How to Build a Custom AI Assistant That Actually Sounds Like You
Most people use AI writing tools the same way they use a search engine — type something in, get something out, move on. The result is content that's technically correct but unmistakably generic. It doesn't sound like you. It doesn't think like you. And your readers can tell.
Why Generic AI Outputs Fall Flat
The default behavior of any large language model is to produce the most statistically average response to your prompt. That's not a bug — it's how these systems are designed. But for creators, that average is the enemy. Your voice is the thing that makes your work worth reading. It's the accumulated weight of your experiences, your opinions, your rhythms, and your obsessions. No off-the-shelf AI prompt will capture that automatically.
Step 1: Build a Voice Document
Before you write a single prompt, you need a voice document — a reference file that captures how you actually write. Pull 10–15 examples of your best work: blog posts, emails, social threads, anything where you felt like you were fully yourself. Look for patterns. What sentence lengths do you favor? Do you use rhetorical questions? Do you open with a story or a provocation? Do you use contractions? How do you handle transitions? Write these observations down in plain language. This document becomes the foundation of every prompt you write.
Step 2: Design Your System Prompt
The system prompt is the persistent instruction set that shapes every response your AI assistant gives. Think of it as a briefing document for a very capable but context-free collaborator. A strong system prompt includes: your role and purpose, your audience and their expectations, your tone and voice characteristics drawn from your voice document, things you never do (passive voice, filler phrases, excessive hedging), and the format you want responses in. The more specific you are, the better. 'Write conversationally' is weak. 'Write in short paragraphs, use em-dashes for asides, avoid the word utilize, and never start a sentence with However' is strong.
Step 3: Calibrate With Test Prompts
Once your system prompt is drafted, run it through a battery of test prompts that represent your actual use cases. Ask it to write an introduction to a topic you know well. Ask it to respond to a reader question in your voice. Ask it to rewrite a paragraph you've already written. Compare the outputs to your voice document. Where does it nail it? Where does it drift? Iterate on your system prompt based on what you find. This calibration phase typically takes 3–5 rounds before you have something reliable.
Step 4: Build Your Prompt Library
A custom AI assistant isn't just a system prompt — it's a library of task-specific prompts that work together. Build prompts for your most common tasks: drafting outlines, writing first paragraphs, generating headline options, editing for clarity, repurposing content for different formats. Each prompt should reference your voice document principles and specify the exact output format you want. Store these in a simple document or tool like Notion so you can access and refine them over time.
The Ongoing Practice
Building a custom AI assistant is not a one-time project — it's an ongoing practice. Your voice evolves. Your use cases change. The models improve. Plan to revisit your system prompt every few months, add new prompts as you discover new use cases, and keep a running log of outputs that surprised you (positively or negatively). The creators who get the most from AI are the ones who treat it as a craft, not a shortcut.
Key Takeaways
- 1Build a voice document from your best existing work before writing any prompts
- 2Your system prompt should be specific enough to exclude things, not just include them
- 3Calibration takes multiple rounds — budget time for iteration
- 4A prompt library compounds over time; start building it from day one
Want help building your AI workflow?
We work with creators to design custom AI systems that amplify their voice and accelerate their work — without replacing what makes it theirs.