Skip to content

Building a Voice System

Tell an AI to "write in my voice" and you'll get a corporate blog post. Every time. The instruction is too vague to produce anything specific, because "your voice" isn't a single thing -- it's dozens of interlocking patterns, preferences, and instincts that you've never had to articulate. You just write, and it comes out sounding like you.

AI doesn't have that luxury. It needs the patterns spelled out.

The Problem

Default AI prose is competent and forgettable. It hedges where you'd commit. It stacks three examples where you'd use one. It opens with "In the rapidly evolving landscape of..." where you'd open with a specific moment. And the longer the project runs -- across chapters, across weeks -- the more generic it gets.

Inline instructions don't fix this. "Be conversational" means nothing to a model. "Sound like me" is a no-op. Even pasting in a sample of your writing produces diminishing returns after a few paragraphs, because the model has no framework for why your writing works -- just a surface-level pattern to mimic.

The fix: encode your voice as structured reference material. Not one file. A system of files that cover different dimensions of how you write, who you write for, and what you refuse to sound like.

The 6-File System

Six files, each with a distinct job. Together they give AI enough constraint to produce output that sounds like you on a good day.

1. Gold Standard Reference

One piece of YOUR best writing. Short -- 400 words or less. Analyzed sentence by sentence for specific techniques: opening moves, metaphor usage, sentence rhythm, vulnerability balance, filler word count (target: zero). This becomes the benchmark every output gets measured against. Not "write well." Write like this, specifically. See The Gold Standard Method.

2. Author Voice Guide

Your voice DNA. The non-negotiable elements that make your writing yours. For this book, that meant: numbered frameworks for teaching, balanced skepticism (every opportunity gets its risks), pop culture bridges that make abstract concepts visceral, honest complexity (admitting when things are hard), and a practitioner perspective (you've built things, not just studied them). This file also calibrates formality, density, and authority -- with "too casual / sweet spot / too formal" spectrums that draw clear boundaries.

3. Blog-to-Book Adaptation

If you have existing writing, you have a voice. But blog voice and book voice aren't identical. This file shows before/after transformations -- how to go deeper without going wordier. Real examples from your blog posts adapted to book format, with the pattern made explicit: keep the voice DNA, remove blog artifacts (emojis, CTAs, padding), deepen the treatment without losing density.

4. Audience Empathy Guide

Two reader personas with internal monologues. Not marketing personas -- real people with specific doubts, time constraints, and trust triggers. For this book: Sarah, a skeptical enterprise COO who's survived three "transformation" initiatives, and Marcus, a Series A founder with 18-month runway. Each persona includes what builds trust ("acknowledge constraints," "share honest failures") and what loses it ("Silicon Valley examples only," "assuming unlimited budget"). Every section needs to satisfy both.

5. Authenticity Markers

The phrase-level details. Specific openings to use ("Here's what most people get wrong about...") and specific phrases to kill ("It's important to note that..."). Structural signatures -- the numbered framework, the balanced take, the reality check, the analogy bridge. Before/after transformations showing generic-to-specific voice shifts. This is the file the AI references most during writing.

6. Quick Reference

One-page checklist for moment-to-moment decisions. The density test (read it aloud, would you skip anything, can you cut 20%). Voice non-negotiables. Instant kills. Do/don't. The Saurav Test. This file exists because the other five are thorough -- and sometimes you need the answer in 10 seconds, not 10 minutes.

How They Work Together

These files aren't standalone documents. They compose into the master system prompt that loads into every AI writing session. The system prompt references all six, so every interaction -- writing, editing, reviewing -- carries this context.

The layering matters. The Gold Standard sets the density bar. The Voice Guide defines what makes you you. The Blog-to-Book Adaptation prevents the AI from regressing to a lighter blog tone. The Audience Empathy Guide keeps the reader in frame. The Authenticity Markers provide granular phrase-level steering. The Quick Reference catches drift in real-time.

No single file does the job. The system works because each file constrains a different dimension of the output.

The Investment

Building this system properly takes about 2 days. Finding your gold standard piece. Analyzing it for techniques. Articulating your voice DNA. Creating audience personas with real internal monologues. Building phrase lists from your existing writing. Drafting the quick reference.

That investment pays off by chapter 2. Without it, you spend more time editing AI output than you saved by using AI in the first place. With it, first drafts arrive close enough to your voice that editing becomes about craft and argument, not about stripping out corporate-speak.

Here's the honest math: if you're writing a 12-chapter book and each chapter takes 4-8 hours of editing without a voice system, versus 2-4 hours with one, the system saves 24-48 hours across the manuscript. The 2-day investment isn't even close.

Next Steps