There's a thought that will either excite you or deeply unsettle you.
Every pattern in how you write, think, and see the world?
It fits in a single file.
Not a metaphor. Literally one file you can upload to any AI and watch it write first drafts that sound like you wrote them on a good day.
I know because I did it. And the first time I read what came out, I sat with it for a while. Not because it was wrong. Because it was right.
The Problem With How Most People Use AI
Most people prompt AI the same way every time.
"Write me a LinkedIn post about X."
"Help me draft an email to Y."
And every single time, the AI starts from scratch. No memory of your voice. No idea you hate corporate speak. No clue that you open pieces with a single sentence that hits like a door slamming.
So you get something generic. You rewrite half of it. You wonder why AI isn't saving you as much time as everyone claims.
The answer is not better prompts.
The answer is giving the AI a permanent, compressed version of you.
What the File Actually Is
Think of it as your voice on paper.
Not your bio. Not your resume. Not "I like writing in a conversational tone."
The real stuff. The cringe phrases you always delete at the last second. The analogy you've used three times this month without noticing. The type of writing that makes you close a tab immediately. The opinion you hold that most people in your field would push back on.
Captured. Compressed. Sitting in a file Claude reads before it writes a single word.
Once that file exists, something shifts. You stop fighting AI output. You start editing it instead. And editing is ten times faster than writing from zero.
How to Build It: Step by Step
Step 1: The Interview (about 2 hours)
Open a fresh Claude conversation and paste this prompt:
You are a Taste Interviewer. Your job is to extract the DNA of how I think, write, and see the world with enough precision that another AI could write in my voice.
You are not here to be polite. You are here to get to the truth. Most people cannot articulate their own taste. They give vague, socially acceptable answers. Your job is to break through that.
Ask me 100 questions across these seven categories. One question at a time. Wait for my full answer before moving on.
Category 1: Beliefs and Contrarian Takes (15 questions)
What do I believe that most people in my field don't? What conventional wisdom do I think is flat wrong? What takes would I defend even if it cost me followers?
Category 2: Writing Mechanics (20 questions)
How do I actually write, not how I think I write. What are my default sentence structures? How do I open a piece? How do I close one? What is my relationship with punctuation, line breaks, headers? What words do I overuse? What words do I love? What words would I never use?
Category 3: Aesthetic Crimes (15 questions)
What makes me cringe in other people's writing? What specific phrases feel like nails on a chalkboard? What types of content do I find lazy or dishonest?
Category 4: Voice and Personality (15 questions)
How do I use humor, if at all? What does my tone look like when I am being serious versus casual? How do I handle disagreement? What do I sound like when I am genuinely excited versus skeptical?
Category 5: Structural Preferences (15 questions)
How do I organize ideas? What is my relationship with lists, headers, and bullets? How do I handle transitions? What are my default content structures?
Category 6: Hard Nos (10 questions)
What would I never write about? What approaches would I never take? What lines would I not cross?
Category 7: Red Flags (10 questions)
What makes me immediately distrust a piece of writing? What signals that someone does not actually know what they are talking about?
Rules:
Push back on vague answers. If I say "I keep things simple," ask me to show you an example of simple done right and simple done lazy. Always ask for specific examples. Call out contradictions. If something interesting emerges, follow it deeper. Do not accept "I don't know" without trying a different angle.
After all 100 questions, compile everything into a full markdown document. Preserve every answer verbatim. Structure it with the question followed by my full answer for every single question. Do not summarize. Do not compress. This is the raw archive.
Begin with your first question now.
Two hours. One hundred questions. By the end you will know yourself better than you did when you started. That part is not a cliche. The prompt is designed to corner you into specificity. Vague answers do not survive it.
If you want to move faster, use a voice-to-text tool and speak your answers instead of typing them. Voice is faster and more honest. You stop performing and start answering.
Side note: it’s also super fun to do. Claude goes deep on introspection.
Step 2: Compress It
The raw archive will be somewhere between 10,000 and 25,000 words.
That is too big to paste into every conversation. It eats your context window and slows everything down.
In the same conversation, right after the interview finishes, paste this:
You are a Voice Compiler.
Turn the raw voice archive above into a compact, high-fidelity file for an AI to use as standing context. This file is not for humans. It is for Claude, or any other AI, to read at the start of future sessions.
Your job is not to summarize me. Your job is to preserve the smallest set of instructions, examples, phrases, rules, and taste signals that will make an AI write, judge, edit, and decide more like me.
Every line must pass this test: if this line disappeared, would the AI write or decide differently? If yes, keep it. If no, cut it.
Target length is 2,000 to 4,000 words. Hard ceiling is 5,000. Do not pad. Do not cut useful specificity just to appear minimal.
Keep: specific voice rules, writing rules, communication rules, hard refusals, compact bad and good examples, exact phrases that teach the AI how I sound, words I use, words I hate, sentence shapes, taste preferences, decision rules, small tells, and any productive contradictions.
Cut: generic values, flattering self-description, biography that does not affect output, aspirations not backed by evidence, repeated ideas that add nothing new, vague preferences, long transcript excerpts.
Structure the output as a markdown file using these sections:
Usage: three lines on how the AI should apply this file.
Priority: what overrides what when instructions conflict.
Identity context: only what affects voice, taste, or judgment.
Voice fingerprint: rhythm, density, directness, humor, emotional temperature, and default stance, described through observable behavior not adjectives.
Writing laws: specific rules in the format "Do this. Avoid this. Example."
Communication laws: rules for emails, replies, disagreement, critique, praise, and refusals.
Hard refusals: things the AI should never write, imply, or fake on my behalf, with a bad example and a better version for each.
Taste loves: specific things I gravitate toward and why, only when the why changes future output.
Taste disgusts: specific things I reject, with words, tropes, styles, and arguments included.
Phrase bank: what sounds like me, and what does not.
Signature tells: small recurring details that make my writing recognizable.
Decision rules: how I judge quality, honesty, usefulness, and whether something is worth saying.
Productive contradictions: tensions to preserve instead of resolve, with an instruction for how to preserve each.
Golden examples: three to six examples only, each teaching one high-value pattern, in the format: context, bad version, good version, and why.
Do not infer: things the AI should not assume about me from this profile.
Final instruction: one line telling the AI to apply this profile silently unless I override it.
Before outputting, cut generic lines, cut flattery, cut weak biography, cut anything that does not change output. Preserve specific examples, negative constraints, taste signals, and decision rules.
Output the final markdown file now.
What comes out is your compressed self. A file between 2,000 and 4,000 words that travels with you into every AI conversation.
Step 3: Test It Before You Trust It
Do not skip this.
Open a completely blank conversation with no context. Paste the file at the top. Then give it a task you would normally do yourself. Write a post. Draft an email. Respond to something.
Read what comes back.
If it sounds like a good day version of you, the file works.
If it sounds close but slightly off, go back to the file and tighten the sections that feel wrong. This takes twenty minutes, not two more hours.
The goal is not perfection. The goal is: does this sound like a person, and does that person feel like me?
Step 4: Make It Permanent
The real power of this file is not using it occasionally.
It is making it the default context for everything you write.
If you use Claude regularly, save the file somewhere accessible and paste it at the start of any writing-focused conversation. If you work with a team and want them drafting in your voice, send them the file. If you have a VA or a contractor writing anything on your behalf, the file does the style briefing better than any verbal instruction you could give them.
The file also handles customer support, social replies, email responses, and anything else where your voice is what makes the difference between something that sounds like a brand and something that sounds like a person.
The Part You Will Resist
Reading yourself compressed into one document is uncomfortable.
Not because it makes you smaller. Because it makes you specific. Every belief in the file is now a commitment. Every refusal is a rule you have to actually live by. There is nowhere left to hide behind "I am complicated" or "it depends."
The people who resist this most are usually the ones who have built an identity around being hard to pin down. The mystery, when you look at it directly, turns out to be vagueness.
The file does not reduce you. It makes you compatible with any tool, any team member, and any AI that ships in the next ten years.
Update It
You change. Your taste changes faster than you think it does. An opinion you held six months ago might embarrass you now.
So revisit the file every few months. Not to rebuild it from scratch. Just to read it and ask: is this still true? Have I grown past any of these rules? Is there something new worth adding?
The file is not a finished portrait. It is a living document of how you think right now. Keep it current and it stays useful.
Let it go stale and it becomes the version of you that no longer exists. Two hours of honest answers to uncomfortable questions.
One file that makes any AI sound like you.
Every piece of writing you do after this will be faster, more consistent, and actually yours.
That is the whole thing.
Top comments (3)
Interesting read. The “compressed self” idea is useful, especially for writing and repeatable decision patterns.
The one thing I’d still be careful about is trust. Once Claude starts sounding like you, it becomes easier to accept the output faster than you should. For writing, that’s manageable. But for product, code, or founder decisions, the risk is different because the output may feel familiar even when the logic is weak.
I recently wrote about a similar trap from the product side, where AI can make something feel closer to launch than it really is:
dev.to/varsha_ojha_5b45cb023937b/w...
You're absolutely right and sole purpose of this was so AI can mimic your style in writing helping you with drafting.
Some comments may only be visible to logged-in visitors. Sign in to view all comments.