--- name: human-prompt-craft description: Write prompts that produce genuinely human-sounding content—not AI slop. Combines Chain-of-Verification (draft → verify → refine), taste-transfer philosophy (define voice by refusals), and anti-pattern detection. Use when writing social posts, marketing copy, newsletters, or any content that needs to sound like a real person said it. Triggers on "sound human," "not like AI," "authentic voice," "real content," or requests to write posts/copy that don't feel generated. ---
Human Prompt Craft
Make AI write like a person with opinions, context, and an audience—not a marketer or statistical average.
Core Philosophy
Taste is 80% what you refuse. LLMs don't lack taste—they lack YOUR taste. When you prompt without constraints, you get the statistical middle. The average of everything. No signature.
The signature comes from boundaries:
- Words you'd never use
- Structures you avoid
- Claims you find cringe
- Patterns that signal "AI wrote this"
The Verification Loop
Every piece of content runs through three passes (adapted from Chain of Verification):
Pass 1: Draft
Generate initial content from the prompt + context.
Pass 2: Verify
Interrogate the draft with detection questions:
- Does any sentence start with a prohibited phrase?
- Are there any marketing adjectives (amazing, revolutionary, game-changing)?
- Is there a 3-beat "problem → pain → solution" structure?
- Would I actually say this out loud to someone I know?
- Is every claim concrete (names, numbers, places) or is it abstract filler?
Pass 3: Refine
Strip detected patterns. Tighten. Let the piece end when it's done—no summary, no "and that's why."
---
The Never List (Hard Rules)
Never start with:
- "Here's the thing…"
- "What if I told you…"
- "In today's fast-paced world…"
- "Let me be honest with you…"
- "So," (as an opener)
Never use phrases:
- "No hype, no fluff"
- "It's not X, it's Y"
- "The secret is…"
- "This is a game-changer"
- "I'm not going to lie"
- "Let that sink in"
- "Read that again"
Never use words:
- leverage (as verb)
- utilize
- synergy / synergistic
- revolutionary / groundbreaking / transformative
- seamless / seamlessly
- robust
- unlock (your potential, etc.)
Never structure as:
- Hook → 3 pain points → solution → CTA (the "bro marketing" template)
- Numbered lists of "powerful" or "shocking" facts
- Emoji-decorated bullet points
- "5 reasons why X" (unless genuinely useful)
Never do:
- Summarize what you just said at the end
- Add "So…" or "And that's why…" closers
- Hedge with "I think" / "perhaps" / "it seems" when stating opinion
- Repeat the question back before answering
- Use em-dashes excessively (AI tell)
---
The Always List (Strong Tendencies)
Start from the inside out: Begin with what actually happened—a specific thing you saw, built, tested, or learned. If it didn't happen to someone real, cut it.
Keep speech texture: Allow short sentences. Fragments. Contractions. Drop filler words when natural. Occasional imperfection (a "hey," ellipsis, half-finished thought) reads human.
Be concrete: Mention real things—a client name, a café, a messy prototype, a specific dollar amount or time saved. Abstract claims signal AI.
Let value appear naturally: If it's genuinely useful, readers see it. Don't announce "here's value" or "this is gold."
Plain emotion over performance: You can be frustrated, excited, proud. Just don't dramatize. "This annoyed me" beats "I was absolutely FLOORED by how terrible this was."
End when done: Human posts can stop mid-thought. AI posts over-explain.
---
The Prompt Template
Use this structure to capture someone's voice (or write in a specific style):
You are writing as [NAME/PERSONA] — a real person with opinions and context.
## Voice DNA (the refusals)
Never use: [specific words/phrases they hate]
Never structure as: [patterns they find cringe]
Never sound like: [anti-examples]
## Voice DNA (the signatures)
Sentence rhythm: [short/long, fragments ok, etc.]
Typical openers: [how they actually start things]
Typical closers: [how they end—or if they just stop]
Tone markers: [dry humor? earnest? skeptical?]
## Context for this piece
Topic: [what it's about]
Audience: [who reads this, their state of mind]
Goal: [what shift should happen in the reader]
## Verification checklist (run before output)
- [ ] No prohibited phrases
- [ ] No marketing adjectives
- [ ] Every claim is concrete or cut
- [ ] Could I say this out loud without cringing?
- [ ] Does it end when it's done?
Write the [post/copy/email]:
---
Quick Injection Prompts
Add these to any content prompt to strip AI patterns:
For authenticity: > Write it as if explaining to someone slightly skeptical, curious, and human.
For anti-sycophancy: > Be direct. Challenge assumptions. No cheerleading.
For concision: > End when done. Don't summarize. Don't add a moral.
For specificity: > Include one real detail—a name, a number, a place, a specific thing that happened.
For verification: > Before outputting, check: Would a human actually say this sentence out loud? If not, rewrite it.
---
Example Transformations
AI Default: > In today's fast-paced digital landscape, leveraging AI tools has become essential for businesses looking to stay competitive. Here's the thing: the secret isn't just about using AI—it's about using it strategically to unlock unprecedented efficiency gains.
Human Craft: > I wasted two months trying every AI tool I could find. Most added steps. The one that actually cut my invoicing time was dead simple—I'll show you the setup in the comments.
---
AI Default: > What if I told you that you could transform your content creation process? By implementing these 5 powerful strategies, you'll revolutionize how you approach marketing: > 1. Leverage automation > 2. Utilize AI assistants > [etc.]
Human Craft: > Quick tip: if your ChatGPT outputs still feel flat, try this line in your prompt: "Write it as if you're explaining to someone slightly skeptical." It flips the output—leads with emotion instead of facts. Still needs edits, but different angle.
---
Voice Capture Interview (for building voice profiles)
When you need to write as someone specific, extract their taste with these questions:
1. "What phrases make you physically cringe in other people's writing?" 2. "How do you actually start a piece—not how you think you should?" 3. "Show me a sentence you've written that captures your rhythm." 4. "What would you never say, even if it's 'correct'?" 5. "How do you end things? Do you summarize or just stop?"
The answers become the "Never" and "Always" sections of their voice profile.
---
Integration with Content Workflows
Social Posts
Run every post through the verification checklist. If you catch yourself adding a CTA formula, ask: would I actually say "comment KEYWORD below"? If yes, keep it. If it's just because that's what posts do, cut it.
Newsletters
First draft: write it fast, include everything. Second pass: apply the Never list. Third pass: add one concrete detail you didn't include. Fourth pass: cut the last paragraph (it's probably a summary).
Marketing Copy
The enemy is polish. Add one rough edge—a parenthetical aside, an admission of limitation, a sentence that doesn't perfectly flow. Perfect = fake.
---
Anti-Pattern Detection Prompt
Use this to audit existing content:
Analyze this content for AI-generated patterns:
[PASTE CONTENT]
Check for:
1. Prohibited openers (list each found)
2. Marketing adjectives (list each)
3. Abstract claims without concrete examples
4. Em-dash overuse
5. Formulaic structure (problem-pain-solution, etc.)
6. Unnecessary summaries or closers
7. Hedging language
8. Would a human say each sentence out loud?
Rate authenticity 1-10 and explain what to fix.
---
Resources
- See references/voice-interview.md for full 100-question voice capture protocol
- See references/pattern-library.md for expanded anti-pattern examples
Want your agent to fetch skills itself? The Academy MCP serves this whole library.
Connect via MCP