One of the most common complaints people have about AI writing is simple: *“It doesn’t sound like me.” (Or in my case, the CEO I was ghost-writing a newsletter article for). *

The problem usually isn’t the model. It’s the prompt.

Most people ask AI to copy style. But good writing is more than tone or vocabulary. It’s how ideas are framed, how arguments are built, and how expertise is translated into insight.

If you want an AI model to write convincingly in someone’s voice (your own or another author’s), you need to teach it how that person thinks, not just how they sound.

We’re going to do that using ChatGPT’s Custom Instructions, but most LLMs have a designated space for these types of instructions.

Now, you may already have Custom GPTs in place as well. If you do, their instruction stack will override most personal settings. Each Custom GPT contains:

• its own system instructions
• optional knowledge files
• defined behavior and scope

**Why Custom Instructions? **

Custom Instructions act as default global guidance for your chats.

They influence:

• tone
• writing preferences
• formatting habits
• general communication style

They apply automatically when you start a normal ChatGPT conversation.

Examples of what they affect:

• writing voice
• preferred structure (bullet points, headings)
• reading level
• professional context

They do not fundamentally change the model, but they bias responses toward your stated preferences.

**Here is the 3-step process I use to create custom instructions. **

Step 1: Extract the Author’s Voice

Start a new chat and paste the following prompt. Then include several samples of the author’s writing underneath. These can be links to online material, uploads, or pasted (into the chat) examples. The goal is to have ChatGPT reverse-engineer the author’s voice and thinking patterns.

----copy from “You” down----

You are creating a writing model of the author whose work appears below.

Identify patterns in:

• Tone and personality

• Sentence length and rhythm

• Vocabulary level

• How ideas are introduced

• How arguments are structured

• Common frameworks or metaphors

• How the author connects ideas to practical outcomes

• How the author concludes pieces

Identify:

• Recurring themes

• Preferred ways of explaining complex ideas

• Typical narrative structure

• How the author speaks to professional audiences

Produce a reusable profile that describes:

• Tone

• Structure

• Vocabulary

• Pacing

• Storytelling patterns

• How the author translates expertise into insight

Produce 8–12 specific rules that should always be followed when writing in this author’s voice.

Confirm you will use this voice and thinking profile for all writing tasks in this conversation.

[paste several samples of your writing here]

----stop copying after your writing samples----

Why this prompt? Most prompts only copy style. This one captures:

• tone
• narrative structure
• conceptual framing
• how you explain ideas

With this, the model learns how the author/speaker thinks about ideas, not just how sentences look. That difference dramatically improves the output. I'm so thrilled with my Claude Project that writes in my brand voice. It uncovered aspects of my writing that I was not conscious of, providing unexpected insights into how I communicate.

Step 2: Refine the Voice if Needed

If the AI output sounds “close” but not quite what you wanted, paste this into the chat window on this chat:

“Edit this so it sounds even more like the original author's voice. Remove generic phrasing and strengthen the author's distinctive style.”

Step 3: Save Your Voice in Custom Instructions

Review all of the model’s outputs (this is called a “human in the loop” guardrail). Make sure there are no hallucinations (made-up or biased material) and that the person whose voice you’re striving to match agrees that the output actually does match, you can have ChatGPT automatically remember your preferred tone by using Custom Instructions. This removes the need to paste voice prompts every time (use Projects in Claude).

How to set Custom Instructions

• Click your profile icon (bottom-left corner).
• Select Settings.
• Open Personalization.
• Click Custom Instructions.
• Fill in the two fields.

You can paste the voice profile directly into the instructions, so the model automatically writes in that tone across conversations.

Generating Content in Your Voice

Once the steps above are complete, you’re ready to generate content in your voice. Give your prompt normally. Use your favorite prompt framework (need some ideas?) and type or paste them into the chat box. Mine happens to be RFT (role, task, format).

Example:

R – Role

Tell the AI who it is supposed to be. “You're a chief marketing officer at a professional association.”

T – Task

Tell it what to do and what inputs to use. “Review the uploaded materials on brand voice and messaging for our webinar program and write four social posts from our board chairman, in her voice, emphasizing value and WIIFM. Use behavioral economics techniques like loss aversion.”

F – Format

Tell it how the final output should be delivered. “The posts will go on LinkedIn.”

Because the model already analyzed your voice, the output typically sounds much closer to your natural writing style.

Why This Matters

Most AI writing advice focuses on tools.

But the real shift is conceptual.

AI is most powerful when it becomes a thinking partner, not just a text generator. When you train a model on voice, frameworks, and reasoning patterns, you’re essentially building a lightweight version of your own intellectual operating system.

Instead of starting from scratch each time, the model begins with:

• your framing
• your narrative patterns
• your way of connecting ideas to outcomes

The result is writing that feels less like AI output and more like augmented expertise.

And that is where AI becomes genuinely useful for professionals.

**TIP: **Don’t forget about the “human in the loop” safeguard! You should be checking the ChatGPT output for hallucinations (false content), bias, and fidelity (I love that word!) to the author/speaker voice at every stage.

If you try this approach and discover refinements that work particularly well, I’d love to hear about them! AI workflows improve fastest when we share what we’re learning.