Part of AI Gold Rush Series 1 of 3

I’ve been using AI tools long enough to know the difference between what they promise and what they actually deliver.

The promise is that AI saves time. The reality, for most people, is that it creates a different kind of work: the work of fixing output that sounds like everyone and no one at the same time. Polished but hollow. Structured but soulless. You can spot it in about three seconds.

I used to think that was just the limitation of the technology. I was wrong. The limitation wasn’t the AI, it was how I was feeding it.

Here’s what I figured out.

Prompts Describe. Examples Demonstrate.

For a long time, I was doing what most people do. Writing detailed prompts. “Write in a conversational but strategic tone. First person. No jargon. Sound like a founder.” And the output would be fine. Usable. But it never quite sounded like me.

The shift came when I stopped describing my voice and started showing it.

I was working on a newsletter and instead of writing a prompt, I shared three newsletters I had actually written and sent. Real ones. To my real list. What came back was different. Not just in tone, but in the specific way I construct a thought. The rhythm of my sentences. Where I break a paragraph. How I move from a personal observation into a commercial point. How I close.

That’s not something you can describe in a prompt. It has to be shown.

Why This Actually Works

Language models learn from examples by nature. That’s literally how they were built. Pattern recognition at massive scale. So when you give them examples of your real work at the point of use, you’re speaking their native language.

A prompt gives AI a concept to aim for. An example gives it a target to replicate. The difference in output quality is significant. A well-prompted AI gets you maybe 70% of the way to your voice. Real examples get you to 90 or 95%.

The more specific the examples, the better the result. Three newsletters in your actual voice will outperform any prompt you’ve ever written for capturing tone. A blog post, a few LinkedIn captions, an email you’re proud of. These are not just references. They are training data.

What I Do Now

Before I ask an AI to write anything in my voice, I give it context in a specific order.

First, I tell it who I am and what I do. Not a bio. A positioning statement. The role, the audience, the lane I operate in. This sets the context so the output isn’t just tonally accurate but strategically relevant.

Then I give it real examples of my work. At least two or three. I match the format to whatever I’m building. Newsletters if I’m writing a newsletter. Blog posts if I’m writing a blog. LinkedIn posts if I’m writing social content. Same format, same medium.

Then I share my guardrails. Things I never do. Things I always do. The specific patterns that make my writing mine.

Only after all of that do I give it the actual task.

The output that comes back isn’t perfect. I always review and adjust. But the gap between what it produces and what I’d actually publish has gotten dramatically smaller. That’s the real time savings. Not AI doing the work for me. AI doing a first draft I can actually work with.

The Bigger Principle

This applies to more than newsletters and blog posts. The same logic works for sales emails, pitch decks, client proposals, and social captions. In every case, you’re doing the same thing: replacing vague instructions with concrete evidence. The AI doesn’t need you to describe your voice. It needs you to prove it.

Most people are prompting their way to generic. A small adjustment, leading with examples instead of instructions, is the difference between content that sounds like you and content that sounds like a polished version of everyone else.

Your voice is your most underutilized asset. The AI doesn’t dilute it. How you use the AI does.

This is the first piece in a series I’m writing called 2026 Is the AI Gold Rush and How Real Operators Are Navigating It. Not the hype version. The real version. What I’m actually building, what’s working, and what isn’t. Next up: how I built my own CRM using AI and why I stopped paying for one.

If this resonates, follow along at stephpliha.com or subscribe to the Leadr. podcast where I get into these conversations in more depth.