Why I Started Treating GPT Like A Junior Ghostwriter
I write a lot, but I do not always feel like typing a lot.
So I started using GPT as a sort of junior ghostwriter for my blog on richardlemon.com. Not to "create content". To keep my publishing cadence up while I am busy shipping weird web experiments and coaching baseball.
The problem: every unconfigured model sounds the same. Smooth. Bland. Corporate. Like it just finished a LinkedIn course on "thought leadership".
So for three months I iterated on a single system prompt. Same idea. Lots of small edits. My goal was simple. Posts that still sound like me after the AI takes a pass at the keyboard.
This post is the current state of that system prompt. The exact text. The banned phrases. The structure rules. Plus what actually moved the needle.
Subscribe to continue reading