The Easiest Way to Get Your Personal Tone of Voice from ChatGPT

You've done it. Everyone has. You opened ChatGPT, pasted in a brief, and typed some version of: "Write this in my voice." What came back was the same vaguely confident, faintly enthusiastic, mildly inspirational LinkedIn paste that everyone else's AI also produced that morning. You tweaked the prompt. Added "casual but professional." Added "like Naval but warmer." Added three exclamation points and a request for "no fluff."

It still sounded like a stranger doing an impression of a marketing manager.

Here's the thing nobody tells you: ChatGPT cannot imitate what it has never seen. Asking a model to "write like you" without giving it your writing is like asking a tribute act to perform a song they've never heard. It will perform something. It will not be your song.

But ChatGPT is excellent at one thing closely related to imitation: analysis. Give it your actual writing and ask it to describe what it sees, and you get a different result entirely. That description, properly captured, becomes the asset. The article you generate later is just a side effect.

The Core Insight: Describe, Don't Imitate

There's an old line from the ghostwriting world that ghostwriting is twenty percent writing and eighty percent extraction. The hard part isn't typing. The hard part is sitting with someone long enough to figure out how they actually think, what phrases they reach for under pressure, which sentence shapes feel like home. Writing in someone's voice without first extracting that voice is impossible. Software doesn't change the math.

So the move is to flip the prompt. Stop asking ChatGPT to produce text in your style. Start asking it to produce a description of your style. The output of that conversation is a document. Call it a Tone of Voice Document. It belongs to you, you can edit it, you can paste it into any tool, and you can use it for years.

Generation happens after that. Generation is the cheap part. Extraction is the part most people skip, which is why most people's AI content still reads like a press release written by someone who has never met them.

The Three-Tier Prompt System

There are three reasonable depths to do this at, depending on how much you want to invest. Pick one.

Tier	Time	Best for
Tier 1, Simple	5 minutes	First-timers, single piece of writing
Tier 2, Intermediate	15 minutes	Regular writers wanting structured output
Tier 3, Advanced	30 minutes	Serious personal brand builders

Tier 1: Simple (5 minutes)

Open a new chat. Paste a single recent piece of your writing, ideally something over 300 words that you're proud of. Then run this:

Analyze the writing voice, tone, and structure of the text below. Output as bullet points.
[paste your text]

You'll get a serviceable summary. Some of it will be obvious, some surprisingly sharp. Read it. Note what you agree with and what feels off.

Then run the follow-up:

If I were to ask ChatGPT to emulate this writing style, how would I describe the style and instruct ChatGPT?

This second prompt is the magic one. It converts observations into instructions, which is what you actually need. Save the output somewhere. You now have a usable, if rough, voice spec.

Tier 2: Intermediate (15 minutes)

If you write regularly, pull together at least three samples of at least 200 words each. Different topics if possible. A LinkedIn post, a long email, a blog intro. The variety matters because your voice shows up in different keys depending on the platform.

I'd like an analysis of the tone and writing style from my writing samples below.
Please identify: 
One primary tone and one secondary tone
One primary writing style and one secondary writing style
Recurring sentence structures
Phrases or words I use often
How I open and close pieces
 [paste your samples]

The structured output makes this version dramatically more useful. You'll see your patterns spelled out: maybe you open with a question two-thirds of the time, maybe every third sentence is short, maybe you keep using the word "honestly" without realizing it. These are the fingerprints. Generic AI output has none of them.

Tier 3: Advanced. The Tone of Voice Document (30 minutes)

This is the one most worth doing if you're serious. The goal here isn't a summary. The goal is a document detailed enough that another AI, or another writer, could read it and produce work that actually sounds like you.

Gather a thousand words of your best, most-you writing. More is better. Dean Seddon, who built a content business on this approach, recommends twenty thousand words for a serious training pass. A thousand is the working floor.

You are an AI language model skilled at analyzing text for writing style and tone of voice. I will provide examples of my writing. Analyze them and create a detailed "Writing Style and Tone of Voice Document" covering:
1. Tone overview (formal/casual, serious/playful, etc.) 2. Sentence structure patterns 3. Paragraph length preferences 4. Vocabulary: words I use often, words I avoid 5. How I open pieces 6. How I close pieces 7. Emotional register 8. Platform-specific variations if applicable
[paste your writing — aim for 1,000+ words total]

Read the output critically. Edit the parts that feel wrong. ChatGPT will sometimes overstate a pattern or invent a tendency you don't actually have. The document is yours, not the model's. Treat the AI's draft as a strong first pass, not a verdict.

Once you're happy with it, run a second pass to make it more portable:

Now describe this writing style step-by-step as if you were explaining it to another AI — so any AI model can consistently reproduce it.

This converts your document from descriptive prose into something closer to an instruction set. It's the version you'll actually paste into other tools.

What To Do With Your Tone of Voice Document

A document sitting in a Notion page is worth nothing. Three places to put it to work:

Custom Instructions

Open ChatGPT settings. Find "What traits should ChatGPT have?" Paste a condensed version of your Tone of Voice Document there. Now every chat, by default, knows how you write. You don't have to remember to invoke it. This single move is the highest-return twenty seconds you'll spend in ChatGPT this year.

Custom GPT

Build a custom GPT. In the system instructions, paste the full Tone of Voice Document. In the knowledge base, upload five to ten of your strongest pieces. Name it something boring like "My Writing Assistant." This is the setup Dean Seddon uses to produce a week of LinkedIn content in roughly an hour. The GPT has both the rules and the examples, which together cover what neither alone can.

Paste-at-Top Workflow

For one-off tasks in a fresh chat, keep your Tone of Voice Document as a text snippet on your clipboard manager. Paste it at the top, then write the task underneath. Slower than the first two options. Useful when you want a different model, a different account, or a temporary chat with no memory.

Common Mistakes (Why It Doesn't Work for Most People)

Most people who try this once and walk away unimpressed made one of four errors.

Too little input. Under five hundred words and ChatGPT will hallucinate a generic professional voice. It needs enough material to find real patterns instead of inventing plausible ones. A thousand words is the comfortable floor. Three thousand is better.

Describing the desired style instead of the real one. People feed in writing they wish they wrote, or aspirational pieces from someone they admire. The output is then a description of that other person's voice, which is exactly the problem you started with. Use what you actually wrote, last week, on a Tuesday, when nobody was watching.

Never updating the document. Your voice moves. The way you wrote two years ago is not the way you write now, and the document should keep up. Re-run the analysis every six months. Add new samples. Strip out patterns that no longer fit.

Using pure descriptors instead of samples. "Write in a witty professional voice with hints of sarcasm" is a prompt every junior copywriter in the world has typed. It produces a junior copywriter's idea of witty. The document approach works because it's grounded in your actual sentences, not in an adjective stack.

The Nielsen Norman Group maps tone across four dimensions: formal to casual, serious to funny, respectful to irreverent, matter-of-fact to enthusiastic. A pure-descriptor prompt picks one point in that space and aims at it. Your real voice, and the document that captures it, is a shape across all four dimensions, with different positions for different contexts. Shapes beat points.

One Last Thing

If all of this sounds like a job, that's because it is. The extraction step is real work, and the maintenance step is real work, and most people do neither.

If you want to skip the whole thing, Co.Actor does this automatically. It analyzes your writing across everything you've published and applies your voice to every post you create, with no document to maintain and no prompt to remember. You can try Co.Actor free.

Either way, the rule holds. Stop asking ChatGPT to imitate you. Get it to describe you, write that description down, and feed it back. Extraction beats imitation. Always has.