I keep catching myself typing the same instructions to Claude over and over. “Write shorter sentences.” “Don’t use that word.” “Match my tone.” Every new chat, same dance. It’s exhausting and it kills any flow I had going.
Then I came across this brilliant post from a LinkedIn creator who cracked the whole problem with a clever two-prompt trick. The mind behind it built a method that turns YOUR writing voice into a portable text file any AI can read. Drop the file in, and the model suddenly sounds like you. No more re-explaining. No more drift across chats.
I was floored when I saw how simple the system actually is. Three steps. That’s the whole thing.
The Three-Step Voice Extraction System
- Run a first prompt that asks you 100 questions about how you write
- Run a second prompt that compresses all your answers into a single .md file
- Drop that .md file into any AI tool and it writes exactly like you
That’s it. The file IS the prompt. Do the work once, every AI remembers forever.
Why Each Step Actually Matters
Let me break down the rationale behind each piece, because the order is doing real work here.
Step 1 forces you to articulate things you’ve never said out loud. Most people can’t tell you why they prefer “build” over “create” or why a certain transition feels off. The 100-question prompt drags those instincts into the open. Your taste, your hard nos, the 2 words you always type and delete. All of it gets surfaced because someone is asking the right questions.
Step 2 turns chaos into a usable artifact. Raw answers are messy and contradictory. The compression prompt distills them into a clean markdown file the AI can actually parse. Markdown matters here because it’s plain text, portable, and editable anywhere you go.
Step 3 is the payoff. Drop that one file at the start of any conversation, and the model has your full style guide loaded before you type a single word. Switch tools, switch models, the file travels with you.
What Goes Into The Voice File
The original poster pointed out the kinds of details this process surfaces. Some examples worth noting:
- Your taste preferences (sentence rhythm, paragraph length, punctuation quirks)
- Your hard nos (banned words, phrases you’d never say, formatting you hate)
- The two words you always type and delete (everyone has them)
- Your storytelling patterns (how you open, how you close)
- Your humor calibration (dry, warm, none)
It’s basically a personal brand manual, except you wrote it for yourself in two hours instead of paying an agency twenty grand.
Why This Beats Generic System Prompts
You might be thinking, “Couldn’t I just write a paragraph about my style and reuse it?” Sure. But the savvy professional behind this method had a sharper insight: a single paragraph forgets the weird stuff. The 100-question approach catches the things you didn’t know you cared about until someone asked you directly.
I think this is a serious upgrade because it scales. The same .md file works in Claude, ChatGPT, Gemini, whatever comes next. You’re not locked into one platform’s memory feature. You own the file.
How To Apply This Right Now
If you want to try the system today, here’s how to approach it:
- Block off 2 hours of quiet time. The 100 questions deserve real answers, not lazy ones.
- Run the first prompt and answer everything honestly. Don’t perform. The AI is not your boss.
- Run the second prompt to compress your answers. Read the output and edit anything that doesn’t sound like you.
- Save the file somewhere you’ll find it (Drive, Notion, Obsidian, wherever).
- Paste it at the start of every new AI conversation that involves writing.
The whole point: your voice is the moat, not the prompt. Once it’s captured in a file, you can edit it like a Google Doc forever. Add new preferences, kill old ones, evolve as your style evolves.
One Reaction From Me
I’ve been watching people try to “personalize” AI for two years now, and most attempts fizzle. Either the prompt is too short (no real signal) or too long (the model ignores half of it). This method splits the difference. Long enough to capture nuance, short enough to fit in any context window without choking the model.
The contributor who shared this hit on something a lot of pros miss: AI doesn’t need to be smarter, it needs to know you better. And the cheapest way to teach it is a text file you fully control.
Check out the original LinkedIn post for the full walkthrough and the exact wording of both prompts. Two hours of work for a voice file you’ll use every single day.