Guide
Choosing tone and humanness: a decision guide
A reference table by content type — what tone, domain, and humanness to pick for what you're writing.
title: "Choosing tone and humanness: a decision guide" description: "A reference table by content type — what tone, domain, and humanness to pick for what you're writing." date: "2026-05-13" tag: "Guide" author: "Inksong"
Seven tones, six domains, a humanness slider from 0 to 100. That's 4,200 combinations on paper. In practice, you don't need them all — most users settle into four or five presets that match the kinds of content they actually write. This guide is the reference table we wish we'd had when we started: what to pick for what you're writing, with reasoning attached.
The mental model
Three controls, three different jobs.
- Tone sets the register — formal vs. casual, academic vs. conversational. It's a categorical choice. Picking
academicdoesn't make the rewrite more aggressive; it shifts the style of the output toward formal academic prose. The seven options (balanced, academic, pastoral, casual, professional, creative, journalistic) cover most writing contexts. - Domain controls vocabulary discipline. Pick the domain matching your content (academic, legal, pastoral, medical, marketing, technical) and the rewrite respects domain-specific terms — it won't paraphrase "supersedes" into "replaces" in a legal document, or "patient-reported outcome" into "what patients said" in a medical one. Domain also influences which transitions and structures the model preserves.
- Humanness is the volume knob. It controls how much rewriting happens, not what kind. At 0, the rewrite is minimal. At 100, the rewrite is aggressive — entire sentences restructured, vocabulary widened, rhythm reshaped.
Tone and domain are categorical. Humanness is continuous. Get the first two right and humanness becomes a fine-tune.
By content type — a reference table
The settings below are starting points. Adjust humanness ±10 for your taste; tone and domain are usually solid as listed.
| Content type | Tone | Domain | Humanness | Notes |
|---|---|---|---|---|
| Academic paper / thesis | academic | academic | 40 | Preserves hedging, formal register. |
| Research grant proposal | academic | academic | 35 | Lower humanness; reviewers expect structure. |
| Literature review section | academic | academic | 45 | Slightly higher — review prose can be drier. |
| Sermon | pastoral | pastoral | 50 | Preserves theological vocabulary and cadence. |
| Pastoral newsletter | pastoral | pastoral | 55 | Slightly more conversational than sermons. |
| Marketing blog post (B2B) | balanced | marketing | 60 | Professional but de-AI'd. Don't use casual. |
| Marketing blog post (consumer) | casual | marketing | 70 | Punchier, more informal, voice-led. |
| Email sequence | balanced | marketing | 65 | Slightly higher than blog — emails read fast. |
| Landing page hero | professional | marketing | 55 | Lower humanness; copy is dense, every word counts. |
| Product documentation | professional | technical | 35 | Low humanness — clarity over voice. |
| Legal contract | professional | legal | 25 | Very conservative; precision is everything. |
| Medical patient communication | professional | medical | 45 | Plain-language but respect medical terms. |
| Fiction draft (literary) | creative | academic | 70 | Yes, academic domain — preserves literary vocabulary. |
| Fiction draft (genre) | creative | marketing | 75 | Marketing domain is fine; preserves character names. |
| Personal essay | creative | academic | 65 | Pair with a voice profile from your own writing. |
| Translated English text | balanced | (match source) | 60 | Pick domain by content; translation often reads stiff. |
If your content doesn't appear above, the heuristic is: match domain to vocabulary discipline, match tone to register, set humanness to 50 and adjust.
The humanness curve
Each band of the slider does a meaningfully different job. Knowing which band you're in matters more than the exact number.
0–25 — Very conservative rewrite. The model touches AI tells only when they're severe. Preserves "therefore," "furthermore," "moreover" if they read natural in context. Sentence structures mostly survive. Use this when the document is mostly fine and you only want to take the edge off — legal contracts, technical specs, formal correspondence.
26–50 — Standard rewrite. Removes most AI tells. Restructures the worst offenders — over-hedged sentences, uniform sentence lengths, transition-heavy paragraphs. Keeps your register intact. This is the default band for most professional writing: academic, technical, formal marketing.
51–75 — Aggressive rewrite. The model is willing to restructure paragraphs and reshape rhythm. Register can shift slightly toward conversational, especially without a voice profile pinning it. Best for content where voice matters more than precision: blog posts, essays, fiction, consumer marketing.
76–100 — Informalize meaningfully. The output reads conversational and casual. Expect to lose formal voice. Use only when you actively want casual prose — Twitter threads, friendly emails, very consumer-facing content. Above 90, the rewrite gets bold enough that you should always pair it with a voice profile.
Voice Profile + humanness interaction
A voice profile changes the math.
Without a profile, higher humanness drifts faster. The model has license to restructure prose, and "what voice should the result have" is left to defaults. By 70, you can feel the register sliding toward generic-conversational.
With a profile, the constraint flips. The profile fixes your sentence-length distribution, hedging rate, vocabulary specificity, and n-gram patterns. Humanness now moves rewrite aggressiveness without dragging voice with it. You can go higher on humanness — 70, 80, sometimes 90 — and still land in your voice rather than a generic one.
This is the main reason we recommend training a profile if you have any meaningful sample of prior writing. It widens the usable humanness range. For team workflows, a shared voice profile lets multiple writers humanize at any humanness and end up at the same brand voice.
What to do when output reads wrong
A few diagnostic patterns:
- Output reads generic, no character. Add a voice profile. This is the single fix for "well-humanized but not me."
- Output reads too informal for the context. Drop humanness by 15–20. If that doesn't help, switch tone from
balancedtoprofessionaloracademic. - Technical terms got paraphrased. Switch to the matching domain (
legal,medical,technical, etc.). If your content doesn't match any domain cleanly,technicalis the safest default for term preservation. - Formatting drifted — headings shifted, lists broke. That's not tone or humanness; it's a format-pipeline edge case, usually with unusual DOCX styling. Report it and we'll look.
Wrap
For voice profile setup, see Setting up Voice Cloning.
Start humanizing today
5 documents free a month, no card needed. Three minutes to your first humanized doc.
- 5 documents/month on the free tier
- No credit card required
- Cancel or upgrade anytime