My best advice for training AI on your voice has always been: find five great examples of your writing, ask Claude to analyze what makes them yours, and turn that analysis into a Voice DNA document you use as a reference.
It works. When you have the right examples, it works well. I wrote about the whole process here: How to Train ChatGPT to Sound Like You.
But here’s the problem I keep running into with clients and readers: most people don’t have five great examples. They have some blog posts they’re not proud of, a few LinkedIn updates that feel generic in hindsight, and maybe one piece that came close — but even that one was edited so heavily by AI that it’s no longer a clean evidence of how they actually write. The samples exist, but they’re inconsistent, they’re contaminated, or they can’t explain what makes the good ones good.
There’s another method. Instead of showing Claude your writing, you let Claude interview you.
What the Interview Method Does Differently
Instead of showing Claude your outputs, you show it how you think. You answer questions about your beliefs, your instincts, your professional experience, what makes you cringe, and what you’d defend in an argument. The Voice DNA that comes out the other end isn’t just a style guide, but a thinking guide.
This works even if you’ve never published a word. It works even if half your existing writing was AI-assisted.
Step 1: The Interview (30–50 minutes)
Open a new conversation in Claude. Paste the prompt below and answer each question as honestly as you can.
The quality of your Voice DNA depends entirely on the specificity of your answers. Vague answers produce generic output. Claude will push back if you’re being too abstract, but you’ll save time if you start specific.
Here’s the difference in practice:
- Vague: “I write in a conversational tone.”
- Specific: “I write like I’m explaining something to a smart colleague who will stop reading if I waste their time with setup they don’t need.”
- Vague: “I don’t like jargon.”
- Specific: “I never use ‘leverage’ as a verb. I never say ‘in the ever evolving landscape of.’ I’d rather say something twice in plain English than once with a buzzword.”
The Interview Prompt
Copy and paste this into Claude:
You are conducting a Voice DNA interview. Your job is to help me articulate my writing voice, my thinking style, and my professional perspective so precisely that you could write and think like me.
Ask me the questions below, one category at a time. After I answer all of them, compile everything into a Voice DNA document — not a summary, but a complete reference that preserves the depth of my answers.
Here are the questions:
Part 1: Your Worldview
- What’s the most dangerous lie your industry is telling your audience right now?
- What do most tools or approaches in your field actually do well — and where do they consistently fail?
- What’s something you believed two years ago that you’ve since changed your mind about?
- Finish this sentence: “AI will never replace professionals who…”
Part 2: How You Think and Work
- Walk me through how you actually work. Not the ideal version — the real one.
- What’s a mistake you made early on that you’d warn others about?
- When you’re evaluating whether a piece of content is good enough to publish, what are you actually looking for?
- What does “sounding like yourself” mean to you — and why does it matter professionally, not just aesthetically?
Part 3: Your Writing Voice
- What are three words you’d use to describe how you write — and three words that would be dead wrong?
- What topics make you genuinely opinionated? Where do you have a point of view you’d defend in an argument?
- Is there something you’ve written that you think is the most “you”? What makes it that?
- What do you never do in your writing, even when others do it constantly?
Part 4: Your Audience
- Who exactly is your reader? Not a demographic — describe a specific person. What are they anxious about? What pressure are they under?
- What does your reader get wrong most often? What’s the misbelief you’re constantly correcting?
- What does your reader need to hear but resist hearing
- When a piece lands well, what’s the reaction you’re getting?
Part 5: The Deeper Stuff
- What professional experience shaped how you think about your work the most? Not the résumé version — the real lesson.
- What makes you different from everyone else writing about your topic? Be specific and don’t be modest.
- If someone read everything you’ve published and had to describe your core argument — the one you keep coming back to — what would they say?
- What would make you cringe if an AI wrote it in your name?
Once I’ve answered all the questions, compile the answers into a Voice DNA document with the following sections:
– Core Identity
– Worldview & Beliefs
– Writing Voice
– The Cringe List
– Quick Reference Card with Always / Never / Voice Calibration Quotes
Step 2: Build Your Voice DNA
Once you’ve answered all 20 questions, ask Claude to compile the document. The prompt is already embedded in the interview instructions above, but here’s what you’re looking for in the output.
Core Identity — two or three sentences that capture the essence of who you are professionally and what you’re trying to do for your reader. Not a bio. A positioning statement written from the inside.
Worldview & Beliefs — the positions you hold that shape how you see your industry. This is what makes your content feel like it has a point of view rather than information.
Writing Voice — how you actually write: sentence rhythm, paragraph length, what you lead with, how you handle transitions. The mechanics, not the aspiration.
The Cringe List — your hard nos. Phrases you’d never use. Framings you find lazy. Tones that are beneath you. This is the most important section. Voice is mostly refusals.
Your Reader — who they are, what they’re anxious about, what they believe that’s wrong, what they resist hearing. The more precise this is, the more precisely Claude can calibrate its output.
Quick Reference Card — a condensed always/never list and a handful of your own sentences that Claude can use as rhythm anchors.
Save the finished document as a .md file. That’s a plain text file with simple formatting — Claude writes it naturally, and any text editor can open it. If you’re not sure how to save it as .md, just ask Claude to explain.
Step 3: Put It to Work
Upload the .md file to a Claude Project. A Project is a persistent workspace in Claude where you can store reference documents that stay available across every conversation — you don’t have to paste your Voice DNA into every prompt.
Once it’s there, the prompt pattern is simple:
Read [your-voice-dna.md] first. Then [your task].
That’s it. Claude reads the file, absorbs your cringe list, your beliefs, your reader description, and your rhythm anchors — and defaults to those instead of the statistical middle.
What changes immediately: Claude stops producing content that could have been written by anyone. It starts making the refusals you’d make. It frames problems the way you frame them. It doesn’t use the phrases you’d never use.
What still requires your judgment: the Voice DNA handles style. It doesn’t handle strategy. The argument, the angle, the decision about what’s worth saying — that’s still your job. The file makes Claude sound like you. It doesn’t make Claude think for you.
Frequently Asked Questions
What if I've never published anything? Can I still do this?
Yes. The interview is about how you think, not what you’ve written. Some of the most precise Voice DNA documents come from people who haven’t published much — they haven’t had time to develop any bad habits yet.
How long does the interview take?
30–50 minutes if you answer honestly. Longer if you’re vague — Claude will push back and ask you to be more specific. That’s the process working correctly, not a problem.
Can I use this with ChatGPT or Gemini instead of Claude?
Yes. The .md file is portable. Upload it to any LLM that accepts file uploads, or paste it into a system prompt. The method works everywhere. I recommend starting with Claude because it handles nuance and long-context documents particularly well — but the Voice DNA you build here will work across all of them.
How is this different from just writing a detailed system prompt?
A system prompt is instructions. A Voice DNA is evidence. One tells Claude what to do. The other shows Claude who you are. Instructions produce compliance. Evidence produces something closer to understanding — and the output reads like it.
