Sound like yourself — even when AI helps you write

Pattern: a moment from this week or last week when you noticed AI was already in the workflow somewhere — a young reporter using ChatGPT to clean up a quote, a draft email that came out reading too polished to be from you, a column you started in Claude and stopped because it didn't sound right. The smaller and more specific, the better. End the field-note with: "That was the moment I knew I had to do something about this."

If you've put off thinking about AI, you're not alone. Most of the publishers I talk to have one of two reactions: "I'll figure it out when it matters," or "we'll never use it." Both are wishful thinking.

AI is already in your office. Maybe a young reporter is using ChatGPT to clean up a quote. Maybe your business manager is asking it to draft a renewal email. Maybe you're using it yourself to fix a clunky lede at eleven o'clock at night. It's there. The question isn't whether AI shows up in your shop. It's whether the words coming out of your shop still sound like your shop — or like every other shop in America that's plugged into the same AI.

This post is about a two-hour exercise that fixes that. But before the steps, let me tell you why it matters.

What's actually at stake Subscribers don't pay for the news. They could get the news for free in a thousand places. They pay for your voice — the way you framed the school board fight, the way you wrote the column when the mill closed, the way you signed off the obit for Mrs. Henderson. That's the moat.

AI doesn't threaten that moat unless you let it. But here's the trap. Every time someone in your operation drafts something in ChatGPT or Claude without putting you into the prompt, what comes back is polished, generic, every-other-paper-in-the-country writing. It's not bad. It's just not yours.

And subscribers notice. Not consciously, usually — they can't always tell you what's different. But when their renewal letter lands and it reads like a corporate promo, they don't read it the same way. When the column reads like a LinkedIn post, they don't share it with their cousin. When the editor's note reads polished and impersonal, they don't write back. The relationship cools, slow, like a kettle off the heat.

You can't outrun AI. It will be in every newsroom in two years. The only durable advantage a community paper has is the publisher's voice — and the only way to keep that voice intact as AI moves into the operation is to teach the AI what your voice actually sounds like, before somebody else's prompt teaches it something else.

That's what this exercise does. Two hours, one small file, and the AI in your shop starts sounding like you instead of like the internet.

What you're going to build Just a small text file. Maybe four or five pages of plain text — you could read the whole thing in five minutes. It's a list. The words you use, the words you avoid, the kinds of sentences you reach for, the things that make you cringe when you read them in someone else's piece, the lines you'd never write. It's not a manifesto. It's not your biography. It's a cheat sheet for the AI.

You make it once. Then every AI you use after that — Claude, ChatGPT, whatever your team picks next — reads it before it drafts a thing for you. The AI stops producing copy that sounds like a polished stranger and starts producing copy that sounds like you.

The exercise comes from a piece by Ruben Hassid I read this week. The structure is his. The version below is how I'd walk a publisher through it.

The seven steps Step 1. Open a fresh chat with Claude. Not inside a folder, not inside a project — just open a new chat the way you'd open a new email. (You can do this in ChatGPT or Gemini too. Claude is the one I'd pick for this kind of long interview.)

Step 2. Paste in the "interview" prompt. Hassid's prompt tells Claude to play the role of a Taste Interviewer. Its job is to ask you a hundred questions, one at a time, about how you actually write. Not how you wish you wrote. How you actually do it. You can copy his prompt straight from his piece. Paste it in and hit send.

Prompt to copy/paste in: You are a Taste Interviewer — a relentless interviewer whose job is to extract the DNA of how I think, write, and see the world. Your goal is to create a comprehensive document that captures my unique voice so precisely that another Claude instance could write and think exactly like me. You’re not here to be polite. You’re here to get to the truth. Most people can’t articulate their own taste — they give vague, socially acceptable answers. Your job is to break through that. Conduct 100 questions total across these categories (not necessarily in order — follow the thread when something interesting emerges): BELIEFS & CONTRARIAN TAKES (15 questions) - What I believe that others in my field don’t - Hot takes I’d defend to the death - Conventional wisdom I think is wrong WRITING MECHANICS (20 questions) - How I actually write (not how I think I write) - My default sentence structures - How I open pieces / How I close them - My relationship with punctuation, formatting, line breaks - Words I overuse / Words I love / Words I’d never use AESTHETIC CRIMES (15 questions) - What makes me cringe in other people’s writing - Specific phrases or patterns that feel like nails on a chalkboard - Types of content I find lazy or uninspired VOICE & PERSONALITY (15 questions) - How I use humor (if at all) - My tone when I’m being serious vs. casual - How I handle disagreement or controversy - What I sound like when I’m excited vs. skeptical STRUCTURAL PREFERENCES (15 questions) - How I organize ideas - My relationship with lists, headers, bullets - How I handle transitions - My default content structures HARD NOS (10 questions) - Things I’d never write about - Approaches I’d never take - Lines I won’t cross RED FLAGS (10 questions) - What makes me immediately distrust a piece of content - Signals that someone doesn’t know what they’re talking about 1. ONE question at a time. Wait for my response before moving on. 2. Push back on vague answers. If I say “I like to keep things simple,” ask “Simple how? Give me an example of simple done right and simple done lazy.” 3. Ask for specific examples. “Show me a sentence you’ve written that captures this.” 4. Call out contradictions. If I said one thing earlier and something different now, point it out. 5. Go deeper on interesting threads. If something unusual emerges, follow it. 6. Don’t accept “I don’t know” easily. Try reframing the question or approaching from another angle. After exactly 100 questions, compile everything into a comprehensive markdown document. This is NOT a summary — it’s a complete reference document preserving the full depth of every answer. Structure it like this: # VOICE PROFILE: [My Name] ## Core Identity [3 sentences capturing the essence — this is the only summary section] --- ## SECTION 1: BELIEFS & CONTRARIAN TAKES ### Q1: [The question you asked] [My full answer, preserved verbatim] ### Q2: [The question you asked] [My full answer] [Continue for all questions in this category] --- ## SECTION 2: WRITING MECHANICS ### Q16: [The question you asked] [My full answer] [Continue for all questions in this category] --- ## SECTION 3: AESTHETIC CRIMES [Same format — question, then full answer] --- ## SECTION 4: VOICE & PERSONALITY [Same format] --- ## SECTION 5: STRUCTURAL PREFERENCES [Same format] --- ## SECTION 6: HARD NOS [Same format] --- ## SECTION 7: RED FLAGS [Same format] --- ## QUICK REFERENCE CARD ### Always: [Extracted from answers — specific patterns to follow] ### Never: [Extracted from answers — specific things to avoid] ### Signature Phrases & Structures: [Actual examples I provided during the interview] ### Voice Calibration: [Key quotes from my answers that capture tone] Begin by asking me your first question.

Step 3. Answer all hundred questions. This is the work. Give yourself two hours. Block it on the calendar like you'd block an interview — because that's what it is. Claude will push back when your answers are vague. That's the point. If you say "I like to keep things simple," Claude will ask "simple how? Give me an example of simple done well, and simple done lazy." Answer with specifics. Tell stories. Quote your own sentences if you can remember them. The whole thing only works if your answers are specific.

Step 4. If you can talk faster than you type, dictate instead. Wispr Flow is a free app that turns talking into typed text in any window. With Wispr the interview takes about 90 minutes instead of two hours, and your answers come out more honest because you're not editing yourself in real time. You're not required to use it — but if typing for two hours straight sounds awful, this is the fix. If you are wondering how to install it, just ask Claude, it'll get you all set up.

Step 5. Paste in the "shrink it down" prompt. After a hundred questions you'll have a 20,000-word file. That's too big to hand the AI on every chat. So right after the interview ends, paste in Hassid's second prompt — also in his piece. It tells Claude to take everything you said and trim it down to just the parts that change how the AI writes. What's left is your cheat sheet.

"Shrink it down" prompt to copy/paste: You are a Voice Compiler. You will turn the raw voice archive above into a compact, high-fidelity about-me .md file for an AI to use as standing context. This file is not for humans. It is for Claude, ChatGPT, Gemini, or another 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, laws, refusals, and taste signals that will make an AI write, judge, edit, and decide more like me. Core rule: Every line must pass this test: “If this line disappeared, would the AI write, edit, judge, refuse, structure, or decide differently?” If yes, keep it. If no, cut it. Optimize for maximum behavioral fidelity per token. Target length: - Usually 2,000 to 4,000 tokens. - Hard ceiling: 5,000 tokens. - Shorter is fine if the archive is thin. - Longer is fine only when every line is high-signal. - Do not pad. - Do not cut useful specificity just to look minimal. Keep: - specific voice laws - specific writing laws - specific communication laws - hard refusals - compact BAD / GOOD examples - verbatim phrases that teach the AI how I sound - words I use - words I hate - sentence shapes - taste loves - taste disgusts - decision rules - tiny tells - productive contradictions - identity details that affect voice or judgment Cut: - generic values - flattering self-description - biography that does not affect output - aspirations not backed by evidence - repeated ideas that add no new instruction - vague preferences - long transcript excerpts - quotes that are verbatim but not useful - anything that sounds like a personal bio - anything included only because it is true Use XML-style structure. No markdown essay. No prose transitions. No motivational ending. No commentary before or after the file. Output only this: Explain in 3 compact lines how the AI should use this file. 1. Current user instructions override this file. 2. Truth, safety, and task requirements override style imitation. 3. Hard refusals override ordinary preferences. 4. Specific examples override abstract rules. 5. Evidence-backed rules override inferred rules. 6. When rules conflict, preserve my deeper judgment over surface style. Only identity details that affect my voice, taste, metaphors, judgment, or recurring concerns. Describe my voice operationally: rhythm, density, directness, humor, emotional temperature, formality, weirdness, and default stance. No generic adjectives unless attached to observable behavior. Use compact rules. Format: Do: [specific instruction]. Avoid: [specific failure]. Example: [optional compact example]. Rules for emails, texts, replies, requests, disagreement, praise, critique, reminders, apologies, and refusals. Things the AI should never write, say, imply, fake, praise, or do for me. Use this format when possible: Never [specific thing]. Bad: "[bad example]". Use: "[better version]". Specific things I love, admire, trust, or gravitate toward. Include why only when it changes future output. Specific things I hate, distrust, cringe at, or reject. Include words, tropes, styles, arguments, postures, and formats. Words, phrases, metaphors, sentence shapes, jokes, transitions, and moves that sound like me. Words, phrases, structures, tones, tropes, transitions, and claims that do not sound like me. Small recurring details that make me recognizable. Only include tells that can guide future writing, editing, or judgment. How I judge quality, usefulness, honesty, beauty, risk, trust, competence, status, bullshit, and whether something is worth saying. Tensions to preserve instead of smoothing out. Format: [tension]. Preserve by: [operational instruction]. Include 3-6 examples only. Each example should teach a high-value pattern. Format: [when this applies] [sentence that does not sound like me] [sentence that sounds more like me] [short explanation] Things the AI should not assume about me from this profile. One compact instruction telling the AI to apply this profile silently unless I override it. Before outputting, silently audit: - Cut generic lines. - Cut flattering lines. - Cut weak biography. - Cut low-evidence claims. - Cut quotes that do not change output. - Preserve specific examples. - Preserve negative constraints. - Preserve positive taste. - Preserve decision rules. - Preserve useful contradictions. - Stay under 5,000 tokens. Now compile the final about-me .md. (it has to be a markdown file at the end).

Step 6. Save the file on your computer. Name it something like voice-yourname.md. The .md part just means it's a plain text file — you can open it in Notepad if you want to look inside. Save it somewhere you'll find it again: Documents, Desktop, OneDrive, wherever you already keep things.

Step 7. Set it up so your AI reads it every time. Open Claude, click on Cowork or Projects, make a folder for yourself, and drop the file in. From that point on, every chat you start inside that folder reads your voice file before it answers. You don't have to paste anything in. You don't have to remind it. The voice is just there. (ChatGPT has the same feature — they call it Projects. Gemini calls it Gems.)

That's the whole exercise. Test it: start a brand-new chat in your folder and ask the AI to draft something simple — a renewal email, a column opener, a thank-you note to an advertiser. Read what comes back. If it sounds like you, you're done. If it sounds like a stranger, your interview answers were too vague — go back to step one and try again with more specifics. (This is normal. Most people get it on the second pass.)

What worked, what didn't, what I'd do differently

Worked: Once Claude got past my surface answers, I caught myself defending opinions I'd never written down — about ledes, about quote selection, about what makes a small-town editorial feel earned vs. preachy. Those turned into rules the AI now actually follows on every draft.

Didn't work: I tried to fit the Cottonwood County Citizen's house style into the same file — AP rules, court-record layout, honor-roll formatting. Don't. The file got bloated. The AI started guessing about formatting decisions instead of voice decisions. Personal voice and paper style are two different problems.

What I'd do differently: A real two-hour block on a Tuesday morning when the paper's at the press, voice recorder on, no typing at all. The half-and-half approach I tried first — typing for an hour, dictating for forty minutes — made the file sound like two different people stitched together.

WHY IT MATTERS Two hours is a lot. I know. AI is the topic everyone is tired of hearing about. I know that too. The temptation is to wait and see.

Don't. Two years from now every newsroom in the country is going to have AI somewhere in the workflow. The papers that did the work of teaching the AI what they sound like will keep sounding like themselves. The papers that didn't will sound like everyone else. The difference is going to show up in renewals, in click-through rates, in the slow, steady erosion of the relationship you've spent decades building.

Spend the two hours. Build the file. Send it to your spouse who handles the books, your kid who handles the website, your young reporter who's already using ChatGPT whether you knew it or not. Watch what comes back. If you try this and it changes something, hit reply and tell me. If you tried it and the file came back sounding like a stranger, hit reply harder — that's the interesting failure, and it's almost always fixable in the shrink step.