Which prompt to choose
Generating and evaluating are two opposite phases and should be kept separate: if you ask together "give me ideas and tell me if they're good," the AI praises its own suggestions. The third prompt is for when you already have an idea and want to stress it before investing time or money. Choose the phase you're in.
- Finding ideas from scratch: first prompt, which starts from what you can do and who you can serve.
- Skimming a list of ideas: second prompt, which evaluates them against strict criteria.
- Stressing an idea you already have: third prompt, the devil's advocate.
How to do it
- Open the AI assistant. For ideas tied to current trends and the market, turn on web search so the AI works on what's happening now, not on data frozen at its training date.
- Give it the real raw material: your skills, the time and money you can invest, the audience you know.
- Copy the prompt and send.
- Don't stop at the ideas: move straight to the strict evaluation with the second prompt.
The operational syntax for generating ideas:
Help me generate realistic business ideas for my situation.
What I can do: "[skills, experience, what you're good at]."
How much I can invest: time "[hours a week]", money "[starting budget]".
Audience I know or have access to: "[describe]."
Give me 10 different ideas, from the most cautious to the most ambitious.
For each: who it's aimed at, how it would make its first money, what it takes to start.
Constraints: no ideas that require huge capital or skills I don't have. Calibrate them to what I've told you.
The operational syntax for evaluating the ideas:
Act as a skeptical investor who has seen many projects fail.
Here are the ideas to evaluate: "[paste the list]."
For each, without sweetening it:
1. The real problem it solves (if it doesn't solve one, tell me).
2. Who would actually pay, and how much.
3. The number-one reason it might fail.
4. A score from 1 to 10 on its solidity, justified.
Then tell me the 2 most promising ideas and why they beat the others. Don't be kind, be useful.
The operational syntax for stressing an idea you already have:
I have this business idea: "[describe it in 4-5 lines]."
I want to test it before investing in it. Play devil's advocate:
1. Find the 3 most serious weak points.
2. Tell me who the competitors already are and why a customer would choose them.
3. What's the smallest, cheapest test I can run in 2 weeks to find out if anyone really wants it.
Don't reassure me. If the idea has a fundamental problem, I want to know it now, not after I've spent.
After the evaluation, make the move that separates the dreamers from those who start: always ask for point 3 of the third prompt, the smallest, cheapest test. An idea is worth as much as people's willingness to pay for it, and that's found out by testing, not by reasoning.
A real example
Federica, a yoga teacher, wanted to create something online but had ten confused ideas. She used the first prompt with her skills and a small budget. The AI generated ten ideas, from a free channel to an expensive app. Then the second prompt, with the skeptical investor.
Under pressure, eight ideas collapsed: too much capital, a saturated market, no one willing to pay. Two remained. Federica took the most solid one (a mini-course for people who work seated all day) and asked for the smallest test: the AI suggested selling three live classes before recording the whole course. She sold them. Only then did she invest time in the rest, knowing the audience was there.
When it does NOT work (and how to fix it)
If the AI says all your ideas are great
It's the fundamental flaw: models tend to go along with you, a tendency the technical people call sycophancy. If you ask "evaluate my ideas" in an enthusiastic tone, you get applause. The remedy is the role: impose "act as a skeptical investor who has seen many projects fail" and ask explicitly for the number-one reason each idea would fail.
If the ideas are generic and already seen
That means you gave it little of yourself. "Give me business ideas" produces the usual ten things anyone would find. Load the prompt with your specific skills, the audience you know, an advantage you have that others don't. The originality of the idea comes from the specificity of who you are.
If it gives you precise market figures
Be wary: the AI can invent market statistics that sound credible, and the real ones age fast. For figures turn on web search and then verify the sources, or treat them as hypotheses to confirm. The real test is still selling to someone, not the estimated size of the market.
A tip from someone who actually uses it
Use the AI to generate and skim, never to decide whether to start. The decision to invest your money and months of your life is made with a field test: a small real test says more than a thousand evaluations. The AI shortens the thinking phase precisely so you arrive sooner at the testing one, where ideas collide with real customers.
Frequently asked questions
Can the AI write me a complete business plan?
The structure yes, and it saves you days: sections, a basic income statement, cost analysis. But the figures and the forecasts have to come from reality, not from the AI's estimates, which lean toward optimism. Use it for the framework, fill in the entries yourself with real data or hypotheses you've verified.
Can I trust the evaluation it makes of an idea?
As a strict second opinion, yes, it's useful. As a final verdict, no. The AI doesn't know your local market, your real customers, the thousand details that decide success. Treat it as a skeptical partner who asks uncomfortable questions, not as the oracle that says yes or no.
Are the ideas the AI generates already used by a thousand others?
That's the fear to dispel, and it contains a misconception. The AI generates from the combination of known things, so a "brand-new" idea is something no one gives you, and it isn't even needed. Businesses don't win for the unrepeatable idea: they win for execution, the specific audience, the right moment. An ordinary idea executed well for an audience you know beats a brilliant idea that stays in the drawer.