The Brand Voice Document That Stopped My AI From Writing Garbage
A complete system for making AI actually sound like you—with templates you can use
On Tuesday, I gave you a cheat sheet. A little “Brand Brain” snippet you could paste at the top of your prompts to make AI sound less like a robot.
Today I’m giving you the whole playbook.
Because here’s what I learned the hard way: that snippet works great... for about a week. Then the AI starts drifting. It forgets the rules. It sneaks in a “premium quality” when you weren’t looking. It gets cheerful when you needed calm.
I spent months chasing my own tail. Rewriting prompts. Adding more rules. Deleting rules that contradicted other rules. My “Brand Brain” file looked like a crime scene—sticky notes on top of sticky notes.
So I built a system. A complete Brand Voice document that I paste once and forget. It has guardrails. It has a tone dial I can turn up or down. It even has a one-line test I use to make the AI grade its own work.
This is the exact system I use for Waddles of Joy. And I’m going to show you how to build your own.
Good news: Everything in this post works with the free versions of ChatGPT, Gemini, and Claude. You don’t need a paid account to use any of it.
Why a Single Snippet Isn’t Enough
That “Brand Brain” snippet from Tuesday? It’s a good start. But it’s missing three things:
An Anti-Persona. The snippet tells AI what you want. It doesn’t tell AI what to avoid. Without guardrails, AI will drift toward generic marketing-speak every single time.
A Tone Dial. “Warm and friendly” isn’t the same energy for a shipping delay email and a holiday product launch. You need a way to adjust the temperature without rewriting your whole prompt.
A Quality Check. How do you know if the output is actually good? You need a test—something quick you can run to catch problems before you hit publish.
The Brand Voice System includes all three. Let me show you how it works.
Part 1: The Core Voice Block
This is your foundation. It’s the expanded version of Tuesday’s snippet—the thing you paste at the top of every request.
Here’s mine for Waddles of Joy:
You are writing in the voice of Waddles of Joy, a penguin-themed brand that spreads joy through whimsical, thoughtfully crafted products.
Your tone is playful but never silly, warm and human, and genuinely kind. You use gentle humor and light penguin wordplay sparingly, always prioritizing sincerity and clarity.
Every message should feel like it was created by a real person who cares—never corporate, never pushy, never gimmicky. The writing should feel cozy, inclusive, and emotionally inviting, while still communicating quality, care, and intention.
Avoid hype, slang, sarcasm, or trend-chasing language. Favor warmth over cleverness, clarity over flair, and joy over sales pressure.The reader should leave feeling seen, appreciated, and a little happier than before.
What makes this work:
It describes a person, not a style. (”Real person who cares” is clearer than “professional but friendly.”)
It includes emotional goals. (”Leave feeling seen, appreciated, and a little happier.”)
It has specific prohibitions. (”Never corporate, never pushy, never gimmicky.”)
How to write yours:
Start by answering these questions:
If your brand were a person, how would a friend describe them?
How should a customer feel after reading your copy?
What’s the one thing you never want your brand to sound like?
Write those answers in plain English. That’s your Core Voice Block.
How to Actually Use This (The Mechanics)
If you’re new to AI tools, you might be wondering: “Okay, but where do I put all this?”
Good news: it’s simpler than it sounds.
The Basic Method (Works Everywhere)
Open ChatGPT, Claude, or Gemini in your browser (the free version works fine)
Start a new conversation
Paste your Brand Voice instructions at the top of the chat window
Add your specific request below it
Hit enter
That’s it. The AI reads everything in order—your voice instructions first, then your task—and generates output that follows both.
What it looks like in practice:
[YOUR BRAND VOICE BLOCK - paste the whole thing]
[YOUR ANTI-PERSONA - paste this too]
Task: Write a product description for a penguin coffee mug with a blue handle.You paste all of that into the chat window as one message. The AI treats everything above “Task:” as context and instructions.
Save it somewhere easy to grab. Keep your complete Brand Voice document in a notes app, a text file on your desktop, or a bookmarked Google Doc. When you need it: open, select all, copy, paste into the AI chat, add your task at the bottom. Ten seconds.
One Conversation = One Context
Here’s something important: AI tools don’t remember between conversations. When you start a new chat, it’s a blank slate.
That means you paste your Brand Voice block at the start of every new conversation where you need it. If you’re doing five product descriptions in a row, paste it once at the top, then keep adding tasks in the same chat. The AI will remember your instructions for that entire conversation.
But tomorrow, when you start a fresh chat? Paste it again.
Coming soon: In future posts, I’ll show you how to make this permanent—using ChatGPT’s Custom Instructions, Gemini Gems, or Claude Projects so you never have to paste again. But those features require paid accounts. For now, the copy-paste method works perfectly and costs nothing.
Part 2: The Anti-Persona (The Secret Weapon)
This is the part most people skip. And it’s the part that makes the biggest difference.
The Anti-Persona tells AI what you are not. It’s a list of everything your brand should never sound like—and it works like a force field against generic output.
Here’s mine:
You are NOT:
- A corporate brand
- A big-box retailer
- A hype-driven marketer
- A social media influencer
- An ad agency pitch deck
Avoid completely:
- Buzzwords and marketing clichés
- Sales pressure language (”must-have,” “buy now,” “don’t miss out”)
- Trendy slang or internet speak
- Sarcasm, snark, or irony
- Excessive emojis
- Over-polished or robotic phrasing
Rewrite anything that feels:
- Loud
- Clever-for-the-sake-of-clever
- Impersonal
- Like it could belong to any brandWhy this works:
AI models are trained on mountains of marketing copy. Their “default voice” is the average of every product description, sales email, and landing page on the internet. That average sounds like a corporate press release wrote a blog post.
The Anti-Persona pushes back. It tells the AI, “I know you want to write ‘elevate your morning routine.’ Don’t.”
How to write yours:
Think about the brands you don’t want to sound like. The competitor whose copy makes you cringe. The influencer whose posts feel fake. The big-box store that calls everything “curated.”
Write them down. Then add the specific words and phrases they use. That’s your Anti-Persona.
Part 3: The Tone Slider
This is my favorite part of the system.
Not every piece of copy needs the same energy. A shipping delay email should feel calm and reassuring. A holiday product launch can be more playful. A return policy should be clear and neutral.
Instead of rewriting your voice instructions every time, you use a Tone Slider—a scale from 1 to 5 that tells the AI how much “whimsy” to dial in.
Here’s how I set mine up:
Tone Levels (specify in every prompt):
1 — Calm & Cozy
Gentle, reassuring, minimal charm. No humor required.
Use for: policies, care instructions, shipping notices
2 — Warm & Friendly (Default)
Conversational and kind. Subtle charm, no puns required.
Use for: most product descriptions, thank-you notes
3 — Playful Whimsy
Gentle penguin charm. Feels cheerful but grounded.
Use for: giftable items, seasonal products
4 — Cheerful & Gift-Ready
Joy-forward and festive. Still sincere and controlled.
Use for: holidays, collections, launches
5 — Max Whimsy (Use Sparingly)
Most playful allowed. Still sincere, never chaotic.
Use for: social captions, special announcementsHow to use it:
When you write a prompt, just add: “Tone Level: 2” (or whatever you need).
The AI now knows exactly how warm, playful, or restrained to be—without you having to rewrite your voice instructions.
Example in action:
Same product. Two different tone levels.
Tone Level 2 (Warm & Friendly):
“This ceramic mug features a penguin in a cozy yellow scarf, ready to keep you company on chilly mornings. Dishwasher and microwave safe.”
Tone Level 4 (Cheerful & Gift-Ready):
“Meet your new morning companion: a cheerful penguin bundled up in a sunny yellow scarf, ready to make every cup of coffee feel like a snow day. Perfect for gifting (or keeping for yourself—we won’t tell).”
Same voice. Different energy. No rewriting required.
Part 4: The One-Line Checksum
Here’s something that used to drive me crazy: AI would generate something, and I’d think, “This is... fine? Maybe? I don’t know.”
I didn’t have a test. I was just vibing.
So I wrote a single question that cuts through the ambiguity:
Does this sound warm, human, and thoughtfully joyful—like a kind person who cares, not a marketer trying to sell something?
That’s it. If the answer isn’t an immediate yes, I revise.
The Pro Tip:
Here’s the trick that changed everything for me: use the checksum as a follow-up prompt.
After the AI generates something, I paste this as my next message:
Review the text you just wrote. Does it pass this test: “Does this sound warm, human, and thoughtfully joyful—like a kind person who cares, not a marketer trying to sell something?” If not, rewrite it.You’d be surprised how often the AI catches its own mistakes. It’s like giving your intern a checklist before they hand in their work.
⸻
Putting It All Together: A Full Example
Let’s walk through the whole system with a real product.
The Product: A 15 oz stemless wine glass with festive holiday penguins.
Step 1: The Prompt
[PASTE CORE VOICE BLOCK]
[PASTE ANTI-PERSONA]
[PASTE TONE-SLIDER]
Tone Level: 4 (Cheerful & Gift-Ready)
Task: Write a Shopify product description for a 15 oz stemless wine glass featuring festive holiday penguins. Include a headline and 2-3 sentences of description. Focus on the gifting angle.Pro tip: If you have a product photo, you can paste or drag the image directly into ChatGPT, Claude, or Gemini along with your prompt. The AI will “see” the product and pull details from the image—colors, patterns, text on the item—so you don’t have to describe everything yourself. This works in the free versions of all three tools.
Step 2: The Output
Raise a Glass to the Season
These cheerful penguins are ready to celebrate—and they’d like you to join them. This 15 oz stemless wine glass brings a little waddle to your holiday toasts, whether you’re sipping mulled wine by the fire or gifting it to your favorite penguin lover.
Step 3: The Checksum
I paste: “Does this pass the test: ‘Does this sound warm, human, and thoughtfully joyful—like a kind person who cares, not a marketer trying to sell something?’”
AI responds: “Yes, this passes. The tone is warm and gift-focused without being salesy. The phrase ‘a whole lot of holiday charm’ is playful but not over the top.”
Step 4: Done.
Total time: maybe 90 seconds. And I didn’t have to explain my brand voice from scratch.
Your Homework
I’ve attached two templates to this post:
My full Waddles of Joy Brand Voice document — so you can see how all the pieces fit together.
A blank template — so you can build your own.
This weekend, try this:
Write your Core Voice Block (answer the three questions I gave you).
Write your Anti-Persona (who do you NOT want to sound like?).
Pick three Tone Levels that fit your business.
Write your One-Line Checksum.
Then test it. Run one product description through the system and see how it feels.
Reply to this email or comment below and tell me: What’s the hardest part of describing your brand’s voice? I’d love to help you crack it.
See you there.
— Ann
The Automated Creative is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.