How to Make ChatGPT Sound Like You: The Voice Capture System

You can make ChatGPT sound like you. Not "kind of like you." Not "close enough." Actually like you, to the point where your audience can't tell the difference on 85%+ of your posts.

The problem is not ChatGPT. The problem is that most people type "write a LinkedIn post about X" and expect it to read their mind. It can't. It defaults to a bland, committee-approved tone that reads as AI-generated within the first sentence.

The fix takes 2 to 4 hours and works on ChatGPT, Claude, and any other large language model. Here is the full system.

Why ChatGPT sounds generic by default

ChatGPT was trained on the entire internet. Its default voice is a weighted average of millions of writers. That average sounds like nobody. It sounds like a press release drafted by a committee.

Specifically, default ChatGPT output has these tells:

• Sentences are all the same length (15 to 20 words, no variation).
• Vocabulary is formal and safe ("utilise" instead of "use," "numerous" instead of "lots").
• Paragraphs open with transition phrases ("Furthermore," "In addition," "It's worth noting that").
• Opinions are hedged ("This could potentially be beneficial for some users").
• Structure follows essay format: intro, body, conclusion. Every time.

None of this sounds like a real person. We covered this in depth in why AI content sounds generic and how to fix it. The short version: you need to give ChatGPT data about how you specifically write.

The voice capture system: 4 steps

This is the same process we use when building voice systems for clients. The DIY version takes 2 to 4 hours. The principles are the same whether you use ChatGPT, Claude, or any other model.

Step 1: Collect 10 writing samples

You need 8 to 12 pieces of writing that represent your real voice. Not your polished, edited, cleaned-up voice. Your natural voice.

Where to find samples:

• Your best-performing LinkedIn posts (the ones that got replies, not just likes).
• Emails you've sent to clients or prospects that felt natural.
• Unedited blog drafts before someone cleaned them up.
• Voice memos or podcast transcripts (messy is fine, that's the point).
• DMs or Slack messages where you're explaining something you know well.

Avoid: copy that someone else wrote for you, heavily edited pieces, formal reports, anything where you were deliberately writing in someone else's style.

Paste all 10 samples into a single document. Label each one with the format (LinkedIn post, email, blog, etc).

Step 2: Extract the 12 voice dimensions

This is the step most people skip. They jump straight from samples to prompt. That's like building a house without blueprints.

Open a new ChatGPT conversation. Paste your 10 samples and use this prompt:

The 12 voice dimensions to extract:

1. Sentence length — average word count per sentence, range (shortest to longest), and rhythm pattern.
2. Paragraph length — typical number of sentences per paragraph.
3. Vocabulary register — formal, casual, technical, colloquial. Specific word choices that recur.
4. Contraction usage — always, never, or situational. Examples.
5. Opening patterns — how do they start posts? Questions? Statements? Numbers? Stories?
6. Transition style — how do they move between ideas? Conjunctions? Line breaks? No transitions?
7. Opinion expression — direct or hedged? Confident or qualified? Example phrases.
8. Punctuation habits — em dashes, semicolons, ellipses, exclamation marks. Frequency and context.
9. Structural patterns — do they use lists? Headers? Numbered steps? Freeform prose?
10. Signature moves — recurring phrases, constructions, or rhetorical devices unique to this writer.
11. Tone range — what's the emotional range? Funny to serious? Warm to clinical?
12. Banned patterns — what does this writer never do? What would sound wrong in their voice?

ChatGPT will return a detailed analysis. Save it. This is your voice blueprint.

Step 3: Build the voice prompt

Now turn the analysis into a structured prompt. The format matters. ChatGPT responds better to structured instructions than to prose paragraphs.

A working voice prompt is 400 to 800 words. Shorter than that and it misses nuance. Longer than that and the model starts ignoring instructions.

The key mistake: describing the voice in abstract terms ("write in a conversational, engaging style"). This tells ChatGPT nothing. Instead, give it mechanical rules: "Average sentence length: 12 words. Range: 4 to 22. Use fragments for emphasis, max 2 per section. Never start a paragraph with a transition word."

Specifics beat adjectives. Always.

Step 4: Test and iterate twice

Version 1 of your voice prompt will be 65 to 75% right. Good enough to see the shape. Not good enough to publish without editing.

Test it by generating 5 pieces of content on topics you've written about before. Compare the output against your original writing. Look for these specific gaps:

• Sentence rhythm: does the output vary sentence length the way you do? Or is every sentence the same length?
• Word choice: does it use your words or its default vocabulary?
• Opening strength: do the openings sound like you or like a ChatGPT intro?
• Opinion clarity: does it state opinions directly or hedge them?
• Overall feel: if you showed this to a client, would they believe you wrote it?

For each gap, add a specific instruction to the voice prompt. "Never open with a question followed by 'Let me explain.'" or "When stating an opinion, state it in under 10 words with no qualifiers."

After iteration 1, you'll be at 75 to 85%. After iteration 2, you'll be at 85 to 90%. That's the range where most readers can't tell the difference between your writing and the AI output.

We've found that going beyond 2 iterations hits diminishing returns. The last 10% of voice accuracy is not worth the effort for most use cases. If you need 95%+, you're in human ghostwriter territory.

The custom GPT shortcut

Once your voice prompt is working, turn it into a custom GPT (ChatGPT Plus or Team required) or a Claude Project. This saves you from pasting the voice prompt into every conversation.

In the custom GPT instructions, paste your voice prompt plus these additions:

• Your typical content formats (LinkedIn post, newsletter, blog intro, etc).
• Your audience description (who reads your content).
• Your topic guardrails (what you write about, what you don't).

Now every conversation with your custom GPT starts from your voice baseline. You type "write a LinkedIn post about [topic]" and it already knows how you write, who you're writing for, and what your content pillars are.

What most voice capture guides get wrong

Three common mistakes that keep people stuck at the "sounds kind of like me" stage:

Mistake 1: Using adjectives instead of mechanics

"Write in a warm, conversational, engaging tone" gives ChatGPT no actionable information. Every writer thinks they're warm and conversational. The fix: replace every adjective with a measurable instruction. "Warm" becomes "use 'you' in the first sentence of every section." "Conversational" becomes "average sentence length 12 words, fragments allowed, contractions always."

Mistake 2: Too few samples

Three samples is not enough. The analysis needs enough data to distinguish your patterns from noise. If you only give it 3 LinkedIn posts, it might latch onto a structure you used once (numbered list) and treat it as your signature move when it was just a one-off.

Ten samples across multiple formats gives the model a reliable signal.

Mistake 3: Skipping the banned patterns list

Telling ChatGPT what not to do is as important as telling it what to do. Without a banned list, it falls back to defaults: "In today's fast-paced world," "Let's explore how," "It's important to note that." Your brand voice guide should have a clear list of phrases and patterns that would sound wrong in your voice.

What happens after the voice prompt is working

The voice prompt is the foundation. Once it's producing content at 85%+, you add workflow on top:

• Batching: generate a week's worth of content in one sitting. We cover the exact workflow in the 90-minute LinkedIn batching system.
• Repurposing: take one LinkedIn post and generate an email, a tweet thread, and a blog intro from it. Same voice prompt drives all formats.
• Iteration log: keep a running document of voice prompt changes. When you update the prompt, note what changed and why. Version control for your voice.

The system compounds. By month 3, you'll have a voice prompt that produces publishable content on the first draft for 80%+ of your topics. The remaining 20% needs a light edit, not a rewrite.

That's the difference between a prompt and a system. Prompts give you one output. Systems give you compounding output over time.

FAQ