I spent three years sitting outside the offices of two top-producing loan officers, listening. How they opened with a nervous first-time buyer versus a seasoned investor. What they reached for when a deal fell apart. The story they told at an open house that they would never tell at a builder meeting.
Then I wrote as them. Emails, speeches, scripts, social.
Then three more years developing voice, story, and content for close to fifty loan officers across a district at Guild Mortgage.
Then I left, and went to use AI for my own content, and had nothing to give it.
That is the embarrassing part. I had built voice documentation for nearly fifty people and had never once done it for myself. I had a LinkedIn profile I had not updated in years. Great look for a brand strategist.
So when your AI content doesn’t sound like you, I would bet money the prompt is not what’s broken.
A better prompt can only work with what you already handed over
I love a good prompt. Clear instructions matter, and most people write vague ones.
But prompting has become the answer to every AI content problem. Generic output, better prompt. Wrong tone, better prompt. Doesn’t sound like you, add three more adjectives.
So people stack instructions. Make this warm but authoritative. Write conversationally. Make it authentic. Don’t sound salesy. Use my voice.
Then they wonder why it still reads like a polite stranger wrote it.
Those instructions describe the effect you want. They do not teach the tool how you produce that effect. “Warm” means humor to one person and reassurance to another and remembering the tiny detail somebody mentioned three weeks ago to a third. “Confident” might mean blunt, or calm, or explaining something so clearly that nobody in the room feels stupid.
Your voice lives in those patterns. No adjective carries them.
Before you read the rest of this, open a voice memo on your phone. You are going to need it in about four minutes.
Why doesn’t AI content sound like you?
AI content sounds generic because the material it was given was generic. Most people hand the tool the same polished answer they give on a discovery call, which is the version built to sound professional rather than the version that is true. The fix is upstream of the prompt. Document the beliefs, stories, and language sitting underneath the polished answer, then let the prompt do the job prompts are actually good at.
The context you would hand it is the same context you hand everyone
Here is where the “just give it more context” advice falls apart.
I asked a client once how she defines wealth. She gave me a three-sentence answer about generational planning that sounded like it came off a financial advisor’s website. Perfectly good answer. Completely useless.
So I asked again. And again.
By the fifth time, she told me about her dad passing away when she was fourteen. About buying a condo three days after her eighteenth birthday with funds he had set aside for her. About graduating four years later and finding out the trust had been embezzled and the condo was the only thing left.
Her first answer was not a lie. It was rehearsed. It was the thing she had said so many times she no longer heard herself saying it.
And it is exactly what she would have pasted into a context document if I had emailed her a template and asked her to fill it out.
That is the whole problem with treating this as a data entry task. You cannot prompt yourself. The material worth giving AI is the material you have never written down, and the reason you have never written it down is that nobody has asked you the question more than once.
Meanwhile the risk sits entirely on you. Bad content does more damage to a referral-driven business than no content does, which is why so many excellent people have gone quiet instead. You are not avoiding this because you are lazy. You are avoiding it because your name is on it.
Five things to write down before you write another prompt
For each one, answer it. Then answer it again. Then a third time, a fourth, a fifth. The fifth answer is the one worth keeping, every single time.
This is less fun than downloading a prompt pack. It also works.
1. What you believe that your industry keeps getting wrong. AI has access to more general information than you could put in a prompt. What it does not have is your read on it. Write the sentence you would get pushback on at a conference. If nobody would argue with it, it is a fortune cookie and you can throw it out.
2. The three stories you already tell out loud. Not the polished origin story. The ones you reach for without planning them: how you got into this, the client you still think about, the mistake that changed how you operate, the example you use every time somebody asks that one question. Those stories teach the tool what you notice, which is something no tone instruction can do.
3. How you write when you are not writing content. Give it real samples, not samples you manufactured to train it. Client emails. Voice note transcripts. The thing you typed fast at 10pm because you were thinking about the person receiving it. That material holds the patterns you cannot describe about yourself: your sentence length, whether you teach with stories or analogies, how direct you get when you disagree, where humor shows up, what you never say.
4. Who is in the room. You already do this in real life. You explain a rate lock differently to a nervous first-time buyer than to a builder you have worked with for a decade. A demographic sketch is not enough. Write what that person already knows, what they are tired of hearing, what they are quietly worried about, and what you want them to understand by the time you are done.
5. What you refuse to do the way everyone else does it. Thousands of people share your job title. What does experience make you notice that someone newer would miss? What do your best clients come back for? Voice is not only how you write. It is what you have earned the right to say.
Do the four-minute version this week
Pick number one. Open the voice memo you started earlier. Answer the question out loud, then keep going past the point where you have already said something reasonable. Do not stop at the first answer, because the first answer is the discovery call answer.
Transcribe it. Paste it into your AI tool as standing context, not as a one-time prompt.
Then ask it to write something you would normally write, and read the output next to what you got last month.
One of my clients, Matt Kovach, was hesitant to use AI for content at all because he had seen the obvious version everywhere. His words: “I didn’t want to get lumped into that category.” So we gave the tool a real foundation. His voice, his values, his stories, how he wanted to show up.
What he said afterward: “The content it created didn’t sound like me. It was me!”
That is the bar. Not AI inventing a personality for you. Not a synthetic version of you, which I will die on a hill about separately. Just a tool that already knows the context, so you can have a half-formed thought and move instead of re-explaining your entire business every time you open a new chat.
Frequently asked questions
Can’t I just upload my old posts and let AI learn my voice from those?
It helps with sentence patterns and it is a reasonable starting point. The catch is what those posts already are. If you wrote them in the safer, more professional voice most people default to online, you are training the tool on the version of yourself you do not like. Old client emails and voice notes are better source material than anything you published.
How much context is enough?
Enough that the tool knows what you believe, the stories you tell, who you are talking to, and how you sound when you are not performing. That is usually somewhere between fifteen and thirty pages of real material, documented once. It is not a longer prompt. It is a foundation the prompt sits on top of.
Isn’t this just a brand book?
A brand book is the artifact. The work is the conversation that produces it: someone asking the same question five different ways until the real answer comes out, because somebody eighteen years into their career has usually forgotten the actual reason they do this. The document is what makes the answer reusable by every tool you own.
Do I have to redo this for every AI tool?
No. Documented once, it ports. The same foundation loads into ChatGPT project instructions, Claude project knowledge, or a custom skill, and it survives the next model release. That is the argument for doing the foundation work rather than chasing whichever tool is winning this quarter.
The order matters more than the tool
Brand before AI is not a slogan I made up to sell something. It is the order I learned backwards, after six years of making other people visible and staying invisible myself.
Your brand gives the tool something real to carry: the beliefs, the language, the stories, the audience, the patterns that make you recognizable. Then the prompt does what prompts are good at. Write the email. Turn this into a post. Pull the strongest point out of this voice note.
So stop trying to prompt your way to a voice you have never written down. Go get the fifth answer out of yourself.
And if you sat down to write number one and could not finish the sentence, that is not a discipline problem. That is what the ninety minutes are for. The Brand Builder interview is me asking, you talking, and you walking out with the thing your AI has been missing this whole time.
Comments +