Why Your First Try Was Disappointing
Most people type something like "write me a post", get back something bland and generic, and quietly decide AI is overrated.
The tool wasn't the problem. "Write me a post" doesn't say what kind, about what, or in whose voice. So it guesses all three, and a guess averaged across the whole internet comes out beige every time.
A prompt is just what you type in. And a good one has three parts.
The Three-Part Recipe
This is the exact mechanic Jerry uses when he demos AI live for an audience — they call out three things and the answer comes back tailored. It works because it removes all three guesses.
Topic
About what?
a bakery · a plumber · retirement · a lost dog · your business · coffee
Tone
In what voice?
professional · funny · dramatic · surfer · pirate · Shakespeare
Snap them together and you get a prompt that works:
Create a social post about my bakery's new sourdough in a warm, friendly style. Keep it short.
Format + Topic + Tone.
That's the whole trick.
✗ Vague
"Write me a post."
It guesses the format, the subject and the voice. You get beige.
✓ Specific
"Create a short Facebook post about my bakery's new sourdough, in a warm and friendly voice."
Nothing left to guess. You get something you can nearly use as-is.
The Follow-Up That Surprises People
Here's the part that lands hardest when Jerry does it live. Once AI has written something, you don't start over to change the voice. You just ask it to redo the same thing differently:
Now rewrite that exact same thing as a surfer.
Now as a pirate.
Now in Shakespearean English.
Now make it polished and professional.
Same content. Four completely different voices. Seconds apart. That's the moment the room goes quiet when he demos it — and it's not a party trick. It's how you find the right tone for a real customer email: write it once, ask for it warmer, ask for it shorter, keep the one that fits.
It remembers — up to a point
Within a chat, you don't have to re-explain yourself. It keeps track of the recent back-and-forth, so "make it shorter", "less formal", "add a line about our hours" all just work — each one building on what came before. Treat it like a conversation, not a search box; that habit is most of the difference between people who get value out of AI and people who don't.
Just know its memory isn't unlimited. It only holds so much of one conversation at once — in a very long chat, the earliest things you said quietly fall off the back end, and it can start to forget them. If it seems to lose the plot, remind it of the important bit, or start a fresh chat.
And here's the part worth knowing exists: that forgetting is only within a chat. The newer tools also have a separate "memory" setting you can switch on that remembers things about you from one conversation to the next — your name, your business, how you like things written. That's the key to AI having a real long-term memory instead of meeting you fresh every time. Actually setting it up is a lesson for another day — for now, just know it's there and it's worth turning on.
Skits' Tip
Your first prompt is a starting point, not a wish granted by a genie. The people who get the most out of AI aren't better at typing the perfect request — they're just willing to say "close, but shorter" three times in a row.
Quick Check
You want AI to write a thank-you note to a customer who just bought from you. Which prompt gets the best result?
Now you can ask. Next: knowing when to believe it.
Format, topic, tone, then refine by just talking to it. That's genuinely most of the skill. But getting a good-sounding answer and getting a TRUE answer are two different things, and the last chapter is about telling them apart.