Here’s a frustration I bet you know well. You spend ten minutes stacking rule after rule into an AI prompt, telling it everything to avoid, and the output still sounds like a robot wrote it. I ran into a brilliant post that finally explains why, and I was nodding along the whole time.
The breakdown comes from a sharp LinkedIn creator who watched this exact pattern play out with their own team, day after day. Their diagnosis is so simple it stings: most people prompt AI like they’re drafting a terms and conditions document. A long list of restrictions. No examples. No voice. No reference point. Just thirty rules about what the model should NOT do, followed by confusion about why every draft reads the same.
Why rules-only prompts keep failing
The original poster nails the core distinction, and it reframed how I think about prompting. Rules tell the model what to avoid. They never show it what to actually do.
So the team would load up commands like these:
- Don’t use buzzwords
- Don’t start with “In today’s world”
- Don’t sound like AI
And the output would still sound like AI. Because you can ban a hundred bad habits and the model still has no idea what your good writing looks like. Restrictions are a fence, not a map. That one realization is what changed the whole team’s approach.
The fix: a reference file instead of a rulebook
According to this contributor, the swap is refreshingly easy. Instead of piling on rules, you hand the model a reference file. Ten to fifteen pieces of your actual writing. Real posts, real emails, real threads, the ones you’re genuinely proud of.
The model reads them and picks up the things rules can never capture: your sentence rhythm, your vocabulary, how you open, how you close. Then you write a single line in the prompt:
“Apply everything you learn as a writing rulebook.”
That’s the entire instruction. The model builds the rulebook itself, from evidence, instead of you guessing at it.
The exact step-by-step build
This is where the expert gets practical, and I love that each step comes with a reason attached. Here’s the reference file they walk through, in order:
- Save 10 to 15 pieces of writing you’re proud of. Posts, emails, and threads all count. This is your raw voice data, so quality matters more than quantity.
- Paste them all into one document. Keeping everything in a single file means the model absorbs your voice as one coherent body of work, not scattered fragments.
- Label each piece with a one-line note on why it worked. Something like “High engagement, conversational tone” or “Short sentences, strong opening.” These notes tell the model which qualities to pay attention to.
- Add a short section at the top describing your voice in plain language. Five or six sentences max. This gives the model a quick summary before it studies the examples in detail.
- Upload this file every time you start a new writing session with AI. Fresh sessions forget everything, so the file is what carries your voice forward each time.
- Tell the model to study it before writing anything, and ask it to raise clarifying questions first. That pause forces the model to understand the assignment instead of guessing and rushing to a generic draft.
Six steps, each one earning its place. Nothing fancy, nothing that takes special tools.
The before and after
What sold me was the contrast the author lays out. The difference in effort and quality is almost backwards from what you’d expect.
- Before: ten minutes writing restrictions into every single prompt, and still getting generic output.
- After: thirty seconds uploading one file, and the first draft already sounds like you.
The lesson here goes beyond one trick. Showing an AI what good looks like will always beat listing everything that looks bad. Examples teach. Restrictions only limit.
A few tips to make it work harder
Building on what this savvy professional shared, here are a couple of practical notes as you try it:
- Pick examples with variety. Include a punchy post, a warm email, and a longer thread so the model learns your range, not just one mode.
- Update the file over time. As you write pieces you’re even prouder of, swap them in so your reference keeps improving.
- Actually answer the clarifying questions. That back-and-forth is where the model locks onto the specific job, so don’t skip it.
I think this is one of those small shifts that quietly changes everything about working with AI. You stop fighting the model with a wall of “don’ts” and start feeding it the one thing it was missing: a picture of your real voice.
The original poster shares the full walkthrough on their LinkedIn post, and it’s worth reading in their own words. Check it out, then go build your reference file and test it on your next draft.