So you’ve watched someone pull something genuinely good out of AI and thought, okay, what prompt do they know that I don’t. Honest answer? Probably none. It’s almost never the words they typed. It’s that they weren’t really typing. They were talking to it.

Most of us treat it like a search bar. Clipped, careful, one tidy sentence, fingers crossed. And it hands back exactly what that earns. Tidy, careful, forgettable.

Stop typing at it. Start talking to it.

This is the shift that changes everything and almost nobody makes it. Tap the voice button, or your phone’s dictation, or a tool like Wispr Flow, and say the whole messy thing out loud. Where you’re stuck. What you already tried. The part you can’t quite word yet. 90 seconds of rambling beats 1 clean sentence every single time.

And the rambling isn’t a mess to clean up first, it’s the point. When you talk, you hand it the stuff you’d never sit down and type. The aside, the second-guess, the “oh, and also.” That texture is the thing that makes an answer come back sounding like you instead of like the whole internet.

Prompt engineering was always the wrong game

The internet sells prompting like a ladder. Memorize the magic wording, unlock the better output. It was never real. A flawless prompt sitting on top of no real context still gives you polished, generic work, and it reads well enough that you won’t even catch it.

The models got good enough that the exact wording barely moves the needle anymore. What moves it is how much of YOU and your actual situation makes it in, and talking is just the fastest way to get more of that in. Think about the best conversation you’ve had with a sharp colleague. You didn’t engineer your sentences. You thought out loud, they pushed back, and you found the real question together. That back and forth is the gear your AI is best in too. It’s a friend to think with, not a form to fill out.

The people getting the most out of AI aren’t better at prompting. They talk to it like a person, and they can hear when it’s quietly wrong.

Then learn to hear when it’s off

Talking gets you a real answer. Judgment tells you whether it’s a good one. So before you even start, decide what a great response would actually have in it. 2 or 3 things a strong version nails, one fact you already know is true. Now you’ve got a ruler instead of grading on vibes.

Then read that first answer like an editor, not a fan. Fluent and correct are two different things, and AI is exceptional at sounding right. So push back, out loud. “Argue against that.” “Where is this wrong.” “What would a skeptic say.” A confident answer will sprout 3 real weaknesses in seconds. (I told mine to argue against me the other day and it came back with 3 things I didn’t want to hear. Rude. Also correct.)

And keep a clear line on what stays yours. It can draft and it can stress-test, but the final call, and the one thing only you can say, don’t get handed off. You own the reins there, always.

It’s a habit, not a hack

Getting better at AI looks less like collecting prompts and more like getting harder to fool, and a lot more willing to just talk. Every time you say the messy version out loud and catch a subtle miss, your standard climbs a little and the work you’ll accept gets better. Small but mighty, that habit.

There’s a ton of AI noise right now, a lot of experts, not many of them actually showing you how to run it inside a real business. I try to be that person.