These are the techniques that separate average outputs from genuinely useful ones. Use them situationally — not as a checklist to paste into every prompt.

3.1 The Assumption Audit

After getting an answer, ask: "What assumptions did you make to answer this? List them explicitly." Then challenge the ones that do not match your situation. This is the fastest way to personalise a generic answer.

Example

You asked: "How do I prepare for a job interview?"

Model assumed: You have 2+ weeks, you are applying for a corporate role, you are experienced.

You say: "Assumption 2 is wrong — I’m applying for my first internship. Revise with that context."

3.2 Steel-Manning

Ask the model to make the strongest possible case for the opposite of what you believe, then ask it to critique its own argument. This is the fastest way to stress-test any idea, essay argument, or business decision.

Example

"I believe social media is net negative for teenagers. Make the strongest possible case that it is actually net positive. Then critique that argument from the original position."

3.3 Constraint Tightening

Generic prompts get generic answers because the model optimises for the average use case. Tighten constraints to force specificity. Think of it like adjusting a camera lens — the tighter the focus, the sharper the image.

Version Prompt
Loose (generic output) "Explain blockchain."
Tight (specific output) "Explain blockchain in exactly 4 bullet points, each under 20 words, for a business student who understands databases but not cryptography."
Loose (generic output) "Give me ideas for my startup."
Tight (specific output) "Give me 5 startup ideas in the edtech space that could work in tier-2 Indian cities, require less than ₹50,000 to start, and can be run by a solo founder. For each, name the core problem and one reason it might fail."

3.4 The Failure Frame

Before asking for a solution, ask: "What are the 3 most likely ways this fails?" This primes the model to think in edge cases, not just happy paths. The solution it then gives you will automatically be more robust.

3.5 The Socratic Chain

Instead of one giant prompt, build a chain where each answer informs the next. This is how

consultants think — they do not ask for everything at once, they drill down.

  1. "What are the key decisions in designing a food delivery app?"