Start by stopping the guessing
Most weak AI answers come from missing context. Context means the details that help the AI understand your situation: who the message is for, what tone you want, what has already happened, what you are trying to avoid, and what a good result looks like. When those details are absent, ChatGPT or Claude will usually fill the gaps with a reasonable-sounding guess.
The fix is simple: before the AI answers, tell it to ask you what it needs. Add a line like, "Before you answer, ask me any questions you need to give me a better result." This is useful when the request has stakes or many possible directions, such as writing an email to your boss, planning a trip, comparing products, preparing for a meeting, or researching a sensitive decision.
Use it when the answer depends on your situation
Because this habit slows the chat down slightly, do not use it for everything. If you ask, "Give me five dinner ideas using eggs," the AI can answer well enough. But if you ask, "Help me write an apology email," the missing details matter. Who was affected? Was it your fault? Do you want to repair trust, avoid legal wording, or keep it brief?
A good everyday prompt has four parts: the task, the background, the limits, and the invitation for questions. For example: "Help me write a polite follow-up email to a hiring manager. I interviewed last Thursday and have not heard back. Keep it warm, under 120 words, and not desperate. Before drafting, ask me what else you need." This gives the AI a starting point, then lets it check the gaps instead of inventing them.
Make the questions easy to answer
If you simply say "ask me questions," the AI may ask too many. Give it a small boundary: "Ask up to three questions first" or "Ask one question at a time." One-question-at-a-time works best when you are unclear yourself, such as planning a hard conversation or shaping a personal statement. A short batch works best when you already know the situation and just want a better first draft.
You can also tell it what kind of questions to ask. Try: "Ask only about audience, tone, and goal before you write." For research, use: "Ask what decision I am trying to make before you summarize." For planning, use: "Ask about budget, timing, and constraints first." A constraint is a rule or limit, like "must be under $500," "cannot involve travel," or "needs to sound friendly but firm."
Know when to answer and when to verify
Once the AI asks, answer plainly. You do not need perfect wording. Short fragments are enough: "Audience: my manager. Tone: calm. Goal: get a deadline extension without sounding careless." Then ask it to draft, compare options, or give you a plan. The better your answers, the less the AI has to assume.
For high-stakes topics, questions are not enough. If the AI gives facts that affect money, health, legal risk, or a major business choice, check them. One practical habit is to ask the same factual question in more than one AI tool and look for agreement, then verify important claims from reliable sources. Use the question-first habit to improve fit. Use verification to improve truth. Together, they keep the AI from sounding confident while being wrong.
Key takeaways
- Add "Before you answer, ask me what you need" when missing details could change the result.
- Use the habit for emails, planning, research, decisions, and sensitive writing, not simple one-off requests.
- Give the AI the task, background, limits, and then invite clarifying questions.
- Limit the number of questions when you want speed: "Ask up to three questions first."
- For important factual claims, verify the answer instead of trusting confidence.