The Voice Training Framework #
The most effective way to kill the "AI smell" is to provide concrete stylistic anchors. Instead of telling the AI to be "casual" or "professional," show it exactly how you write.
Use this prompt to clone your style before starting a task:
Here are three examples of my writing style:
[Paste 3 paragraphs you've written that you like]
Analyze my writing style. Note:
- Sentence length patterns
- Tone and formality level
- Use of humor or sarcasm
- Vocabulary preferences
- Structural patterns
Then write [your task] matching my exact style.
Advanced Refinement Techniques #
Even with voice training, AI tends to be too "perfect." Real human writing has texture—opinions, quirks, and occasional rough edges. I use a "de-robotizing" pass after the first draft is generated.
The "Human Texture" Prompt:
This sounds too perfect and too AI-generated.
Make it more human by:
- Adding a personal anecdote or opinion
- Including an imperfect transition between sections
- Adding a sentence that starts with "Look," or "Here's the thing"
- Making the tone slightly more casual and opinionated
- Adding one mild controversial take
AI as an Editor, Not an Author #
For a truly professional AI workflow, flip the script: write the first draft yourself and use the LLM as a high-end editor. The key is using negative constraints to prevent the AI from adding corporate fluff.
The Constraint-Based Editing Prompt:
Edit this for clarity and flow. DO NOT:
- Change my tone or voice
- Add corporate buzzwords
- Make it more formal
- Add filler phrases
DO:
- Cut unnecessary words
- Improve sentence variety
- Strengthen weak transitions
- Suggest where I need more specific examples
By treating the LLM as a collaborator rather than a ghostwriter, you maintain ownership of the perspective while leveraging the AI for structural polish.
Next Building an LLM agent doesn't actually require a massive →