The Strategy Guide: The AI Ghostwriter

You’ve built engines for growth and ops. Today, we build the engine for your voice—a repeatable system that turns your ideas into on-brand drafts in minutes, without sounding like a bland content mill. This is how you create a “digital doppelgänger” that scales your writing without sacrificing authenticity.

1) Core Philosophy — Why “write in my style” fails

Big mistake: asking an AI to “sound like me” without giving it evidence, constraints, or purpose. You’ll get generic tone, recycled clichés, and off-brand claims.

Right approach: treat voice as a system, not a vibe:

• Evidence (your best writing samples)

• Constraints (what to never say/do)

• Objectives (what each piece must achieve)

• Checks (how you validate authenticity)

AI should draft; you set the rules, give the proof, and do the final pass.

2) Build the Voice Corpus — Your raw ingredients

Start with a focused, high-quality set (you don’t need everything you’ve ever written).

What to include (aim for 20–40 pieces / 15k–40k words):

• 8–12 of your best long-form posts or newsletters

• 6–10 short posts (LinkedIn/X) that performed well

• 3–5 emails you’re proud of (announcements, tough decisions, heartfelt notes)

• 2–3 transcripts (talks/podcasts) to capture your spoken cadence

What to exclude: drafts you didn’t publish, ghostwritten pieces by others, anything off-brand.

Label each sample with metadata: date, audience, purpose, topic, and why it worked (hook, story, insight). This teaches the model what to emulate.

3) The Voice Card — Your master specification (copy/paste)

Create a one-pager the model sees every time it writes for you.

VOICE_CARD:
  Who_I_Write_For:
    ICP: "Founder-operators and senior ICs building AI-powered systems"
    Promise: "Actionable playbooks with zero fluff"
  Lexicon:
    Preferred: ["playbook", "engine", "guardrail", "kill switch", "mission control"]
    Banned: ["cutting-edge", "synergy", "game-changer", "leverage (as verb)"]
    Terminology: ["RAG", "routing", "scorecard", "SLA", "risk-adjusted ROI"]
  Rhythm_&_Pacing:
    Sentences: "short-to-medium, 1 idea each"
    Structure: "hook → setup → numbered steps → examples → CTA"
    Cadence: "decisive, plainspoken; avoid hedging"
  Signature_Moves:
    - "Define a 3–5 step framework with bold labels"
    - "Show math or a back-of-napkin formula"
    - "Include a drop-in prompt or template"
    - "Add a 7–14 day rollout plan"
  Tone:
    Traits: ["direct", "practical", "slightly playful", "no fluff"]
    Empathy: "acknowledge pain, then give tools"
  Evidence_Rules:
    - "Ground claims in data/examples; no hallucinated facts"
    - "If uncertain, offer options not guesses"
  Formatting:
    - "Use H2/H3, tight bullets, code blocks for prompts"
    - "≤ 1,200 words unless specified"
  Anti_Patterns:
    - "Generic inspirational fluff"
    - "Name-dropping without value"
    - "Overlong intros; burying the lede"
  Redlines:
    - "Never promise outcomes; frame as experiments"
    - "No invented quotes or fabricated results"

4) RAG vs. Fine-Tuning — Which path and when?

RAG (Retrieval-Augmented Generation) = supply your Voice Card + recent samples at prompt time.

• Use when: you iterate often, want transparency, and have <100k words.

• Pros: flexible, easy to update, safer against drift.

• Cons: small context window limits breadth; repeated prompting overhead.

Fine-tuning = train a model on your corpus.

• Use when: you have >150k words of strong, consistent writing, your voice is stable, and you need speed at scale (many drafts/day).

• Pros: faster, cheaper per draft at scale, “feels” more native.

• Cons: locked to the training cut; needs re-trains; risk of baking in quirks or bias.

Pragmatic rule: start with RAG + Voice Card. Consider fine-tune only after 6–8 weeks of stable output and when you’re producing high volume.

5) The Ghostwriter Workflow — From idea → publish (30–45 min)

A) Brief (you, 3–5 min)
Fill a tiny brief: audience, goal (what action should the reader take), one sentence thesis, 3 proof points (story, data, example), distribution channel.

B) Draft (AI, 2–5 min)
Feed Voice Card + 2–3 relevant corpus pieces + brief into the prompt below.

Drop-in “Ghostwriter” prompt

SYSTEM: You are my ghostwriter. Follow the VOICE_CARD exactly.
CONTEXT:
- VOICE_CARD: <paste YAML>
- EXAMPLES: <paste 2–3 short, high-performing excerpts>
BRIEF:
- Audience: [who]
- Goal/CTA: [what action]
- Thesis: [one sentence]
- Proof assets: [bullets or links]
- Channel: [newsletter | LinkedIn | blog | email]
RULES:
- Start with the payoff in ≤30 words.
- Use the structure and lexicon from VOICE_CARD.
- If proof assets are thin, insert TODO markers like [ADD METRIC].
- Output Markdown. ≤[word_limit].
TASK: Write a V1 draft that is 80% complete and ready for my polish.

C) Authenticity polish (you, 5–10 min)
Run the Editor’s Checklist (below), add 1 lived detail (screenshot, metric, anecdote), and tighten the CTA.

D) Final pass (AI, 1–2 min)
Ask the model to shrink, sharpen, and de-jargon without changing meaning.

Editor’s Checklist (quick pass)

• Hook test: First 2 sentences name the pain + payoff?

• Lexicon test: Voice Card words present; banned words absent?

• Specificity test: At least 1 metric, 1 example, 1 “do this now” step?

• Cadence test: Any long, mushy paragraphs to split?

• CTA test: One explicit next step aligned to goal?

6) Personalization & Guardrails — Keep it you (and safe)

• Channel adapters: keep channel-specific mini-cards (newsletter vs LinkedIn vs email).

• Disclosure policy: decide when to disclose AI assistance (e.g., sensitive comms).

• Source hygiene: when citing stats, include links/footnotes or mark [CITE].

• Drift control: review 1 piece/week with your Voice Card open; add new phrases you actually used, prune ones you no longer say.

• Ethics: never synthesize private messages without consent; avoid mimicking others’ voices.

7) Your 60-Minute Setup — Ship today

00:00–10:00 Gather 12–15 of your best pieces; label with purpose & why they worked.
10:00–25:00 Draft your Voice Card (YAML above).
25:00–35:00 Create a 5-field Brief template (audience, goal, thesis, proof, channel).
35:00–50:00 Build a “Ghostwriter” custom GPT or snippet with the prompt + Voice Card.
50:00–60:00 Run a real brief; publish the V1 with a 5-minute polish.

8) Upgrades (week 2+)

• RAG library: store your best paragraphs/snippets; retrieve 3 per draft.

• House style linter: a simple rule check for banned phrases, passive voice, over-length sentences.

• Prompt macros: “Repurpose” macro to auto-create LinkedIn + email + 3 tweets from a cornerstone.

• A/B hooks: ask the model for 5 alternative hooks; test on social first.

9) The Quirky Bit — Have fun with your doppelgänger

• Cliché Roaster: feed your own draft and ask the model to roast your clichés and rewrite them in your voice (“replace ‘game-changer’ with a concrete outcome”).

• Storytime Mode: generate a 120-word parable in your cadence that teaches the same lesson as your post—great for intros.

• “Future Me” Emails: write a note from “you in 12 months” describing what changed because you followed this playbook—excellent for internal alignment.

TL;DR

• A great ghostwriter needs evidence (corpus), constraints (Voice Card), and a job (brief).

• Start with RAG + Voice Card; consider fine-tuning only after stable wins and >150k words.

• Use the Ghostwriter prompt to get an 80% V1 fast, then do a 5–10 minute authenticity polish.

• Evolve your Voice Card weekly to keep the doppelgänger honest—and unmistakably you.