The biggest difference between people who are amazing with AI and people who aren't has nothing to do with prompts. It's how they talk to it. This guide gives you the four rules I use every single day, with the exact words to type, so you get better answers from ChatGPT, Claude or any AI starting today.
What you get: The one way of thinking that makes AI 10x more useful, how to give AI a goal instead of a list of steps, exactly what to say when it gets something wrong, the two words that make it do the work for you, what to do the moment you get stuck, and 6 copyable prompts.
AI Is Not Google
The plan · 5 steps
- 1Start every big task with the manager prompt.
- 2Give it the goal, not the steps.
- 3When it's wrong, say exactly what's wrong.
- 4Make it do the work for you.
- 5When you're stuck, ask it how to get unstuck.
Before you start
Your account
Any AI works: ChatGPT, Claude, Gemini. Free plans are fine.
Your materials
One real task you've been putting off. Your inbox, a trip, a budget, a project at work.
Your first goal
Get one task done today by managing the AI instead of searching it.
Most people treat AI like a search engine. They type something in, get an answer, and if it's wrong, they either give up or type "that's wrong, try again."
You have to treat AI like an employee. And not just any employee.
Think of it as your smartest new hire, on day one
Picture the smartest person you've ever hired, on their very first day. They know more than you about almost everything: writing, money, travel, code, marketing. But they know nothing about you. Not your goal, not your situation, not what "good" looks like to you.
So your job isn't to do the work. Your job is to be a good manager:
- Tell them the goal and let them figure out the best way to get there.
- Give real feedback when the first draft misses.
- Hand them the work, even the boring parts.
- Expect them to solve problems, not quit when something breaks.
That's the whole guide. The rest is the exact words to type.
Maverick's tip
This works the same in ChatGPT, Claude, Gemini or anything else. Every rule here is about how you talk, not which app you use.
The Manager Prompt
Start any big task with this. It tells the AI the goal, gives it the context only you have, and puts it in charge of the how.
My goal is [WHAT YOU WANT DONE]. Here's what you need to know about me: [YOUR SITUATION, WHAT MATTERS TO YOU, ANYTHING IT CAN'T GUESS]. Figure out the best way to do this, do as much of it yourself as you can, and only stop to ask me when you truly need me.
What a filled-in one looks like
"My goal is to plan a 5-day trip to Japan in April that I'll actually remember. Here's what you need to know about me: I'm 34, I hate tourist traps, I love food and walking, and my budget is $3,000 not counting flights. Figure out the best way to do this, do as much of it yourself as you can, and only stop to ask me when you truly need me."
Notice what's missing: no steps, no "first find hotels, then make a list of restaurants." Just the goal and you.
Maverick's tip
Not sure what context to give it? Add this to the end: "Before you start, interview me about anything you'd need to know to do this really well." It'll ask you, one thing at a time.
Give It the Goal, Not the Steps
This is the one almost everyone gets wrong.
Say you want your inbox organized. A lot of people will tell the AI exactly how: make this folder, move these emails here, set up this filter. And if you actually know exactly how you want it, great.
But most people think they know and they don't. The AI has seen more ways to organize an inbox than you ever will. When you micromanage something smarter than you, you get your idea, not its best one.
Steps vs. goals
| Instead of the steps | Give it the goal |
|---|---|
| "Make a folder called Bills, one called Work, one called Family, and move these emails into them." | "Organize my inbox in the best way possible so I can find anything in seconds and only see what matters." |
| "Make a spreadsheet with columns for rent, food and gas." | "Help me figure out where my money goes and where I could save $300 a month." |
| "Write five bullet points about our new product." | "I need customers to understand why this product is worth the price. Write whatever would do that best." |
| "Find me three hotels near the Eiffel Tower under $200." | "Find me the best place to stay in Paris for a first trip, under $200 a night." |
Then let it show you its plan. If you don't like part of it, tell it, the same way you'd tell an employee.
Here's my goal: [THE OUTCOME YOU WANT]. Don't follow my usual way of doing this. Tell me the best way you'd do it and why, then do it.
When It's Wrong, Say Exactly What's Wrong
AI gets things wrong sometimes, and that's normal. It's a first draft, the same as an employee's would be.
The mistake is how people respond. "That's wrong, try again" gives it nothing to work with, so it just guesses again. Tell it what's wrong, why, and what not to do next time.
This isn't right. [WHAT'S WRONG AND WHY]. Don't [WHAT TO STOP DOING]. What I actually want is [WHAT GOOD LOOKS LIKE]. Try again with that in mind, and remember this for next time.
Bad feedback vs. good feedback
| What most people say | What you should say |
|---|---|
| "That's wrong, try again." | "This is too formal. Don't use bullet points. Make it sound like a text to a friend." |
| "No, not like that." | "The budget is wrong because you left out rent. Add $1,800 a month for rent and redo the totals." |
| "Make it better." | "The first paragraph is boring. Start with the $195K hospital bill story, then explain the fix." |
| "These restaurants suck." | "These are all tourist spots. I want places locals go, under $30 a person, within walking distance of my hotel." |
When it keeps making the same mistake
Ask it to fix the cause, not just the answer:
You've made the same mistake a few times now. Tell me why you think it keeps happening, then write yourself a rule so it doesn't happen again, and redo it following that rule.
Maverick's tip
If a chat has gone way off track, don't keep fighting it. Ask it to "summarize everything we've figured out so far into one prompt," then paste that into a brand new chat. Long, messy chats make AI worse.
Make It Do the Work
This one is so common. The AI says "first, you'll need to install this," or "go into your settings and change this," and people just go do it themselves.
Your answer should almost always be "do it for me."
What to say back
| When the AI says | You say |
|---|---|
| "You'll need to install this software first." | "Install it for me." |
| "Go into your settings and change this." | "Change it for me." |
| "Here's a template you can fill in." | "Fill it in for me with what you know about me." |
| "You could email them and ask for a refund." | "Write the email for me, ready to send." |
| "Here are the steps to set it up." | "Set it up for me and tell me when it's done." |
With the newer agent modes, like ChatGPT Work, Claude Cowork and Claude Code, the AI can actually open a browser, use apps and do the job. And even in a normal chat, it can write the email, fill in the form text or give you a finished file instead of instructions.
Don't give me instructions. Do as much of this as you can yourself, and only hand me the parts you truly can't do, with the exact words or clicks I need for each one.
HEADS UP: Agents should still ask you before anything that spends money, sends something for you or can't be undone. That's a good thing. Read what it's about to do, then approve.
When You're Stuck, Ask It How to Get Unstuck
People get stumped and close the tab. A good employee doesn't quit the first time something breaks, and you shouldn't either.
If you hit a wall, the AI is the first one you should ask, not the last.
- Paste in the error. Copy the exact message you're seeing.
- Take a screenshot. Drop it in the chat. It can read your screen.
- Say what you expected. "I clicked Save and nothing happened."
I'm stuck. Here's what I was trying to do: [YOUR GOAL]. Here's what happened instead: [WHAT YOU SEE, OR ATTACH A SCREENSHOT]. What would you do next? Walk me through it one step at a time, and wait for me after each step.
Still stuck?
Ask it to come at the problem a completely different way:
"Forget the way we've been trying. What are three completely different ways to get to the same result? Pick the one most likely to work and let's do that."
Common Mistakes and Quick Fixes
| If this happens | Try this |
|---|---|
| The answer is generic and could be for anyone | You skipped the context. Tell it who you are, what matters to you, or say "interview me first." |
| It keeps making the same mistake | Use the "write yourself a rule" prompt from Step 2, or start a fresh chat with a summary. |
| It did exactly what you said, but it's not what you wanted | You gave it steps, not a goal. Tell it the outcome you're after and let it pick the how. |
| It gives you instructions instead of doing it | Say "do it for me." Switch to an agent mode (ChatGPT Work, Claude Cowork) if it needs to click around. |
| It says it can't do something | Ask "What's the closest thing you can do?" or "What would make this possible?" |
| You don't know if the answer is right | Ask "How confident are you, and what would you double-check?" Check money, health and legal stuff yourself. |
| You've been going back and forth forever | Summarize into one prompt and start a new chat. |
Quick Reference: The Four Rules
| Rule | What to type | Why it works |
|---|---|---|
| Give it the goal | "My goal is ___. Figure out the best way." | It knows more ways to do it than you do. |
| Say exactly what's wrong | "This isn't right because ___. Don't ___. I want ___." | It fixes the real problem instead of guessing again. |
| Make it do the work | "Do it for me." | It can do way more than it offers to. |
| Never stay stuck | "I'm stuck. What would you do next?" | It's the fastest way out of almost any problem. |
DO THIS TODAY
Pick one task you've been putting off. Paste the manager prompt, fill in your goal and a few lines about you, and send it. When the first answer misses, don't say "try again." Tell it exactly what's wrong, like you would a new employee.
Stop searching AI. Start managing it.
Published · Checked