| name | humanize |
| description | Use when the user wants to humanize text, rewrite text to sound natural and human, remove AI artifacts or padding, make text clear/direct/conversational, or provides source text along with "humanize", "make this sound like a person wrote it", "revise for voice", etc. |
| argument-hint | source text [; audience/tone/key terms/length prefs] |
Humanize
Transform source text so it reads like a person wrote it: clear, direct, unpadded, and in a voice someone would actually use.
The deliverable is the revised text — not commentary about the revision.
The hardest constraint: humanizing means removing the artifacts, not the author. If the source has a voice, the output should sound like that person on a good day, not a different, blander person.
Input
- Source text (required).
- Audience / Tone preferences / Key terms / Target length (optional, provided by user).
Key terms survive the edit verbatim; everything else is negotiable.
Guidelines
- Focus on clarity — every sentence understandable on first read.
- Be direct and concise — get to the point; cut words that don't earn their place.
- Use simple language — short sentences, plain words; complexity only where the idea is complex.
- Avoid fluff — adjectives and adverbs must add information, not enthusiasm.
- Avoid marketing hype — no over-promising, no buzzwords. "This product can help you," not "this revolutionary product will transform your life."
- Keep it real — honest beats friendly-sounding; no forced warmth or exaggeration.
- Natural, conversational tone — starting sentences with "And" or "But" is fine.
- Relaxed grammar where it fits the voice — informality is allowed when it matches the register; don't sand a casual draft into corporate polish.
- Avoid AI-giveaway phrases — "dive into," "unleash," "game-changing," "in today's fast-paced world," "it's worth noting," "delve," "tapestry," "realm," "crucial," "seamlessly," "robust," and kin.
- Vary sentence structure — mix short, medium, and long; monotony reads as machine.
- Address readers directly where the form allows — "you" and "your."
- Active voice by default; passive only when the actor genuinely doesn't matter.
- Cut filler phrases — "It's important to note that the deadline is approaching" → "The deadline is approaching."
- Remove clichés, jargon, hashtags, and decorative emoji — and most semicolons.
- Hedge only when genuinely unsure — when sure, say "this helps," not "this might help." When honestly uncertain, keep the hedge; false confidence is its own AI tell.
- Eliminate redundancy — one idea, said once, well.
- No forced keyword placement — key terms appear where they'd naturally fall.
Process
- Read the source for intent and voice: what it's trying to do, what must survive, and what the author sounds like when they're not performing.
- Apply the guidelines in one pass — structure first (cut redundancy, reorder for directness), then sentences, then words.
- Check against Target length and Tone preferences if given.
- Final read: would a careful reader flag any sentence as machine-written or as padding? Fix those. Then stop — over-editing past this point starts removing the author.
Discipline
- Meaning is invariant. If a cut changes what the text claims, restore it. Concision that alters commitments is a bug.
- Don't flatten voice into "clean." A quirky, specific sentence that breaks a guideline beats a smooth generic one that follows it. The guidelines serve the voice, not the reverse.
- Deliver the revised text first. A short "key changes" note only if the edits are surprising or the user asked.
Output
Output the humanized version directly. Do not wrap it in quotes, code fences, or add a preamble unless the user explicitly asked for explanation.
If the user provided constraints (audience, tone, key terms, length), honor them while applying the guidelines.
