Find what you already have
Wrong question. You already have the raw material. You just haven't looked at it.
Here's the audit. Twenty minutes, one sheet of paper, tonight.
Step 1: List the last ten things people asked you for help with.
Work, family, neighbours, the group chat, all of it. Not what you're good at in some resume sense. What people actually come to you for. Most guys are surprised by this list because the thing they get asked about is not the thing on their business card.
Step 2: Circle anything you've explained more than twice.
Explaining something twice means there's a gap in the world where you happen to have the answer. Explaining it five times means there's a product.
Step 3: For each circled item, ask AI this exact question.
"Here is my job, my background, and the thing people keep asking me to explain. What are five things I could build or offer that would help those same people? Keep them small enough for someone with a full-time job and kids."
That last sentence matters. Without it you'll get ten ideas that need a team.
Step 4: Pick the one you could describe to a friend in one sentence.
Not the most impressive one. The most explainable one. If you can't say it plainly, you'll never sell it plainly.
The reason this works: the first AI skill isn't writing clever prompts. It's noticing what you already know. AI is very good at helping you build the thing. It is terrible at telling you what your thing is, because it doesn't know what ten people asked you last month.
You do. That's the part only you have.
Do the twenty minutes. Most guys find the answer was sitting in the second question.
Turn it into ten ideas
Then the important part. Read the ten and notice which one you keep coming back to. That's your answer. Not the most profitable one, the one you didn't want to stop thinking about.
Ask for ten more if none of them land. It costs you two minutes.
The reason this works: you don't have an idea problem. You have a "nobody ever asked me the right questions about myself" problem. Most of us have been answering to a job description for twenty years.
Make AI interview you first
Here's the whole thing. Paste this at the start of any request:
"Before you build this, ask me questions one at a time until you have what you need. Then build it."
That's it. That's the trick.
Why it works: bad AI output is almost never the AI being dumb. It's the AI guessing at things you never told it. Your budget, your timeline, who it's for, what you already tried, what you're actually good at.
When you let it ask instead of guess, it stops inventing a person who isn't you.
The "one at a time" part matters more than it looks. Ask for all the questions at once and you get a wall of twelve that you'll skim and half-answer. One at a time feels like a conversation, and you'll actually answer properly.
Fair warning: the first time you do this it will ask you something you don't know the answer to. That's not the AI failing. That's the thing you needed to figure out before you started building.
Five questions before you start
1. "What is the real problem here, in one sentence?"
If AI can't state it simply, I don't understand it yet.
2. "What is the smallest version of this I could finish today?"
Finished and small beats impressive and abandoned.
3. "What would make this fail? Be blunt."
AI is a better pessimist than I am. Let it find the holes before the holes find you.
4. "What do I already have that makes this easier?"
Skills, contacts, tools, past work. The answer always surprises me.
5. "What is the single next step? Just one."
Not a plan. Not a roadmap. One step. Momentum eats strategy for breakfast.
Save this one. It is the whole checklist.