anything else scores 0.15. It's a label comparison, not a continuous measurement.

How do I make AI get to the point?

Tell it directly: main clause first, conditions after. Ask it to open paragraphs with the conclusion rather than build up to it, and to move "because" and "if" clauses to the end of the sentence. If your ScriptGrain profile reads front-loaded, every ScriptGrain draft carries that label as a hard constraint, and the profile reaches ChatGPT, Claude and other tools through the API and MCP server.

How do I make ChatGPT lead with the point?

Ask explicitly for front-loaded sentences: point first, reasoning after. Push it to open each paragraph with the verdict, not the scene-setting, and to shift subordinate clauses like "because" and "if" toward the end. If your ScriptGrain profile reads front-loaded, ChatGPT can read that label through the API and MCP server, so it sits in the profile instead of a prompt you retype every session.

How do I write front-loaded sentences by hand?

Draft the sentence normally, then find the main clause: the part that survives once you strip out anything starting with "because", "if", "when" or "although". Move that main clause to the front. Read it aloud. If it sounds like a headline rather than a story, you've got it.

Related

Sources