← Back to blog
·13 min read

How to Make AI Content Sound Like You (Not Like AI)

62% of marketers say AI content sounds generic. The fix isn't better prompts — it's a voice extraction methodology. Here's the exact system I use.

You can tell within three sentences.

The rhythm is too even. Every paragraph is the same length. The vocabulary is suspiciously clean. There's a "delve" in paragraph two and a "comprehensive guide" by paragraph four. And the whole thing reads like it was written by someone who has opinions about nothing.

That's AI content. And your readers know it. Your customers know it. Google knows it.

The advice you'll find online is useless. "Just add your personal touch." "Write better prompts." "Tell ChatGPT to be more casual." I tried all of it. The output went from robotic to slightly-less-robotic-but-still-obviously-AI.

The problem isn't prompting. The problem is that the AI doesn't know how you sound. It doesn't know your sentence rhythm, your vocabulary, your storytelling patterns. It doesn't know that you never say "comprehensive" and always say "tiny." It doesn't know you lead with failure stories, not success stories. It doesn't know that your paragraphs are short and your opinions are strong.

So it defaults. To the average of everything it's ever read. Which sounds like everything and nothing at the same time.

There's a fix. It's not a prompt trick. It's a methodology I call brand voice extraction — analyzing your best existing content, pulling out the specific patterns that make you you, and codifying them into a profile that AI references every single time it writes for you.

I've been using this system for months. The difference is night and day. Here's exactly how it works.

Why "Write in a Casual Tone" Doesn't Work

Let me show you what happens when you try to fix AI voice with prompts.

The prompt:

Write a landing page headline for an AI marketing tool.
Be casual and conversational. Avoid corporate speak.
Sound human.

The output:

"Take Your Marketing to the Next Level with AI That Gets You"

Still generic. Still sounds like AI pretending to be casual. "Gets you" is doing zero work. "Next level" is a phrase no human with taste would write. The AI followed your instructions — it avoided the worst corporate jargon — but it has no idea what "your voice" actually means.

Because "casual" isn't a voice. "Conversational" isn't a voice. Those are vibes. And vibes are too vague for a language model to do anything useful with.

A voice is specific. A voice is: "Short declarative sentences. Fragments for emphasis. Technical specificity — actual numbers, real tools, exact frameworks. Never uses 'game-changer,' 'unlock,' or 'dive into.' Default tone: smart friend explaining something at a coffee shop."

That's a voice. And when you give AI that level of specificity, the output transforms.

The Brand Voice Extraction Method

Here's the system. Five steps. Takes about 30 minutes the first time. Then you never re-do it — you just reference the profile.

Step 1: Gather Your Best Writing

Pull 3-5 pieces of content you're proud of. Not the stuff you published because you had to. The stuff that sounded like you. The email that got replies. The social post that got shared. The landing page section that converts.

If you're starting from scratch and don't have written content, use:

  • Voice memos or podcast transcripts (how you talk is your best voice data)
  • DMs or emails you sent to friends explaining your product
  • Social media comments where you were being yourself, not performing

The material doesn't need to be polished. In fact, raw is better. You're looking for patterns in how you naturally communicate — not how you think you should sound.

Step 2: Analyze Sentence Structure

Read through your samples and look for these patterns:

Sentence length. Are your sentences short and punchy? Long and flowing? A mix? Most people have a default range. Mine is 5-12 words. Fragments everywhere. That's a pattern the AI needs to know.

Paragraph length. Do you write walls of text or short bursts? I write 1-3 sentence paragraphs. Lots of white space. One idea per paragraph. That's a specific instruction AI can follow.

Transitions. How do you move between ideas? "And then." "Here's the thing." "That's when." "But." Your transitions are fingerprints.

Example from my own analysis:

PATTERN: Short declarative → Fragment → Evidence sentence
"I tried everything. Total disaster. Spent 3 hours rewriting
what ChatGPT gave me."

Step 3: Map Your Vocabulary

This is where most people skip ahead. Don't.

Words you use: List the words and phrases that show up repeatedly in your writing. Not industry jargon — your personal vocabulary. I say "ship" instead of "launch." I say "tiny" and "little" to keep things small-scale. I say "that's when" to connect cause and effect.

Words you never use: This matters just as much. I never say "leverage." Never say "comprehensive." Never say "passionate about." Never say "unlock your potential." Those words are AI tells. They're also just not how I talk.

Jargon level: How technical do you get? Do you explain terms or assume knowledge? I use "MRR" and "churn" without explaining them because my audience knows what they mean.

Voice Vocabulary Map:

USE: "ship," "build," "tiny," "that's when," exact numbers ($4, not "a few dollars")
NEVER USE: "leverage," "comprehensive," "game-changer," "unlock," "passionate about"
JARGON: Light — assume audience knows "MRR," "MVP," "product-market fit"

Step 4: Identify Your Storytelling Patterns

Everyone has default narrative structures. Here are common ones:

  • Failure-first: You always start with what went wrong before explaining what worked. (That's mine.)
  • Data-first: You lead with numbers and evidence, then explain the implication.
  • Question-first: You open with the question the reader is thinking, then answer it.
  • Contrarian-first: You state the common belief, then challenge it.

Also look at:

  • How you open. Do you start with a story? A bold claim? A question?
  • How you close. Do you end with a CTA? A philosophical line? A callback to the opening?
  • How you prove points. Do you use personal stories? Data? Examples? Screenshots?

This matters because AI defaults to a generic essay structure: introduction, body paragraphs, conclusion. If your natural style is "punch them in the face with the contrarian take in sentence one," the AI needs to know that.

Step 5: Build the Profile

Now compile everything into a single document. This is the file AI will reference every time it writes for you.

Here's the structure I use:

# Voice Profile: [Your Name]

## Summary
One paragraph describing how you sound. Not aspirational — descriptive.

## Sentence Structure
- Default sentence length: [range]
- Paragraph length: [range]
- Fragment usage: [frequent/occasional/rare]
- Transition patterns: [your go-to transitions]

## Vocabulary
### Words to USE
[your specific words and phrases]

### Words to NEVER USE
[the words that aren't you]

### Jargon level
[how technical you get]

## Storytelling Patterns
- Opening style: [how you typically start]
- Narrative structure: [failure-first, data-first, etc.]
- Proof style: [stories, data, examples]
- Closing style: [how you typically end]

## Tone
- [3-5 specific tone descriptors with examples]

## Examples
### This sounds like me:
[2-3 sentences from your best writing]

### This does NOT sound like me:
[2-3 sentences that are wrong for your voice]

That's it. One document. Maybe 500 words. But those 500 words change everything about AI output.

Before and After: The Same Request, Two Different Worlds

Let me show you the actual difference.

The request: "Write an intro paragraph for a blog post about why most AI marketing content fails."

Without voice profile:

In today's rapidly evolving digital landscape, AI has become an
indispensable tool for marketers. However, many businesses are
discovering that AI-generated content often falls short of expectations.
This comprehensive guide explores why AI marketing content frequently
misses the mark and what you can do about it.

Count the AI tells. "Rapidly evolving digital landscape." "Indispensable tool." "Falls short of expectations." "Comprehensive guide." "Misses the mark." Six cliches in four sentences. This is what AI produces when it has no voice to follow.

With voice profile loaded:

You can spot AI marketing copy in about three seconds. It's too clean.
Too balanced. Too agreeable. It has opinions about nothing and says
everything in the same rhythm.

62% of marketers say AI-generated content "sounds generic." The other
38% probably aren't reading their own output.

The fix isn't better prompts. I tried that. Spent months on it. The fix
is giving AI something it doesn't have by default — a methodology for
how good content actually works. And a voice profile so it knows how
YOU work.

Same AI. Same request. Completely different output. The second version has short sentences. Fragments. A specific statistic. A self-deprecating joke. An opinion. It sounds like a person wrote it because the AI had a specific person's patterns to follow.

How to Feed the Profile to AI

Once you have the voice profile document, the execution is simple.

In Claude Code (with skills)

If you use Claude Code, save your voice profile in your project directory. Claude reads it automatically. Every skill — Direct Response Copy, SEO Content, Email Sequences — references it when generating output.

my-project/
├── brand/
│   └── voice-profile.md    ← Claude reads this
├── .claude/
│   └── skills/
│       ├── direct-response-copy.md
│       └── seo-content.md
└── content/

You don't re-paste it. You don't re-explain it. It's persistent project context. Write once, reference forever.

That's the real advantage. Not just that the profile exists — but that it's always loaded. Every email, every landing page, every social post runs through the same voice filter. Consistency without effort.

In ChatGPT or other AI tools

Create a custom instruction or system prompt that includes your full voice profile. Paste it at the top of every new conversation. Yes, it's manual. Yes, you'll forget sometimes. That's the tradeoff without project-level context.

Some people create a saved "voice prompt" they copy-paste. Better than nothing. But you're still re-explaining context every session — your product, your audience, your brand positioning. The voice profile alone doesn't solve the full picture.

The 7 Tells That Your Content Was Written by AI

Even with a voice profile, you need to edit. Here's what to scan for on every piece before you publish:

1. Rhythm uniformity. Every sentence is the same length. Every paragraph has the same structure. Real writing has rhythm changes — short, short, long, fragment, short.

2. The AI vocabulary. "Delve," "comprehensive," "cutting-edge," "landscape," "game-changer," "leverage," "robust," "harness," "dive into," "navigate," "unlock." If you see these, rewrite.

3. Hedge phrases. "It's worth noting that," "it's important to remember," "one might argue." Real people don't talk like this. They state things.

4. Empty transitions. "That being said," "on the other hand," "furthermore." Replace with how you actually move between ideas.

5. Lack of specifics. AI loves vague claims. "Many marketers struggle with..." How many? Who? "Results can be significant..." What results? What numbers?

6. The opinion problem. AI is trained to be balanced. It hedges. It presents "both sides." Real content takes a position. If your post doesn't make at least one group of people disagree, it's too safe.

7. The opening paragraph. If it starts with "In today's [adjective] [noun]," delete it. Write a real opener. A story. A bold claim. A question. Anything but that.

Common Mistakes with Voice Extraction

Mistake 1: Being aspirational instead of descriptive

Don't write the voice profile you wish you had. Write the one you actually have. If your writing is informal and fragmented, document that. Don't write "eloquent and sophisticated" because you think it sounds better. AI will follow your instructions literally — and produce content that sounds nothing like you.

Mistake 2: Being too vague

"Professional but friendly" gives AI nothing to work with. "Short sentences, max 15 words. Occasional fragments. Uses specific numbers instead of vague claims. Opens with failure stories, not success stories." — that gives AI everything.

Mistake 3: Skipping the "never use" list

The words you avoid define your voice as much as the words you use. If your profile only says what TO do, AI will still sprinkle in "delve" and "comprehensive" because those are its defaults. You need the explicit blocklist.

Mistake 4: Not updating the profile

Your voice evolves. Every 3-6 months, re-analyze your recent writing. Add new patterns. Remove ones that no longer fit. The profile is a living document, not a one-time exercise.

FAQ

How do I make AI content not sound like AI?

The short answer: give the AI a detailed voice profile instead of vague tone instructions. Analyze your best existing content for sentence structure, vocabulary patterns, storytelling habits, and tone. Document these patterns in a voice profile file. Reference it every time the AI writes for you. The result is output that matches your specific voice — not the generic AI average.

What is brand voice extraction?

Brand voice extraction is the process of analyzing your existing content to identify the specific patterns that make your writing sound like you. This includes sentence length, paragraph structure, vocabulary preferences (words you use and words you avoid), storytelling patterns, and tone. The output is a voice profile document that AI tools can reference to produce content in your voice.

Why does AI content all sound the same?

Because AI models are trained on massive datasets and default to the statistical average of everything they've read. Without specific voice guidance, they produce "average internet writing" — grammatically correct, stylistically dead, full of filler phrases like "in today's landscape" and "it's worth noting." Every AI produces similar output because they're all drawing from the same well. A voice profile breaks that default.

Can I use brand voice extraction with ChatGPT?

Yes. Create your voice profile using the method in this article, then paste it into ChatGPT's custom instructions or at the top of each conversation. It works — but you'll need to re-paste context about your product, audience, and brand positioning each session. Tools with persistent project context (like Claude Code) handle this more seamlessly since the profile stays loaded across every interaction.

How long does it take to create a voice profile?

About 30 minutes for the first version. Gather 3-5 pieces of your best writing (10 minutes), analyze them for patterns (15 minutes), compile the profile document (5 minutes). After that, you reference it forever. Update it every few months as your voice evolves.

Do I still need to edit AI content after using a voice profile?

Yes. Every time. A good voice profile gets AI output to 80-90% — which is massive compared to starting from zero. But that last 10-20% is your judgment, your specific expertise, your taste. The goal is to skip the blank page and the first few hours of generic drafting. Not to remove yourself from the process entirely.

Built by Augustin Brun — a solopreneur who spent months fighting generic AI output before building a system that actually works.

Want to see brand voice extraction in action? Get the free Direct Response Copy skill — it includes automatic voice matching from your profile. Or grab the full 16-skill marketing pack for $69 with Brand Voice Extraction, SEO Content, Email Sequences, and 13 more skills that all reference your voice.