"Write a professional email to my client about the project delay."
This is how most people use ChatGPT for email. It is the task-first approach: tell the AI what to write, let it figure out how.
It works. You get an email. It is grammatically correct. It is professional. It covers the topic.
It does not sound like you.
Here is the specific reason why, and what the alternative is.
The problem with task framing
When you say "write a professional email about X," you are specifying what but leaving all of the how to the model's defaults.
"Professional" is not a voice. It is a register: a point on a formality spectrum. But within "professional," there are thousands of different ways to write. The consultant who writes three-sentence emails. The account manager who always starts with a reference to the last call. The founder who uses the client's first name in the subject line. The advisor who never writes more than one paragraph to important clients.
All of those are professional. None of them is "professional" in the generic sense that ChatGPT defaults to.
When you do not specify, ChatGPT produces the statistical center of gravity of professional email: medium-length paragraphs, polite opener, clear structure, comprehensive sign-off. Competent. Generic.
The instructions people add that do not help
The typical response to "this does not sound like me" is to add instructions. "Make it shorter." "Warmer." "Less formal." "Remove the bullet points."
These instructions are directional corrections, not voice injections. They push the output toward a different part of the generic distribution (shorter, warmer, less formal), but they do not move it toward you specifically. You are still getting the average of everyone who writes short, warm, informal email, not how you write short, warm, informal email.
What actually bridges the gap
Two things, in order of impact:
First: First-person framing. Instead of "write a professional email," tell the AI to write as you, in your own voice, from your first-person perspective. "I am writing a follow-up email to a client. Here is how I typically write: [your specific patterns]. Draft this as if you are me."
The shift from third-person task to first-person impersonation is significant. The model's job changes from "produce professional email" to "sound like this specific person." That produces meaningfully different outputs.
Second: Specific patterns, not general adjectives. Instead of "write in a warm, direct tone," describe your actual patterns:
- "I open emails with one sentence that states the purpose, then supporting context."
- "I use the client's first name in the greeting but not in the body."
- "I end with a single question when I want a reply, or nothing when I do not."
- "I never use 'please let me know if you have any questions'; I state the next step instead."
Adjectives ("warm," "direct") give the model latitude to interpret. Specific patterns give it instructions it can actually follow.
The first-person persona prompt
This is the framework behind the FinalDraft Persona Prompt Generator: a structured process for building a first-person prompt that captures your specific email patterns.
The output looks like: "I am a [role] who writes 30-50 emails per day. My writing style: I open with the point... I tend to keep body paragraphs under three sentences... My closers..."
This prompt can be pasted into any ChatGPT conversation as a system message. It does not fully replace voice-trained email (for that you need a tool that reads your sent history), but it closes a meaningful gap between "write a professional email" and "sound like me."
The generator is free, takes about 5 minutes, and you can use the output in ChatGPT today.