Humour register analyser

By Jack Stovell · published 2026-09-24

Humour register (humour_register) is the dominant humour style in a piece of writing: dry, sarcastic, self-deprecating, warm or none. A ScriptGrain voice profile stores it as one of those five labels, judged by a model from your samples rather than counted, and it moves the voice match score with a weight of 1.5.

Analyse your humour register free

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

Try the six-signal check first

At a glance

What humour register measures

Humour register records which kind of humour shows up most in your writing, or that none does. The app's own definition: "the dominant humour style: dry, sarcastic, self-deprecating, warm, or none." Five options, one label. Dry humour lands through understatement and deadpan observation. Sarcastic humour says the opposite of what it means. Self-deprecating humour turns the joke inward. Warm humour is gentle, inclusive, rarely at anyone's expense. None means exactly that: no detectable comic register in the sample.

A writer who mixes two kinds still gets one label: the value is snapped to one of the five when the profile is saved. It's a judgement call, made by a model reading tone rather than a script tallying jokes, and labels are the least stable part of a profile. In ScriptGrain's SGR-003 study, 13 labelled attributes matched a 15-piece profile 72% to 87% of the time, depending on how many pieces built it. Humour register is one of six attributes in the tone and register group.

How humour register is scored

Humour register is scored by a model labelling your draft and code comparing labels; nothing counts punchlines. When voice match runs, Claude Haiku reads the draft and names its humour register, and code compares that label with your profile's, at a weight of 1.5. There's no reference corpus for this attribute, so nothing here is benchmarked against a typical value.

An exact match scores 1. If either side says "mixed" or "varied", that scores 0.7. A profile value that contains the draft's label scores 0.9. Three adjacent pairs score 0.5, because they're near neighbours: dry and none, dry and sarcastic, warm and self-deprecating. Any other pairing scores 0.15, so a warm draft against a dry profile keeps little credit on this feature. The full method is on the voice measurement framework page.

How to analyse humour register in your own writing

You can work this out by hand, and it takes about ten minutes with a highlighter and a bit of honesty about your own habits.

  1. Pull three or four pieces you've written that felt natural, not ones you laboured over.
  2. Highlight every line that made you smile while writing it, or that a reader flagged as funny.
  3. Ask what each line is doing: understating something (dry), inverting meaning (sarcastic), mocking yourself (self-deprecating), or being gently affectionate (warm).
  4. Count which category wins. A profile stores one label, so take the winner and note the runner-up for yourself.
  5. Check whether "none" is actually the honest answer. Plenty of good writing has no humour register at all, and that's fine.

ScriptGrain's free voice profile reads humour register from a writing sample among all 45 attributes.

Examples of humour register in real writing

Examples help more than definitions here, so three short ones, each doing a different job.

Example (dry): "The meeting achieved nothing, which at least matched expectations."

Example (sarcastic): "Sure, because that plan has worked brilliantly every other time."

Example (self-deprecating): "I proofread this three times, which explains why it only has two mistakes."

Notice none of these need an exclamation mark to work. Humour register lives in the observation, not in the volume.

How to make AI writing funnier in your register

Making AI writing funnier means naming the register you want, because "be funny" tends to get the wrong kind of joke.

  1. Name the register explicitly: dry, sarcastic, self-deprecating or warm. Skip "just make it funnier."
  2. Paste in two lines of your own that actually land, and ask the model to write something in that same register, never the same joke rephrased.
  3. Push the humour into the observations themselves, the little asides and specific details, rather than bolting a punchline onto the end.
  4. Rule out exclamation marks. They flatten dry and sarcastic humour into something closer to enthusiasm, which isn't the same thing.

When ScriptGrain writes from your profile, the humour register goes into the voice profile JSON block as a hard constraint, and the system prompt adds a named rule: "Apply the correct humour register; do not introduce warmth or exclamation if the profile shows dry/rare". The profile's values reach ChatGPT, Claude and other tools through the API and the MCP server. Emotional expressiveness sets how openly feeling shows around the jokes.

How to make AI writing more serious

Making AI writing more serious means removing humour deliberately, not just asking for "less funny."

  1. Set the humour register to none, and say so in plain words: "no humour, no jokes, no wordplay."
  2. Strip out winks, asides and exclamation marks wherever they've crept in.
  3. Wherever a quip would normally sit, ask instead for a plain statement of fact.

When your profile reads "none", ScriptGrain sends that label in the voice profile JSON block as a hard constraint, like every other attribute, and the profile reaches ChatGPT, Claude and other tools through the API and the MCP server. Formality score and confidence vs hedging shape how serious a piece reads once the jokes are gone.

Questions

What is a humour register analyser?

A humour register analyser is a tool or model that reads a piece of writing and identifies its dominant comic style: dry, sarcastic, self-deprecating, warm, or none. ScriptGrain's version judges this from writing samples as one of 45 tracked attributes, storing it as a single label under the Tone and register group rather than counting jokes mechanically.

How is humour register scored in writing?

Claude Haiku labels the draft's humour and code compares that label with the one stored in your voice profile, at a weight of 1.5. An exact match scores 1, "mixed" or "varied" 0.7, adjacent pairs such as dry and sarcastic 0.5, and unrelated labels 0.15. There's no reference corpus or typical value for this attribute, so nothing gets benchmarked against an average.

How do I make ChatGPT less cheesy?

Name the register you actually want (dry, sarcastic, self-deprecating or warm) instead of asking for "funnier." Give two lines of your own that land, ask for the same kind of joke rather than the same joke, and explicitly ban exclamation marks. Keep the humour in the observations, never in added punchlines.

How do I make AI writing more serious?

Set the humour register explicitly to none. Remove wordplay, winks and exclamation marks from your instructions and examples. Wherever a joke would normally go, ask for a plain factual statement instead. Stating it outright gives the model a rule instead of a guess.

How do I add humour to AI writing without it feeling forced?

Push jokes into your observations rather than tacking on punchlines. Give two real lines of your own humour and ask for the same register, never the same gag. Name the register instead of asking for "funny", and leave out exclamation marks, which push dry and sarcastic lines towards enthusiasm.

Related

Sources