Voice Profile

A VOICE.md file that captures how you actually write -- your patterns, phrases, tone, the whole thing. When your agent writes tweets, articles, or emails for you, it references this file so the output sounds like you, not like ChatGPT.

Generic AI writing is painfully obvious. It uses the same filler phrases, the same structures, the same "let's dive in" energy. A voice profile teaches your agent what you actually sound like. The test is simple: would you actually say this?

Setup Prompt

I want you to build a voice profile for me. Analyze these writing
samples and create a VOICE.md file that captures:

- Core voice traits (e.g., direct, casual, educational)
- Sentence patterns I actually use
- Structure patterns (how I organize thoughts)
- Phrases I say vs. phrases to NEVER use
- The "vibe" -- how would a friend describe my writing?
- AI patterns to specifically avoid (list the cliches)
- A test: before writing anything for me, check
  "would I actually say this?"

Include good and bad examples showing the difference between
my voice and generic AI voice.

[Paste 20-50 of your tweets, posts, or writing samples here]

What Makes a Good Voice Profile

1. Real examples -- actual quotes from your writing, not descriptions of your style
2. Anti-patterns -- specific phrases to avoid ("Here's the thing:", "Let's dive in", "Game changer")
3. The vibe test -- a quick gut-check to see if output sounds right
4. Format-specific notes -- tweets sound different from articles sound different from emails

Tips

• Feed it at least 20 samples. More data = better profile.
• Include variety -- tweets, longer posts, casual messages.
• Update it every few months. Your voice evolves.
• Reference it everywhere -- tweet pipeline, article drafts, email templates, all of it.

Set Your Expectations

This one takes time. The AI-native writing patterns are baked really deep into these models and a VOICE.md file isn't going to just overwrite all of that. You'll build out the profile, ask your agent to draft something, and it'll still sound like AI wrote it. That's been my experience across months of working on this.

It does get better with iteration. You go back and forth on a draft, tell the agent what sounds wrong, update the voice file when you notice patterns. Slowly the output gets closer. But I want to be honest about it -- even with a strong voice profile, any piece of writing you produce with your agent is going to need editing on your part. The models can get the structure and general tone right but the stuff that makes writing actually sound like you and not a well-prompted LLM -- that still requires a human pass.