One Prompt, Properly Built

One Prompt, Properly Built: The Email That Sounds Like You

A good prompt is built, not sprinkled with magic words. Sera Voss builds one real prompt line by line so your AI email finally sounds like you.

A good prompt is built, not sprinkled with magic words. There is no phrase you can add that suddenly makes the output brilliant. What makes a prompt work is structure. Let me show you the difference on one real task: writing an email that actually sounds like you.

The weak prompt everyone writes

Here's the version most people type:

"Write a friendly email to a client following up about our project."

It's not wrong, exactly. It will produce an email. But it will produce anyone's email — generic, slightly corporate, the kind of message that could have come from any business on earth. That's not the model failing. That's the model doing its best with almost nothing to go on.

The four things it's missing

A prompt that produces usable writing needs four ingredients. The weak prompt has none of them.

Context. Who is this client, what's the project, what's the actual situation? "Following up" could mean a dozen things. The model has to guess, and it guesses toward the average.

Role. Who is writing? A prompt that starts by telling the model whose shoes to stand in produces far more specific output than one that doesn't.

Constraints. How long? What tone? What should it not say? Constraints don't limit good writing — they're what makes writing sound deliberate instead of generic.

Examples. The single most powerful thing you can add. Show the model one email you've actually sent, and it will match your rhythm far better than any adjective like "friendly" ever could.

Building it, line by line

Watch what happens as we add each ingredient.

Role: You are writing as me — the founder of a small studio. My tone is warm, direct, and a little dry. I don't use exclamation points or corporate filler.

Context: I'm following up with a client, Maria, two weeks after delivering the first draft of her brand refresh. She went quiet. I want to check in without sounding anxious or pushy.

Constraints: Keep it under 120 words. No "just circling back," no "I hope this email finds you well." One clear, low-pressure question at the end.

Example of my voice: "Hi Maria — no rush at all, but I wanted to make sure the draft landed. If anything felt off, I'd rather hear it now than polish the wrong thing. What's your read so far?"

Now write the follow-up.

That's the whole trick. Nothing magical. Just enough information that the model can hit your target instead of the average of everyone's.

The reusable version you keep

Here's the part that saves you time forever. Once you've built this, you don't rebuild it. You save the role, tone, and example — the parts that are always true about you — as a template, and swap only the context and constraints for each new email.

[Fixed] Role + tone + one voice example. [Swap each time] Who, what situation, length, and the one thing to avoid.

Build it once. Reuse it for a year.

Why the "example" does the heavy lifting

Of the four ingredients, one matters more than the rest, and it's the one people skip: the example. Adjectives are weak instructions. "Friendly," "professional," "warm" — every one of those means something different to every reader, and the model has to average across all of them. Your idea of friendly and the internet's idea of friendly are not the same, and a word can't close that gap.

An example closes it instantly. When you show the model one real sentence you'd actually write, it stops guessing what "your voice" means and starts matching something concrete — your sentence length, your punctuation habits, the words you'd never use. One good example teaches the model more about your voice than a paragraph of description ever could.

This is why the reusable template works so well. The example is the part of you that stays constant, so you write it once and carry it everywhere. You're not re-explaining yourself each time. You're handing the model the same reference sample and simply changing the assignment.

What to do when you don't have an example yet

Sometimes you're writing something you've never written before, so there's no past example to hand over. This is where most people give up on the example entirely. Don't — there are two easy fixes.

The first: use an example of the feeling, even if it's not the same format. Writing your first-ever difficult client email? Paste in a text you sent a friend that had the exact warmth and directness you want. The model doesn't need a matching document. It needs a sample of your voice, and your voice shows up everywhere you write.

The second: describe what you don't want, which is often sharper than describing what you do. "Not corporate, not apologetic, no exclamation points, nothing that sounds like a template" gives the model firm edges to stay inside. Negative constraints are underrated precisely because they're so specific — there are a hundred ways to be "friendly," but "no exclamation points" has exactly one meaning.

Between a feeling-example and a few things to avoid, you can steer the model well even on something entirely new. The example was never really about the format. It was always about giving the model a fixed point that is unmistakably you.

The takeaway

If your AI writing sounds generic, the problem is almost never the model and almost never a missing magic word. It's that you gave it a sentence and expected a paragraph of you.

Let's remove the vocabulary for a second: a prompt is just a clear brief. Give the model what you'd give a talented new assistant on their first day — who you are, what's going on, the rules, and one example — and it will write like it's known you for years.

promptingai-writingtutorialemail
← All posts