Five Prompts That Fix Most Bad AI Answers

Most bad AI answers aren’t the AI failing. They’re the AI answering the question exactly as asked, which is a different thing from answering the question you meant.

After a year of running my business on these tools and coaching my Roundtable members through theirs, I keep reaching for the same five sentences. Each one fixes a specific, predictable failure. Steal them.

1. “Before you start, do you have any questions for me?”

End every complex request with this. Ambiguity is where AI projects die: it doesn’t know your constraints, so it guesses, and its guesses compound. Invited to ask questions, it surfaces the ambiguity up front, and two minutes of clarification replaces two hours of rework. This is the single highest-value sentence on this list.

2. “Give me the top ten solutions, ranked by what matters to me.”

The first solution an AI offers is an option, not the best option. When the answer matters, ask for the field: give me the top ten approaches, ranked by least effort and lowest cost, or whatever your actual priorities are. Best fit to worst. Suddenly you’re choosing from a menu instead of accepting the daily special, and the right answer for you is frequently sitting at number four.

3. “I need you to think outside the box. Find me the free way.”

When AI says something can’t be done free, or can’t be done simply, push back. I’ve learned to say it directly: I’ve never found something that can’t be free. It might be hard, but I need you to think creatively right now. More often than not, it comes back with a path I hadn’t considered. The first no is a reflection of the average answer on the internet, not the limit of what’s possible. Treat it like a person, and don’t accept the first no.

4. “Tell me why this is a bad idea.”

AI agrees with you by default. Ask if your new offer is a good idea and you’ll get applause every time, which is worth nothing. So make the criticism mandatory: what could go wrong here? Give me the reasons this fails, so I can make it better. I do this any time I’m creating something that doesn’t exist yet. The ideas that survive interrogation are the ones worth your next three months.

5. “Here’s exactly the output I need, down to the details.”

Vague format requests get vague deliverables. If you need a checklist with boxes you can actually click in the document, say clickable checkboxes, because the printed-look kind is what you’ll get otherwise. If you need a file sized to a print shop’s template, give it the template specs. If a deliverable comes back close but wrong, don’t redo it yourself: tell it precisely what to change. Specificity in, usability out.

None of these require technical skill. They require remembering that you’re managing a very fast, very agreeable new hire, and managing means asking better questions than “can you do this?”