If you built a brand voice file and Claude still doesn’t sound like you, the problem isn’t your brand voice file. It’s that the AI still doesn’t have enough context to fill in the “blanks” around how you think.
I watched 7 different “make AI sound like you” walkthroughs, and read a 21,000-word style guide from Tiago Forte that is the most sophisticated version of this I have found anywhere. All of them teach the same thing. None of them teach the piece that actually fixed my output.
Same Prompt; Different Context
I even ran a test. (you can see it live on the video) Same prompt, same brand voice file but in the second run I added an additional file. The output was two completely different pieces of content.
The first one could have been written by any consultant who does what I do. It contained this line: “AI is a mirror, and most people are looking into it and seeing a stranger.” I would never say that. It doesn’t even make sense.
The second one had my thinking all over it. It talked about AI as a word prediction engine, which is a thing I say constantly. It referenced the AI Marketing Assistant I build with clients. It used my 10-80-10 framework — 10% you on the front end, 80% letting the AI work, 10% you cleaning up the back end. It got the closing steps wrong, but it was in the neighborhood.
One file separated those two outputs.
Why Your Brand-Voice File Isn’t Good Enough
Open your brand voice file and actually look at what’s in it. Tone. Cadence. Do’s and don’ts. Emoji policy. Em dash policy. Word choice.
At its core, it is a list of rules.
That document recreates your voice patterns. It tells the model nothing about how you think. So when the model needs an actual opinion or a reason or a because – it goes looking in its training data and returns the average next word. A synthesis of what everyone in your field is already saying, delivered with your punctuation.
Picture one of those survey distribution graphs. The fat middle is what most people answer. That is what Claude writes by default. Then there are the responses out on the tails…the outliers. What you want is to feed Claude your personal outliers.
Use My POV Framework to Feed Claude Your Personal Outliers
I call the missing file POV (point of view). It lives in my context folder right next to my brand voice file, and I have every client build one.
It has three components.
Frameworks. Your systems and processes. If you have 20 years of experience, you have a framework, you may just have never written it down. Have AI interview you until you can see it, or work with someone who can pull it out of you. Either way, give it a name. For some reason the AI platforms love a named framework.
Opinions. Your strategic stances. Your contrarian views. These are the things you would say to a smart peer who disagreed with you. If everyone in your field already agrees with it, nobody needs your version of it. Be brave here.
Stories. The lived experience underneath the opinions. This is what converts a hot take into something people pay attention to.
Those three together do three jobs. They build real trust, because they show expertise instead of claiming it. They make your content connect – every piece points back to the same core ideas, which is what turns scattered posting into a strategy that moves people toward hiring you. And they put your material in exactly the categories that get you found and recommended by AI search.
What this file is not: your bio, your credentials, or beige statements like “I put clients first” and “I have a bias toward action.” Garbage in, garbage out.
How to build a POV.md file in 20 minutes without writing anything
You do not have time to write out 20 stories, and writing is the wrong capture method anyway. When you type, you polish. You edit out the raw specifics that make it interesting.
So talk instead.
- Right now, 15 minutes. Open Claude and paste this: This is my [niche] and this is my [audience]. Ask me one question at a time about opinions I hold that other people in my field would disagree with. Then ask me to tell you a story about where that opinion came from.
- Answer out loud. Voice dictation on your laptop, the voice memo app on your phone, Wispr Flow, whatever you have. Talk, don’t type. Three questions is enough to start.
- Same sitting, 5 minutes. Have Claude compile your answers into a document called pov.md. Put it in the same folder as your brand voice file. If you are working in the browser instead of Cowork, upload it to the project where you do most of your writing.
- Ongoing. Add two or three more questions whenever you are on a walk. The file gets stronger every time.
Next Steps
Once you’re getting better content drafts with Claude, the next step is to create a content strategy that builds trust when your potential clients check you out online. Want to pinpoint your biggest content opportunities? Get your free Online Signal Report + AI Visibility Score.
