By
Katie Shive
•
August 11, 2026
For six years, I ghostwrote for top-producing loan officers. I sat outside their offices and eavesdropped every conversation. I listened to the way they opened a call with a nervous first-time buyer versus a seasoned investor and the phrases they reached for when a deal was falling apart at the closing table. I took notes, absorbed the patterns, and wrote in their voices so completely that I disappeared into the work.
Looking back, I was doing by hand what AI does now. Absorbing a person's voice, their language, and their stories, until I could produce content that sounded exactly like them. So when loan officers ask me why their AI content sounds hollow, I know exactly where it breaks.
It breaks because most people hand the machine a blank slate and ask it to sound like someone. It can't accomplish this task because it has nothing to work from except the average of everything ever written on the internet, which is why the output reads like everyone and no one at the same time.
Here is how to fix that.
AI without a brand foundation just scales your confusion faster. Before you type a single prompt, you need documented answers to a few questions: who you truly are, who you serve, what you believe about this business that not everyone agrees with, and the stories only you can tell. Without this context, the model fills the gap with generic filler, and generic is the one thing a referral-driven business cannot afford.
Think about it from your client's perspective. Someone Googles you at 9pm before they refer their sister to you. If what they find sounds like it could belong to any loan officer in any market, you have given them no reason to remember or choose you. Your story is the one thing AI can't fabricate, and it is also the only thing that makes the content worth reading. Foundation first, every time.
The people getting flat output are prompting like they're using a search engine. A prompt like: "Write a post about rate locks" is a recipe for a paragraph that says nothing.
The people getting output that sounds like them are prompting like they just hired a sharp assistant who knows nothing about them yet. They feed context before they ask for anything: 'here is my audience, target client, and a story from a closing last week' is how I actually talk about rate locks at my kitchen table. Then they ask for the draft.
The rule is simple. Whatever you put in front of the model before you ask it to write is the difference between your voice and a stranger's voice. Thin input, thin output. Rich input, something that naturally sounds like you.
This is where most people quit too early. They read a draft, decide it's fine, and hit publish. Fine is the problem. It’s also the exact type of AI content that no one remembers.
Read every draft out loud. If a sentence sounds like a brochure or a LinkedIn post from 2019, don’t stop there. Cut the hollow authority words that sound impressive and mean nothing. Replace the vague claim with the specific scene. For example, "You're exhausted" does nothing. "The kind of tired that doesn't go away on the weekend" lands, because your reader has lived it.
Your job in the edit is to put yourself back into the words. The catch in your voice, the opinion you're a little nervous to say out loud, the detail only someone who does this work would know. That is the part the machine misses, and it is the only part that matters.
Nobody follows you for information. Information is free and the models produce infinite amounts of it. People follow you for perspective, and the fastest way to carry perspective is a real story.
You have hundreds of them: the buyer who cried at the closing table; the deal you saved at 6pm on a Friday; and the moment you decided to leave a bigger shop and build your own book. Those specifics are believable in a way no statistic ever will be. When you feed a real story into your AI session and let the model help you shape it, you get content that is both efficient to produce and impossible to confuse with anyone else's voice.
Here is my hill, and I'll die on it. Never outsource your likeness. No AI video of your face, and no cloned version of your voice reading a script you didn't say. The moment a client learns you doubled yourself, trust dies, and trust is the entire business.
There is a real distinction worth holding onto. A well-built AI system writes in your words, and is trained on your natural voice and stories. A synthetic version of you pretending to be you is a liability. One helps you show up more consistently as yourself, while the other replaces the reason anyone trusted you in the first place.
AI is not the shortcut it was sold to be, and anyone promising you ‘one prompt and done' is selling a finish line that doesn't exist. The real work is slower and more intentional. Get your brand foundation right, prompt with strong context, edit like it carries your name because it does, and lead with the stories only you can tell.
Do that, and AI stops sounding like AI. It starts sounding like you, on your best day, with more time back in your week to actually close loans.
Katie Shive is a brand strategist and AI consultant who helps mortgage professionals build personal brands that sound like them instead of the algorithm. After six years ghostwriting for top-producing loan officers, she founded KS Marketing to help advisors document their voice and build AI content systems on top of a real foundation, brand before AI. She hosts the AI After Hours podcast and works with loan officers, real estate agents, and service-based entrepreneurs who refuse to outsource their voice. Learn more at katieshive.com.
