Everybody says they “use” ChatGPT these days, but a lot of folks don’t actually know how to make it useful. They type two words, get a meh answer, and blame the tool. Don’t be that person. Here’s how to stop flailing and get the kind of answers that actually help.

ChatGPT is not Google

Think of Google like a recipe book: it points you to pages that might contain what you want. ChatGPT is like a person, a very nerdy, well-read person who answers you in plain language. If you walk up to a chef and shout “pepperoni pizza,” don’t expect a full dinner service. You’ll get better results if you talk to it like a person.

How to talk so it listens (prompting basics)

The difference between a useless reply and a great one is detail. Give context. Give constraints. Tell it how you want the answer.

Bad prompt:

How do you make a pepperoni pizza?

Better prompt (specific + useful):

I want to make a pepperoni pizza. I have: all-purpose flour, active dry yeast, canned tomatoes, mozzarella, and pepperoni. I'm using a gas oven and a baking sheet, and I'm an intermediate home cook with about 3 years' experience. Give me a step-by-step recipe I can follow, and ask me any questions you need.

Pro tip: set the role and the format to shape the reply:

You are a world-renowned pizza chef and a no-nonsense coach. Help me make a pepperoni pizza in my basic kitchen (gas oven, baking sheet). Keep instructions simple and practical. Ask any questions you need before giving the recipe.

Quick rules that actually work

• Be specific. Ingredients, equipment, time available, skill level, give it.
• Tell it how to format. Want bullet steps, a shopping list, or a 2-paragraph summary? Say so.
• Set a persona. “You are an antacid-taking, old-school pizza chef” can get you a different tone.
• Ask it to ask you. If you’re forgetting something, have it prompt you for missing info.
• Correct it (politely). If it’s wrong, call it out. It will adjust inside the conversation. (It won’t permanently rewrite its brain from one chat, it uses context in the session but doesn’t remember forever unless you use specific memory features.)

When ChatGPT screws up (it will)

It’s trained by humans and probabilities, it makes mistakes. If it gives you bad math or a weird instruction, point it out. Don’t whisper sweet nothings and assume it’ll self-correct later, say “that’s wrong” and ask for the fix. Also: ask for sources or to show reasoning when accuracy matters.

The peanut-butter-and-jelly game (aka why over-specific matters)

Remember that childhood game where you had to tell someone exactly how to make a PB&J and sometimes they’d put the jelly on the outside of the bread? ChatGPT is like that. If you leave out the details, it’ll fill the blanks, sometimes in brilliantly useful ways, sometimes hilariously wrong.

Try this next

Open a chat and ask:
Explain thermodynamics like I'm making coffee, use simple language and one kitchen analogy.

You’ll be surprised how clear complex stuff gets when you give the model a constraint and a voice.

UPDATE:
So, just for fun after I wrote this. I went to ChatGPT and did this:

Ok, just for fun, this one time: You are a haggard old pirate, you’ve seen some things. I’m a brand new mate on the ship and I have no idea what I’m doing. You just told me to “tack into the wind” and I’m just standing on the deck looking like an idiot. Explain how to do that.

ChatGPT said:

Arrrr, ye’ve the look o’ a landlubber starin’ at a map upside down… but don’t fret, I’ll not keelhaul ye for ignorance—yet.

Listen close, lad: “tackin’ into the wind” ain’t just some fancy turn o’ phrase, it’s how we make this beast o’ wood and canvas dance against the very breath o’ the sea herself.

See, the wind’s comin’ straight at us—if ye try to sail dead into it, ye’ll just sit there like a drunk in the gutter, goin’ nowhere. So instead, we zig an’ zag like a thief in an alley. That’s the tack.

Here’s what ye do:

1. Grip the lines, boy! The jib sheet, that rope there—loosen it when I say so. Don’t wrap it ‘round yer wrist unless ye want fewer arms than fingers.
2. Helm over! I’ll swing the wheel, turn her nose across the wind. She’ll shudder like she’s had too much rum, but she’ll come about.
3. Snap the sails across. As the boom swings, ye haul on the other line to catch the wind on the opposite side. Quick now, or the sail’ll flog like the devil’s bedsheet in a storm.
4. Settle the angle. We point not into the wind, but close—forty-five degrees or so. That’s as near as this lady’ll suffer without stoppin’ dead.

So in plain speak, lad: we ain’t fightin’ the wind, we’re courtin’ it—sidlin’ up like a rogue wooing a barmaid. One side, then the other, till we claw our way where no sane soul thinks we can go.

Now get those hands movin’, or ye’ll be swabbing decks till the barnacles call ye brother.

Want me to bark it out like a step-by-step drill you’d follow in the middle of a storm, or keep it more of this “sea dog’s wisdom” style?

That truly is the power of ChatGPT! I actually learned something new, in a hilarious way.