I Know What I Want AI to Do. What Should I Ask It?

You know what you want AI to help you do. Write an email. Plan your week. Compare a few options. Organize a project. Create a meal plan. Work through a business problem.

Start by explaining what you are trying to accomplish, then give AI the information it needs to understand the job.

Quick Answer: What Should I Ask AI?

Think of that as a checklist, not a rigid prompt formula. Your job is simply to give AI enough useful information to understand what you are asking it to accomplish.

Start With the Goal, Not the Perfect Prompt

One of the easiest ways to get stuck with AI is to assume you need to know the exact wording before you begin. You don’t.

Start with the outcome you want. Instead of asking yourself, “What is the perfect prompt for this?” ask, “What am I trying to get done?”

Your answer can be completely ordinary:

• I need to reply to this email.
• I need to figure out what to work on first today.
• I want meals I can make with the food I already have.
• I need to compare these three options.
• I have notes everywhere and need them organized.

Those are not finished prompts. They are goals, and a clear goal gives you somewhere to start.

Give AI the Context That Changes the Answer

Suppose you tell AI:

AI can write an email, but it does not know who the email is for, why you are writing, what happened before this message, or what you need the recipient to do.

Now add the information that changes the job:

That extra information is context. Context is not every fact you know. It is the information AI needs to handle this particular task.

If you are asking AI to plan dinner, the food in your kitchen may matter. Your internet provider probably does not. If you are asking AI to prioritize your workday, your deadlines and available time may matter. What you ate for breakfast probably does not.

The goal is not to give AI more information. It is to give AI the right information.

Tell AI About Preferences That Matter to You

Two people can ask AI to perform the same task and want very different outputs. You may want an email to sound warm and conversational, while someone else wants it short and formal. You may want familiar meals with minimal prep, while someone else wants to experiment with new recipes.

Those preferences belong in the prompt when they materially affect what a useful answer looks like.

For example:

Or:

Preferences help AI adapt the response to your situation instead of an imaginary average user.

Add Constraints Before AI Goes in the Wrong Direction

A constraint tells AI what boundaries it needs to work within. You might have a deadline, budget, word limit, limited amount of time, ingredient restriction, required format, tool limitation, or an option that is already off the table.

If you only have 30 minutes, say so. If your budget is $100, say so. If the email must be under 150 words, say so. If you cannot move a meeting, say so.

Constraints reduce the number of directions AI can take. That matters because an AI response can be perfectly reasonable and still be useless if it ignores a limitation the model did not know existed.

Tell AI What You Want Back

Sometimes the problem is not the task. It is the output.

Consider the request, “Help me organize these project notes.” What should AI return? A summary? A checklist? A timeline? A table? A list of unanswered questions? A project brief?

If you know what would be easiest for you to use next, tell AI. For example:

Or:

You are defining the output requirements. This is one of the simplest ways to make an AI response easier to use.

Put the Pieces Together

You do not need a complicated prompt formula. A useful prompt can often be built from a few straightforward pieces:

You do not normally need to keep those labels in the final prompt. They are simply a way to check whether you have given AI the information that matters.

Example: Turn “Organize My Day” Into a Useful Prompt

Suppose your starting thought is:

That is enough to identify the goal, but AI still needs more information. Maybe you have six tasks competing for your attention, a meeting at 4 PM, about five hours of focused work time, and a preference for finishing one meaningful task before switching to another.

Now AI has context, a time constraint, and a useful preference. You can also tell it what you want back: an ordered plan for the day, plus anything that should move to tomorrow.

Put those pieces together:

The goal never changed. You simply gave AI enough information to understand what “organize my day” means for you.

What If I Don’t Know What Context AI Needs?

Ask.

You do not have to anticipate every useful detail yourself. Start with the goal and tell AI to gather the missing information before it tries to complete the task.

For example:

Now AI’s first job is not planning the week. Its first job is gathering the missing context.

This approach is especially useful when the task has several moving parts or you are not sure what information matters. You can also tell AI to ask questions one at a time if receiving a long list of questions would feel overwhelming.

What If I Haven’t Made All the Decisions Yet?

Tell AI what is still undecided.

Suppose you need to plan a small birthday dinner for eight people, but you have not chosen the menu or theme. Instead of letting AI fill in those decisions without you, say:

AI can help you organize a decision without pretending the decision has already been made.

There is an important difference between “Help me execute what I decided” and “Help me think through a decision I haven’t made yet.” Tell AI which job you need.

Tell AI What You Don’t Want When It Matters

Sometimes a useful constraint is simply ruling something out.

You might say:

• Give me dinner ideas, but don’t include seafood.
• Help me rewrite this, but don’t make it sound more formal.
• Suggest ways to organize this room without recommending that I buy storage products yet.
• Help me plan the project, but don’t assign deadlines until I give you everyone’s availability.

You do not need to give AI a long list of prohibitions. Use negative constraints when they prevent the response from heading toward an option you already know will not work.

If the First Answer Misses the Mark, Don’t Automatically Start Over

Your first prompt does not need to produce the final answer. Look at what AI gave you and identify what needs to change.

If the answer is too broad, tell AI what to narrow. If it is too detailed, ask for a shorter version. If AI made an incorrect assumption, correct it. If you forgot an important constraint, add it. If the format is difficult to use, tell AI what format would work better.

For example:

Or:

Or:

This is prompt refinement. You are adding useful information to the conversation instead of rebuilding the entire request every time something needs adjustment.

What If I Can Explain What I Want but Still Can’t Turn It Into a Prompt?

This is exactly where many people get stuck.

You may be perfectly capable of saying:

But then you feel like you need to translate that into some special form of “AI language” before you can continue.

You don’t.

One option is to tell the AI what you want and ask it to interview you for the missing information. Another is to use a tool designed to do that translation for you.

Before You Send the Prompt, Check These Five Things

You do not need to analyze every prompt you write. But when a task matters or AI keeps missing the point, a quick check can save time.

Ask yourself:

1. Goal: Did I clearly say what I am trying to accomplish?
2. Context: Did I include the information that changes the answer?
3. Preferences: Did I explain anything that would make the response more useful to me?
4. Constraints: Did I identify important limits, requirements, or things AI should avoid?
5. Output: Did I say what I want AI to give me?

If one does not apply, leave it out. The goal is not to make every prompt longer. The goal is to remove ambiguity that matters.

Related guide: AI Basics: A Practical Guide to Using AI in Everyday Life