You want AI writing that sounds like you. So you give it a pile of old posts. The result still sounds bland and safe.

Why?! Old posts show finished words, not the decisions behind them. They show your vocabulary and rhythm, but they often hide what you reject, what you believe, and which trade-offs guide your choices. A better voice onboarding process asks direct questions about your opinions, technical positioning, and reasons, then adds those answers to the voice profile.

✦ Key Takeaways
  • Old posts show the final result of your thinking, but hide the trade-offs and rejected ideas that make your voice distinct.
  • A complete voice profile needs four layers: vocabulary, rhythm, contrarian positions, and architectural beliefs.
  • An interview can capture how you think, what you believe, who you talk to, and how you sound, including information old posts may leave out.

Old posts teach the surface, not the decision

When you give AI your old writing, you show it the final sentence. You don't show it the three other sentences you deleted. You don't show it why you chose one argument over another. A post is finished output. It contains your favorite words and your usual sentence length, but it hides the choices behind them.

I believe founders are often terrified to say what they actually think. They want the sale without the risk of being disliked. So they choose sales over social friction. That makes public posts polished and safe. If you train AI only on those safe samples, it learns to be safe too. It misses the strong opinion that makes a founder's voice worth hearing.

Here's an illustrative example. An AI model can learn that a founder uses clear, short sentences. But without the founder's internal logic, it won't know which product choices that founder would attack or defend. It copies the style and misses the point of view.

Use four layers to define a real founder voice

Generic writing doesn't fix itself. You need a way to check whether your AI profile captures how you write and how you think. I use four layers for this.

  • 1. Vocabulary
  • Which words and terms should you use or avoid? Give AI a basic language guide. For example, use direct product words such as "user," "problem," and "trade-off." Avoid soft business phrases such as "unlock value" when a simple phrase says the same thing.
  • 2. Rhythm
  • How should the writing move from one idea to the next? This sets the pace. Lead with the point. Use a short statement. Then explain the reason.
  • 3. Contrarian positions
  • What common advice do you reject, and what do you believe instead? This gives AI a real position instead of another safe answer. For example, reject the idea that more features always make a product stronger. Prefer solving one clear user problem well.
  • 4. Architectural beliefs
  • What deeper rules guide your decisions across different topics? I believe you shouldn't hide a real opinion just to protect a sale from social friction. That rule helps AI carry your decision logic into topics that don't appear in your old posts.

Test style against judgment

You can see the difference when you change the information behind a draft. Here is an illustrative, fictional example about whether a startup should add a new feature.

Draft A (Surface Style Only):

Adding more features can help startups meet customer needs, improve retention, and stay competitive. Teams should review user feedback and prioritize features carefully.

Draft B (Style + Judgment):

More features will not rescue a product that has not solved one clear problem. I would rather make one useful workflow excellent than add five options nobody needs.

The topic and length are similar. The difference is the judgment. Draft B makes a clear choice and shows a trade-off. It also shows what it rejects.

Style makes writing look consistent. Judgment makes it sound like a specific person.

Want to test your own drafts?! Remove your name from the text. Then ask if the writing still contains a choice you recognize, or if it is just clean language anyone could have written.

Why an interview finds what posts leave out

Waiting for your beliefs to appear in old posts misses too much. A direct interview asks for your thinking, beliefs, audience, and sound instead.

RollKind starts with one interview to learn your voice. It learns how you think, what you believe, who you talk to, and how you sound. You still own the judgment behind what it says.

These inputs map to the four layers. The sound part supports vocabulary and rhythm. Thinking and beliefs support positions and decision rules. The founder still decides what they believe, checks whether a draft reflects that belief, and chooses the final direction.

Run the profile on a new question

The best test for a voice profile is a topic you haven't written about before. Ask AI to answer it using your four layers. Require one clear position and one reason.

When you review the draft, check four things:

  1. Does it use your natural words?
  2. Does the rhythm feel familiar?
  3. Does it make a clear choice?
  4. Does the reason match your decision rules?

If the draft feels safe and generic, don't just add more old posts. That's the lazy fix. Add the missing information instead: what you reject, what you prefer, and why.

This builds a voice brief that teaches AI how you write and, more importantly, how you decide.

Bottom Line: AI sounds generic when it lacks your decision logic. Build a voice profile that captures what you reject and why you believe it.

Frequently Asked Questions

Why does my AI writing sound generic even after I add many old posts?
Old posts often leave out the "why" behind your choices. If your public writing is polished to avoid social friction, AI learns to be safe and agreeable instead of distinct.
What should a founder voice profile capture?
It should capture four layers: your specific vocabulary, your sentence rhythm, the common advice you reject, and the deeper rules that guide your choices.
Can an interview replace the founder's review of AI content?
No. You still own the judgment behind what it says. You must approve the final direction and check that the draft reflects your actual judgment.