The email prompt you are using right now probably looks something like this:
"You are my professional email assistant. Write in a concise, warm, and direct tone. Avoid filler phrases like 'I hope this email finds you well.' Keep it short."
That is a fine prompt. It is also functionally identical to the prompt being used by approximately 40 million other ChatGPT users.
Let me show you what the difference looks like in practice.
The setup
I took two prompts:
Prompt A (typical Reddit-sourced):
"You are a professional email assistant for a consultant. Write emails that are concise, direct, and warm. Avoid corporate jargon. No filler. End with a clear call to action."
Prompt B (first-person persona prompt, built from actual sent emails):
"I am a [role] who sends 40-60 emails per day. My writing style: I open with the point, not pleasantries, usually one sentence that frames what I need or why I am writing. My sentences are short (under 20 words average). I never use 'per our conversation' or 'as discussed.' I end emails with a single question when I want a reply, or a statement when I don't. My closers: first name only for warm contacts, 'Thanks' + first name for neutral-to-formal. I never say 'best regards' or 'kind regards.' When I am asking for something I am not entitled to, I acknowledge it directly ('I know this is a lot to ask') rather than burying it in passive framing."
Same task: Draft a follow-up to a client who has not responded in two weeks.
The outputs
Output from Prompt A:
Subject: Following Up
Hi [Name],
I wanted to follow up on my previous email regarding [project/topic]. I understand you may be busy, but I wanted to check in and see if you had any questions or needed any additional information from my end.
I look forward to hearing from you at your earliest convenience.
Best, [Your Name]
Output from Prompt B:
Subject: Quick check-in
Hey [Name],
Still thinking about [specific thing from last conversation]. Any update on your end?
[First name]
The analysis
Both emails accomplish the follow-up task. But notice what is different:
Subject line. "Following Up" is a template subject. Every person in their inbox has received three "Following Up" emails this week. "Quick check-in" is mildly more personal, but the real signal is in what you do not see. A persona-trained prompt would eventually learn whether you use the recipient's name in subject lines, whether you reference the specific thing you were discussing, or whether you write bare-minimum one-word subjects with warmth established in the body.
Opening. Prompt A produces "I wanted to follow up on my previous email regarding..." This is a sentence that has appeared verbatim in hundreds of millions of emails. It is AI-shaped. Prompt B opens directly on the thing (the specific reference from the last conversation), which is how people who write 40-60 emails a day actually write. Every word is functional.
The ask. Prompt A buries it in "I wanted to check in and see if you had any questions." The subtext is: please respond to my email. Prompt B says exactly that: "Any update on your end?" One sentence. The point.
Closer. "Best, [Your Name]" vs. just "[First name]." This is not a small thing. Your closer is part of your email signature in the relationship sense: it is how the person on the other end has come to expect your emails to end. Getting this wrong is a subtle tell.
Why prompt engineering alone cannot close this gap
The quality of your ChatGPT email prompt is bounded by your ability to describe your own writing. And here is the thing: most people are not very good at describing how they actually write. They describe how they want to write, or how they think they write, which is often aspirational.
The only way to get a prompt that captures the fine-grained specifics of your style is to derive it from your actual writing. Not from a description you wrote. From the emails themselves.
That is the premise behind the FinalDraft Persona Prompt Generator. It walks you through the structural patterns of your writing (your openers, your closers, your sentence rhythm, your pattern for different relationship types) and produces a first-person prompt grounded in specifics rather than adjectives.
It takes about 5 minutes and you can use the output in any AI tool you already have. No account required.