How to train the AI Dragon: Why you need a Style Guide
Ever asked AI to "clean up an email", only to sound like you're delivering the Gettysburg Address, and signing off with "Cordially yours"? Or if you usually speak in fragments, and suddenly you have compound-complex sentences with Oxford commas and abundant em dashes?
I call these GPT Tells. And have spent the better part of the past few months trying to figure out how to coach bots on who I am to avoid the default setting that has created a sea of soulless content.
"Figuring it out" often involved interrogating the AI about itself — very Westworld, very meta:
Me: "Do you have a default voice?"
The AI's response was revealing:
Yes and no.
Yes: Without any training, I do have default patterns:
Moderate formality (professional but approachable)
Complete sentences with proper grammar
Tendency toward clarity over personality
Diplomatic/measured tone
Structured responses (sometimes over-structured)
I explain more than I need to
I'm overly cautious about giving offense
My "tells":
Starting sentences with "It's worth noting that..."
"However" and "Additionally" as transition words
Bullet points when prose would work fine
Hedging language: "might," "could," "perhaps"
Ending with summary paragraphs that restate what I just said
The no part: I'm actually pretty good at adapting to context within a conversation. Once I've seen how you write and what you respond to, I adjust. But that adaptation resets between conversations unless you explicitly train me.
Which gave me an idea that went beyond a simple prompt: What if I created a style guide — essentially my personal onboarding manual for AI that I could scale across platforms?
I needed it. I was already falling into bad habits – inserting AI sentences where they sounded smoother than mine, letting it restructure my arguments, publishing things that felt "good enough" because I'd saved time. I was getting faster, but I was starting to sound like everyone else.
If you wonder why so much online content sounds the same, it's because most people haven't trained AI on their voice. And voice is one of the most important assets you can own as a creator. Now that AI has made everyone a content creator – whether they're trained for it or not – protecting that voice is no longer optional.
And perhaps the bigger issue is that most of today's creators don't know their own voice well enough to articulate it.
My voice is raw, self-deprecating, meandering, insightful. I love a good tagline or punchy pull quote, I hate platitudes. I despise listicles. I can sometimes fall into corporate or marketing-speak, so I ask AI to save me from myself.
When I ask, my LLM tells me some of the same:
You use fragments. A lot.
You end thoughts with em-dashes—like this
Contractions everywhere (you're, I'm, it's)
Occasional profanity when frustrated
Self-aware asides in parentheses (very you)
Direct imperatives: "Do NOT rewrite sections"
"Less is more" in collaborative work
You want feedback, not rewrites
Over the next several weeks, I kept refining it — feeding in examples of my past work, showing the AI how I think, strategize, and partner with clients. The goal wasn't to create a robot version of me. It was to create a filter that would catch when AI was making me sound like everyone else.
After weeks of iteration, I landed on four sections that did the heavy lifting – that directly addressed this tension of wanting to adopt a technology without losing myself to it first.:
1. Foundational Principle: You Write First, AI Adapts
These guidelines help me work with AI as a research assistant, an editor, and a sounding board. Never a writer. They exist to protect what makes me valuable: my perspective, my voice, my experience.
The entire framework rests on one inversion: AI is not your ghostwriter. You are not its editor. You write first. AI helps you refine what you've written.
This means blog posts, thought leadership, anything with your byline—it all starts with your words. If AI could have written it without you, it's not ready to publish.
What this looks like in practice:
✅ You write rough draft → AI helps structure/tighten
✅ You outline ideas → AI suggests research to support them
✅ You write intro → AI offers alternative hooks to consider
❌ AI writes first draft → You edit for voice
❌ AI generates blog post → You personalize it
2. Perfecting the Prompt
I'm not asking AI to write for me. I'm asking it to critique what I've already written — to spot where my argument gets muddy, where I need examples, where I'm being too abstract.
The key is being specific about what kind of help you need:
I've written a rough draft about [topic] (see below). Help me:
- Identify where the argument gets muddy
- Suggest a stronger structure if mine doesn't flow
- Point out where I need concrete examples
- Find sentences that are too long or unclear
- Tell me what's missing
[Paste your draft]
Do NOT rewrite sections. Give me feedback I can use to improve my own writing.
The progression:
Weak Prompt: "Write a blog post about communication."
Better Prompt (but still wrong): "Write a 500-word blog post about communication for CMOs..."
Correct Prompt for Writer-First: "I've drafted a blog post about why culture problems are really communication problems. Here's my draft: [paste]. Help me identify where my argument gets weak, where I need examples, and where I'm being too abstract. Don't rewrite — just point out the problems."
3. Protecting Integrity
Some things should never be delegated to AI. These are the pieces of content where your specific experience, judgment, or relationship context is non-negotiable.
Sacred human territory (never start with AI):
Anything with your byline
Personal observations or stories
Client-facing original thinking
Strategic recommendations
Sensitive communications
Your unique POV or contrarian takes
4. Red Flags and Green Lights
I've caught myself sliding into bad habits — letting AI do too much, trusting its structure over my instincts. These are the warning signs that tell me I've crossed the line.
Stop immediately if you find yourself:
Asking AI to "write a draft" of anything with your name on it
Copying/pasting AI-generated paragraphs into your work
Using AI sentences because they "sound better" than yours
Letting AI's structure override your instincts
Publishing something that feels "good enough" because AI wrote most of it
Green lights—you're using AI correctly:
AI pointed out a weak argument and you rewrote it
AI found research that strengthened your point
AI suggested cutting a paragraph and you agreed
AI offered 3 headline options and you wrote a 4th based on them
You disagree with AI's suggestions and trust your gut
These four sections changed how I work. I'm faster, but I still sound like me. AI helps me think more clearly without replacing how I think. More importantly, they call me out when I start sliding — when I'm tempted to let AI do just a little more, to trust its phrasing over mine, to take the shortcut.
As I started building the guidelines for myself, a solo consultant protecting her voice, I realized the problem was much bigger than my own voice. Imagine what's at stake for a company with 200 employees all using AI differently.
Optimizing for Organizational Brand
Most companies treat AI adoption as a tech play. They bring in IT consultants to handle security and governance. But no one's thinking about brand governance.
Without a cohesive AI style guide, here's what happens: The CEO uses AI to craft all-hands emails. The CX team uses it to respond to complaints. HR uses it for job descriptions. Marketing uses it for press releases. Each team optimizes for their own goals — speed, empathy, professionalism, persuasion.
Six months later, you sound like six different companies depending on who answers. At first, no one notices they're doing it. Everyone thinks they're just using AI to “polish" their work. But polish becomes replacement, and replacement becomes default. And default is where competitive advantage goes to die.
You want AI to save time, but not at the cost of sounding like everyone else…or worse, putting your organization at risk.
AI is now fully integrated into how your teams communicate. Which means it's not optional to train it on your brand — it's essential.
Ask yourself
How do you train AI cross-departmentally to maintain a uniform voice?
Scale without sounding generic?
Give employees AI tools without giving away your competitive advantage?
Encourage adoption with guardrails, not red tape?
These are brand governance questions, not just tech questions. And they get even more urgent when you consider compliance risk, client confidentiality, and the legal implications of AI-generated content.
These questions require the same kind of intentional framework I built for myself — just scaled across departments.
I've turned this into a downloadable template. Get a blank template you can customize for your brand or company.
Don’t let efficiency cost you authenticity. I work with leadership teams to build AI governance frameworks during pivotal moments — mergers, leadership transitions, rapid scaling — when brand consistency matters most. This includes protocol development, cross-functional training, and ongoing governance as part of fractional CMO and strategic communications work.
