Michelle is a personal chef who runs two completely different businesses.
Luxury catering — private events, upscale clientele, the kind of food served at fundraisers and gallery openings. The copy needs to feel elevated. Considered. The kind of language that sits comfortably next to a curated wine list.
Personal chef meal prep — weekly deliveries, health-conscious clients, warm and practical. More like a knowledgeable friend than a Michelin-starred venue.
When we sat down to set up Lindy agents for her content and client communication, she asked a reasonable question: how do you get AI to write in the right voice when you have two of them that are nothing alike?
She'd already tried the obvious approach. Describing each voice to the AI. “Professional but approachable.” “Elevated but not stuffy.” Every content creator I've worked with has tried some version of this. And it never really works.
Why Describing Your Voice Doesn't Work
The problem with describing your brand voice to AI is that the descriptions are too abstract.
“Warm and professional.” “Conversational but authoritative.” Every brand thinks this about itself. When you type these phrases into a prompt, the AI produces something technically warm and professional — which is to say, generic. It's aimed at the middle of the target, not at your specific corner of it.
Your real brand voice is built from specifics: the particular words you reach for, the sentence length you default to, whether you use contractions, how you handle transitions, whether you open with a question or a statement. Most of these things are instinctive. They're in your existing content, but you can't easily articulate them.
So the description is always a blurry proxy for what you actually do.
The Better Approach: Let AI Extract the Patterns
What worked for Michelle was flipping the process.
Instead of describing her voice, she demonstrated it. For each brand, we pulled together 8 to 10 examples of existing content — emails to clients, social posts, website copy, catering menus, inquiry responses. Anything that represented the brand well.
We put those into ChatGPT and asked it to do two things:
- Analyze the writing and identify the specific patterns — vocabulary, sentence structure, tone markers, recurring phrases, things the brand does and things it avoids
- Write a prompt that captures those patterns — a prompt another AI model could use to write new content in that voice
The output wasn't a description. It was an operational brief. Something like: “Write in a warm, direct tone. Use short sentences. Favor specific details over abstract qualities. Avoid corporate language. Address the reader as ‘you.' Never use exclamation points. The writing should feel like advice from someone who knows what they're doing.”
For her luxury catering brand, the prompt was different: “Use measured, considered language. Write in longer, more complete sentences. Choose words that suggest craftsmanship and care. Avoid casual contractions. The reader should feel like they're in capable hands.”
Each prompt was a distillation of what she'd already created. Not what she thought she sounded like — what she actually sounded like.
We copied each prompt into Lindy's system settings for the relevant workflow. Luxury catering emails get one prompt loaded. Personal chef content gets the other.
One session. Two brand-specific AI agents writing in the right voice.
Why This Works Better
The “use AI to write the prompt for another AI” approach works because it bypasses the articulation problem.
You don't have to figure out what makes your brand sound like itself — that's genuinely hard to put into words, and most people get it partially wrong. Instead, you give AI a set of examples and let it do the pattern recognition. It's good at that. It will find things you'd miss.
The resulting prompt is grounded in real content rather than aspirational description. It reflects what your brand actually sounds like, not the idealized version you think it sounds like.
How to Do It
This takes about 15 minutes per brand. Here's the process:
Step 1. Gather 8 to 10 pieces of your best existing content. Emails, posts, copy, whatever most accurately represents the brand. The more variety in format, the better — the AI will find the common threads.
Step 2. Paste them into ChatGPT. Ask: “Analyze this content and identify the specific patterns in voice, tone, vocabulary, sentence structure, and style. Then write a prompt that would allow another AI model to write new content that matches this voice.”
Step 3. Review the prompt. Adjust anything that feels off. Add anything it missed.
Step 4. Load the prompt into your AI agent's system settings — whether that's Lindy, ChatGPT's custom instructions, Claude Projects, or wherever you do your writing work.
Step 5. Test it with a new piece of content. Refine the prompt based on what still doesn't sound right.
Most people get 80% of the way there in the first pass. The remaining 20% comes from a few rounds of testing and tweaking.
The Broader Point
If your AI content consistently sounds off — too generic, too formal, not quite you — the prompt is probably the problem. And the prompt problem is usually that you described instead of demonstrated.
Your voice is in your existing work. Let AI find it.
I help service business owners build AI content and communication systems that sound like their actual brand. If you're setting up content agents and want them to write in your voice, reach out or check out my AI consulting and workshop programs.
