TutorialsMulti-Turn Dialogue + 3 Prompts Hands-On
The five-step order for multi-turn dialogue, the six questions that make one task clear, and three rounds of hands-on work on the main case — prompts A / B / C in copy-ready full text.
What this note is
The last lesson covered the formula. This one is the body of the course: putting the formula to work on one real task.
The formula helps you say one sentence completely, but a real task rarely finishes in one round. The point of multi-turn dialogue is not saying more — it is knowing, before each round starts, what that round must produce.
There is only one way to communicate: talk to AI the way you would talk to a very smart colleague — plainest, most direct language, say whatever comes to mind. Not knowing how to write a 'professional prompt' is completely fine.
The five-step order for multi-turn dialogue
The easiest order to get started with:
Define the final result
What do I ultimately want to get?
Define the current round
Which piece should be done first this time?
Add materials and details
What does AI need now in order to continue?
Review the result and correct course
Where does it match, and where does it not?
Decide the next round
Expand further, change materials, or change direction?
Steps 1 and 2 are the crux: set the final result first, then set this round. Skip step 1 and start working, and you will often find halfway through that the direction is wrong.
Six questions that make one task clear
Putting steps 1 and 2 onto one sheet gives these six questions:
| Question | What you answer |
|---|---|
| 1. Who ultimately uses it? | Yourself, a client, a colleague, a manager, or the public |
| 2. What exactly is delivered, and by when? | A document, a list, an article, a decision, or a next action — plus when it is needed |
| 3. What material already exists? | Raw records, data, examples, rules, and existing versions |
| 4. What have you tried, and where are you stuck? | How you did it before, which step failed, where the most time went |
| 5. Who judges quality, and which three things get checked? | Who makes the call, and the specific criteria — complete, accurate, usable |
| 6. What is the next 30-minute small test? | Which small slice of the result to try first, and how you will tell whether it is worth continuing |
Answer the first five yourself first; the sixth decides whether you should start now. Anything verifiable in 30 minutes should not be planned as a three-month project.
The stronger AI gets, the more judgment stays with you
AI can help you search, organize, generate, rewrite, and speed things up. But these things are still decided by a person:
Is this really what I want
Whether the result is right, only you know. AI executes your request — but whether the request itself is what you truly want is your call.
What is trustworthy and what is only inference
AI will state inferences as facts with total confidence. Anything involving numbers, dates, names, or conclusions must be checked by you.
What can actually be sent out
Your name is on it and so is your responsibility. Run your own final pass before publishing, submitting, or sending.
Which materials may be given to AI
De-identify confidential and sensitive information first, or swap in simulated material. Client privacy, accounts, verification codes, and unauthorized files never go in directly.
Whether to continue at all
Expand, change materials, or stop — only you can make that call.
Hands-on practice: the main case
- Main case: take a long course transcript and turn it into study material that can be reviewed, checked, and continued next time.
- Demo note: demo material uses de-identified examples or simulated fragments. No real client names, chat contents, or unauthorized materials are shown.
Three rounds, three prompts. Copy them and go.
Prompt A: use the formula to state the result you want this time
Do not let it start working in round one — align on direction first.
【Role】You are a document organizing assistant.
【Task】Help me turn the course transcript below into reviewable study material.
【Context】The user is me, and the purpose is after-class review. This is my first time organizing it, so I want to confirm the direction first.
【Format】First give me a one-page task brief: the final result, what it must contain, and the acceptance criteria.
【Constraints】Do not generate the full material yet. If key information is missing, ask me one question at a time.
Before you start, ask me three questions and begin only after I confirm.Prompt B: define what this round covers — and only that
Once the task brief is in hand and the direction is confirmed, this round you give it material — and only a small slice.
【Task】Based on the final result we confirmed, this round only organizes the first 10 minutes of the course transcript.
【Context】Here is that part of the transcript:
【Paste material】
【Format】Produce a small sample first, containing: key points from the original, what I should remember, what you added, and where verification is needed.
【Constraints】Do not handle the rest in this round. Do not expand to the full material on your own.
Do this small piece first after confirming, and continue after I check it.Prompt C: take the result, check it, then decide the next round
The point of round three is not continuing — it is first turning your review comments into a checklist.
【Context】This is the small sample you just produced. My review comments are:
【Fill in what to keep, change, delete, and add】
【Task】First turn my comments into a revision checklist, sorted into must-fix / nice-to-improve / needs-my-confirmation.
【Constraints】Do not rewrite the whole text directly.
After I confirm the revision direction, handle the next section of material:
【Paste the next section or describe the scope】
The new result must still satisfy:
【Fill in the acceptance criteria for this task】The most important line in prompt C is 'do not rewrite the whole text directly.' Ask for the checklist first, the rewrite second — that way every change is one you confirmed, rather than a decision it made for you.
Checkpoints to watch during the demo
While watching the demo, pay attention to these:
| Checkpoint | What to look at |
|---|---|
| The questions AI asks | Do they actually affect the result, or are they trivial details |
| Are the five parts clear | Have role, task, context, format, and constraints all been stated |
| Is the scope held | Did this round honestly cover only the first 10 minutes |
| Can the sample be checked | Can you verify it item by item, rather than just feeling it is fine |
| Where the next round goes | Expand further, or re-state what must be delivered |
Demo materials: a 10–15 minute de-identified course transcript or simulated meeting notes, three key points prepared by the instructor, a clearly defined purpose, and a simple acceptance sheet.
Next lesson, we turn this process into templates you can reuse.
Understand the Mechanics + One Formula
Four key concepts, five easily-confused things, 8 common scenarios, and the soul of the lesson — the formula role + task + context + format + constraints.
Template Pack + Blank Cards + Quality Red Lines
5 ready-to-copy templates, 3 blank cards, 3 general red lines + 5 lesson-specific ones, and an appendix on choosing tools.