Make the AI Slow Down Before It Answers
The simplest upgrade is to stop treating the first answer as the finished answer. ChatGPT and Claude are good at sounding complete, even when they are guessing, skipping context, or choosing the easiest path. Your job is not to become technical. Your job is to build a small habit: ask for an answer, then ask for a check.
Start with a prompt that gives the AI enough shape to work with. Use four plain parts: the task, the background, the limits, and the question. For example: “Write a polite email asking my manager to move our meeting. Background: I have a medical appointment. Keep it short and not too personal. Before writing, ask me anything you need to know.” That last sentence matters. It tells the AI not to pretend it has all the facts.
Ask for a Review, Not More Polish
Once you have a draft, do not only say “make it better.” That often gives you smoother wording without fixing the real problem. Ask the AI to review the draft against the job it was supposed to do. A useful prompt is: “Check this against my goal. What is unclear, too strong, missing, or likely to be misunderstood?”
This works for everyday tasks. For an email, ask whether the tone fits the relationship. For a travel plan, ask what assumptions could break the plan. For a short research summary, ask which claims need a source. The tradeoff is that the answer may feel less flattering. That is the point. You are asking for useful friction, not encouragement.
Use a Skeptic When the Stakes Are Higher
For important work, add a second check. A “skeptic” is simply a role you give the AI: it should look for problems instead of helping you finish faster. Try: “Act as a careful reviewer. Challenge this answer. List the three weakest claims, the missing information, and what would change your recommendation.”
This is especially useful when comparing options. If you ask, “Which laptop should I buy?” the AI may pick one confidently. Better: ask it to compare two or three options, then ask for the case against its own choice. For medical, legal, financial, or job-changing decisions, do not rely on one AI answer. Compare with reliable sources or another tool, and treat agreement as a signal to investigate further, not as proof.
End by Asking What Would Make It More Reliable
The final move is to ask the AI to name the missing evidence. This turns a vague answer into a checklist you can act on. Use: “What information would make this answer more reliable?” or “What are you assuming that I should confirm?” The AI might say it needs the date, audience, budget, source, policy, exact wording, or local rule.
That question closes the loop. You are no longer just receiving text from a machine. You are setting up a small review process: give context, get an answer, ask for weak spots, then ask what would improve confidence. The goal is not to make AI perfect. The goal is to stop accepting a polished guess when a checked answer is only one prompt away.
Key takeaways
- Do not treat the first AI answer as final; ask it to check its own work.
- Before the answer, ask: “What do you need from me to give your best answer?”
- After the answer, ask what is unclear, missing, too strong, or likely to be misunderstood.
- For important choices, ask the AI to argue against its own recommendation.
- Ask what information would make the answer more reliable before you act on it.