Getting better AI results by asking AI to interview you

Executive overview

Most people expect AI to produce great output from a single prompt. It can't — it makes assumptions when context is missing. The fix is to flip the dynamic: let AI ask you questions before it does any work.

Telling AI to ask questions until it's 95% confident is the fastest path to high-quality output.

The core problem with standard prompting

• AI fills gaps with assumptions, not clarifying questions
• A vague prompt produces a generic result regardless of model quality
• Giving AI all context upfront feels like writing a "massive tome"

The 95% confidence technique

• Before asking AI to produce anything, tell it your goals first
• Add to any prompt: "before you respond, ask me any questions until you're 95% confident in doing a great job"
• "95%" is precise — vague thresholds like "mostly confident" produce inconsistent behaviour
• AI typically asks one focused round of questions, then proceeds
• Prompt AI to ask more if you sense it may have missed something

What AI tends to ask

• Goals and intended outcome
• Audience and context
• Specific details relevant to the task (e.g. client's problem, tone, budget)

Handling questions you can't answer

• Saying "I'm not sure" is acceptable — AI proceeds with what it has
• Alternatively: "I'm not sure — give me suggestions on different directions"
• This opens another round of Q&A rather than stalling progress

When to add extra direction

• For most tasks, the 95% threshold alone is sufficient
• For specialist work (deep analysis, brand writing), specify the areas you want probed
• Examples: tone and style for audience fit, or a specific analytical angle