Voice Extraction Framework

Most people using AI for content have a voice problem. The output sounds like AI because they never defined what they sound like. They skip straight to prompting without answering the basic question: what does my writing actually sound like when it's good?

This framework fixes that. Five steps. End result is a document you can hand to any AI tool that produces output in your voice.

Step 1: The Questionnaire

Answer about yourself. Not who you want to be. Who you actually are. The gap between aspiration and reality is where AI voice falls apart.

Identity Questions:

How do friends describe the way you talk? (Not how you'd describe yourself. How do they describe you?)
What can you talk about for 2 hours straight without notes?
What opinion do you hold that most of your peers disagree with?
What's your default communication style? (Blunt? Diplomatic? Sarcastic? Dry?)

Voice Mechanics:

Do you use contractions? (Most people do in speech, then drop them in writing. Keep them.)
Do you curse? How often? Which words?
What kind of humor? (Deadpan? Self-deprecating? Dark? None?)
Do you use analogies or direct statements? (Some people explain through comparison. Others just state the thing.)

Content Boundaries:

What topics are off-limits? (Not because they're controversial, because you genuinely don't care about them.)
What topics do you always have an opinion on?

Influences:

Whose writing do you enjoy and why? Not "admire from a distance." Actually enjoy reading.
Whose writing annoys you and why? Just as important.
Read your 3 best posts/emails/messages. What do they have in common?

Step 2: Extract the Rules

From your answers, extract 8-12 rules that describe how you already communicate at your best. Not aspirational rules. Descriptive rules.

Format each one:

Rule: What it is
Why it matters: What it does for your voice
Right: Example that follows the rule
Wrong: Example that breaks it

Example:

Rule: Use contractions always. Why it matters: Dropping contractions makes writing sound formal and stiff. You don't talk like that. Right: "Don't overthink it. You won't get it perfect." Wrong: "Do not overthink it. You will not get it perfect."

Step 3: Build the Ban List

Go through AI content in your niche. Find the words everyone uses. Ban them.

Common bans that apply across niches: "delve," "tapestry," "landscape," "leverage" (as a verb), "harness," "elevate," "foster," "streamline," "cutting-edge," "game-changer."

Then add niche-specific bans. In fitness: "transform," "unlock your potential." In tech: "robust," "scalable solution," "paradigm." In business: "synergy," "ecosystem," "stakeholder alignment."

The ban list isn't about avoiding specific words forever. It's about forcing you to find your own way to say things instead of defaulting to the same vocabulary as every other AI-assisted post.

Step 4: Identify Structural Tells

Beyond words, find the structural patterns that are overused in your niche. Ban those too.

Common structural tells:

The Tricolon: Three parallel sentences. "It builds trust. It drives engagement. It creates loyalty."
The Inversion: "Not X, it's Y."
The Setup-Reveal: Building tension then a dramatic one-liner after a paragraph break.
Fortune-Cookie Compression: Perfectly clean quotable aphorisms.

Read 20 posts in your niche. Mark every structural pattern that appears more than 3 times. Those go on the ban list.

Step 5: Test and Iterate

Write 5 pieces using your rules. Read each one aloud. For each piece, answer:

Does this sound like me?
Would a friend recognize my voice in this without seeing my name?
Could any AI have written this without my voice document?

If the answer to #3 is "yes," find what's generic and fix it. The rules need updating.

First version of your voice document is never right. Expect 3-4 iterations before it feels accurate. That's normal. You're trying to codify something intuitive. It takes a few passes.

The Result

A document, usually 2-4 pages, that describes exactly how you communicate. Rules, bans, examples. Hand it to any AI tool as context and the output gets dramatically closer to your actual voice.

Not perfect. Never perfect. But close enough that editing takes 5 minutes instead of a complete rewrite. That's the difference between AI as a tool and AI as a crutch.