You ask AI to help with a task. The answer looks polished, but it is too general to use. You try again, adding “make it better,” and get a longer version of the same thing.
That is a useful moment to pause and explain what “better” means for this particular task. You do not need a special vocabulary. Start with the information you would give a helpful colleague.
This guide uses a fictional client meeting as an example. You can practice with it before trying a task from your own work.
Explain the task before asking for the result
Compare these two requests:
Help me prepare for a meeting.
And:
I have a 20-minute check-in with a client tomorrow.
We need to agree on the next step for their website project.
The first draft is ready. The client has not sent their photos.
We have not agreed on a launch date.
Help me prepare a short agenda and three questions to ask.
Separate confirmed facts from things we still need to decide.
Do not invent a launch date or assume the photos have arrived.
The second request gives the assistant a purpose, relevant facts, and a shape for the answer. It also identifies two things the assistant should leave open.
For your own request, try answering four questions:
- What am I trying to accomplish?
- What does the assistant need to know?
- What should the result look like?
- What must it avoid assuming?
You can write those answers in ordinary sentences. There is no need to turn every request into a long template.
Improve one part at a time
Read the first answer and name the part that needs work. A specific follow-up is easier to act on than “try again.”
For the meeting example, useful follow-ups could be:
- “The agenda is too long for 20 minutes. Keep the three decisions that matter most.”
- “These questions sound formal. Rewrite them as something I could naturally say to a client.”
- “Add a question about who will supply the photos and when.”
- “The answer assumes we are ready to launch. Remove that assumption.”
Keep the parts you like. Ask for a change to the part that is getting in the way. If an important fact changes, state the correction explicitly so the next version uses it.
You can ask for an explanation
If an answer uses a term you do not understand, ask: “Explain that in plain language, with an example from this task.” Asking questions is part of learning the tool.
Check the answer against what you supplied
A confident tone does not establish that a statement is true. Before using the agenda, compare it with your original notes. Did the assistant invent a date, promise a deliverable, or treat an unanswered question as a decision?
Ask it to list the assumptions it made. Then check those assumptions yourself. For this example, you should still see the photos and launch date as unresolved, with questions that help the meeting move them forward.
For a real client, check names, dates, commitments, and any statement you intend to repeat. Read the finished version aloud if you will use it in conversation. Edit anything that does not sound like you.
Practice with a small, finished task
Choose an output you can review in a few minutes: a meeting agenda, a short outline, or a list of questions. Keep your original notes beside the result so you can compare the two.
Before entering workplace information, follow your organization’s rules for the tool you are using. A fictional or anonymized example is enough to practice this exercise. You can learn how to explain a task without sharing a client’s private details.
If you want more examples, try three everyday work tasks to practice with AI. To choose an example for guided practice, read what to bring to an AI workshop.