The one technique that makes AI sound like you — and why "don't do X" prompts backfire.
1
Stop writing "don't" lists
Telling AI what to avoid — "don't use buzzwords, don't sound like AI, don't use em-dashes" — gives it nothing to aim at. It fills the gaps with its default patterns, which is exactly what you were trying to avoid.
2
Show, don't restrict
Examples teach the model what to do; rules only tell it what not to do. One is a creative brief, the other is a list of restrictions — and the brief wins every time.
3
Build a voice reference file
Collect 10-15 pieces of your own writing you're proud of — posts, emails, threads — and paste them all into one document. This is the single highest-leverage prompting move most people never make.
4
Annotate it lightly
Add a short section at the top (five or six sentences) describing your voice in plain language: "short, punchy sentences," "warm but direct," "opens with a hook." Tag a couple of the examples with why they worked.
5
Make the model study it first
At the start of a session, tell the model to read the file and study your tone, structure, and rhythm before writing anything — and to ask you a few clarifying questions first. Then let it write.
6
Reuse it every session
Upload the same file every time you start writing with AI; it should be the first thing the model reads. One good reference file beats fifty random uploads or a thirty-rule prompt.
7
The payoff
Before: ten minutes writing rules and still getting generic output. After: thirty seconds uploading one file, and the output sounds like you from the first draft.
One of 8 free step-by-step guides
Practical playbooks for actually using AI well — plus 459 prompts, 227 ideas, and 172+ tools.