Emotional expressiveness analyser

By Jack Stovell · published 2026-09-24

Emotional expressiveness (emotional_expressiveness) is how openly emotion shows in the prose. A ScriptGrain voice profile stores it as low, medium or high, judged by a model from your samples; no code counts it. It moves the voice match score with a weight of 1, and a draft one step away from your profile keeps half the credit.

Analyse your emotional expressiveness free

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

Try the six-signal check first

At a glance

What emotional expressiveness measures

Emotional expressiveness measures how visible feeling is on the page, not how much feeling the writer had while typing. You can be furious and write a flat memo. You can be mildly annoyed and write something that reads like a breakup letter. The attribute cares about what lands in the text.

In ScriptGrain's model, emotional expressiveness sits in the Tone and register group, alongside things like formality score and humour register. See the full group at tone and register. The definition the app uses is blunt: "How openly emotion shows in the prose." No sub-clauses, no wiggle room.

Low, medium and high are judgements, not tallies. An extraction model (Claude Sonnet 5, if you want the specifics) reads each writing sample in a first pass, then synthesises everything into one profile in a second pass. That's a judged reading of a pattern, not a stopwatch on your exclamation marks. It gets snapped to one of the three canonical labels when your profile is saved.

How emotional expressiveness is scored

Emotional expressiveness enters your voice match score as one of the attributes Claude Haiku judges from a draft; no code counts it. Code then compares the draft's label with your profile's, at a weight of 1. The comparison happens on an ordered scale: land on the same label as your profile and you score a full 1; be one step off (low versus medium, say) and you score 0.5; be two steps off (low versus high) and you drop to 0.1. An unreadable label scores 0.15.

The ordered scale means a near miss keeps half the credit on this feature and a two-step miss keeps almost none: a gushing draft against a low profile scores 0.1. Full detail on how this and 27 other attributes combine into one number lives at voice match and the voice measurement framework.

How to analyse emotional expressiveness in your own writing

Emotional expressiveness is something you can rate by hand in a few minutes with a decent sample of your own writing. Here's a method that doesn't need software:

  1. Pull three or four paragraphs you'd call typical, not your best or your worst.
  2. Underline every word that names or implies a feeling: happy, frustrated, relieved, worried, thrilled.
  3. Check whether those feelings are stated plainly or dressed up with intensifiers like incredibly, deeply, truly.
  4. Count exclamation marks and rhetorical flourishes separately; they're a different lever, but they inflate the same impression.
  5. Ask whether a stranger reading it cold would say "this person clearly feels something" or "I can't tell how they feel about this."
  6. Sort the sample: no visible feeling reads low, stated but restrained reads medium, openly emotive reads high.

It's judgement, same as the model's, just done by eye instead of by Claude. ScriptGrain's free voice profile reads emotional expressiveness from a writing sample among all 45 attributes.

Examples of emotional expressiveness in real writing

Low: "The client cancelled the contract on Tuesday. We're reviewing the terms." (Example: no named feeling, no intensifier, pure fact.)

Medium: "I was disappointed by the cancellation, though I understand their reasoning." (Example: feeling named once, plainly, no stacking.)

High: "I was gutted. Genuinely gutted. This one hurt more than I expected." (Example: repetition and intensifier doing the emotional lifting.)

How to make AI writing more emotional

AI writing gets more emotional when you tell it to name feelings plainly instead of hiding behind neutral description. Try this:

  1. Ask the model to name the feeling directly: "frustrated," "relieved," "thrilled," not vague euphemisms.
  2. Instruct it to keep first-person reactions in the text rather than editing them out as unprofessional.
  3. Tell it to let word choice do the emotional work: one well-chosen word beats three stacked adjectives.
  4. Explicitly ban exclamation marks as the shortcut; they're a crutch, not an emotion.
  5. Review the draft and cut anything that reads like performed enthusiasm rather than a stated feeling.

When ScriptGrain writes from your profile, emotional expressiveness travels in the voice profile JSON block of the system prompt as a hard constraint, same as every other attribute. There's no special rule bolted onto this one. The profile reaches ChatGPT, Claude and other tools through the API and the MCP server, so they can read the same label.

How to make AI writing less emotional

AI writing gets less emotional when you ask for the facts and the decision, and nothing else. Concretely:

  1. Ask for the facts and the outcome, explicitly without adjectives of feeling.
  2. Cut intensifiers on sight: incredibly, deeply, truly, and their cousins.
  3. Remove exclamation marks entirely; a full stop carries a fact just fine.
  4. Strip rhetorical flourishes, the questions and asides that exist purely to signal enthusiasm.
  5. Re-read for anything that sounds like it's trying to make you feel something rather than telling you something.

That is also how you stop ChatGPT gushing: strip the levers above one at a time and check what's left. If your samples read as low, a ScriptGrain profile carries that label into every draft it writes through the same JSON block, and to other tools through the API and MCP server.

Questions

How do you measure emotion in writing?

You measure it by checking how openly feeling shows on the page: named feelings, intensifiers, exclamation marks, first-person reactions. ScriptGrain's extraction model reads a sample and judges the level as low, medium or high rather than counting anything in code. For your own writing, underline emotion words and intensifiers, then judge the overall impression a stranger would form.

How is emotional expressiveness scored?

It's judged, not counted. An extraction model reads your writing samples and assigns low, medium or high. In voice match scoring, Claude Haiku judges the same scale from a draft, weight 1, on an ordered comparison: same label scores 1, one step apart scores 0.5, two steps apart scores 0.1, unreadable scores 0.15.

How do I make AI writing sound more emotional?

Ask the model to name feelings plainly, keep first-person reactions rather than editing them out, and let word choice carry weight instead of stacking adjectives or exclamation marks. Avoid performed enthusiasm; one honest word beats three loud ones. When your samples read as high, a ScriptGrain profile sends that label as a hard constraint every time ScriptGrain writes.

How do I make ChatGPT sound less gushing?

Strip intensifiers like incredibly, deeply and truly, cut exclamation marks, and ask explicitly for facts and decisions without adjectives of feeling. Read back for anything performing enthusiasm rather than stating a fact. A ScriptGrain profile that reads your writing as low carries that label to ChatGPT through the API or MCP server.

What's the difference between an emotional tone analyser and a formality check?

An emotional tone analyser judges how openly feeling shows; a formality check judges how casual or formal the register is. They are separate readings in a ScriptGrain profile, so a piece can be highly formal and still openly emotional, or casual and flat. Both sit in the tone and register group with humour register, contraction rate, confidence vs hedging and audience adaptation.

Related

Sources