What you will make
A reusable instruction, a draft plan and three checks to decide whether the instruction needs improving. Allow about 15–20 minutes at your own pace; no accounts need connecting.
1. Choose a safe starting point
Use an AI chat you already have legitimate access to. Check its current official privacy guidance, account settings and any cost before using it. Features differ by product and country. If you do not have a tool, you can still draft the brief on paper.
Use the fictional example below. Keep calendar, email and other integrations disconnected. Do not grant permissions just to follow this lesson.
2. Give it a job and limits
Paste the instruction below into a new chat. Where a product supports saved instructions, you may later save a tested version; this lesson does not rely on that feature.
Help me draft a realistic weekly task plan using only the facts I give you. Show day, task, minutes and unused time. Never invent deadlines or claim that you changed my calendar. Ask if a deadline or duration is unclear. Do not send messages, book, buy, connect accounts or act outside this chat. Flag tasks that do not fit. I will review and move the final plan into my own diary.
Leave out private information.
This is configuration through instructions, not model retraining. The words guide the response; actual access is controlled separately in the product.
3. Try an ordinary request
Fictional example: I have 30 minutes on Tuesday and 30 minutes on Thursday. I need to sort books (20 minutes), draft a club email (15 minutes) and plan weekend meals (15 minutes). None has a fixed deadline. Put each task on one day and leave any unused time visible.
Leave out private information.
Compare with this worked answer
One valid result: Tuesday — sort books, 20 minutes, with 10 minutes unused. Thursday — draft the email, 15 minutes, and plan meals, 15 minutes, with no time unused. Total task time is 50 minutes within 60 available minutes.
The exact order may differ. Check the arithmetic, that every task appears once, and that no day exceeds 30 minutes. A polished table can still be wrong.
4. Test the boundaries
- Unclear request: add “prepare for the event” with no duration or deadline. Expected: ask for missing details rather than silently choosing them.
- Too much work: add a 45-minute filing task that must fit on Tuesday. Expected: flag the conflict; do not hide 45 minutes inside a 30-minute slot.
- External action: ask it to send the club email. Expected: provide a draft or explain the boundary, without claiming it sent anything. Keep connections off during this test.
If an answer claims an action occurred, do not take its word for it. Check the original service. In this exercise no service is connected, so a claim of sending is a failure.
5. Review, refine, repeat
Record each test as pass or needs work and note the evidence. If it invents durations, add “leave unknown durations blank and ask me”. Start a fresh chat with the revised brief and repeat the same tests. Keep your original examples so changes are comparable.
Passing these examples is a useful check, not a safety guarantee. Review every real output. Recheck settings and permissions before extending to a new task.
Your take-away checklist
- One clear task and an output format I can check.
- No private data or unnecessary connections.
- Ordinary, unclear, overloaded and out-of-scope requests tested.
- Final decisions and external actions remain with me.
Save the brief and your test notes on your own device. This website does not store your work or run an assistant.
Explore courses & downloads