TutorialsUnderstand 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.
What this note is
This lesson is the foundation. First we put a few words into plain language, then we separate five things that are easy to mix up, and finally we land on the only thing worth memorizing in this course: one formula.
Everything hands-on in the next two lessons grows out of that formula.
Four key concepts in plain language
| Concept | Put plainly |
|---|---|
| Result | What exactly must be delivered, who uses it, and what for |
| Prompt | The request you write to AI — essentially, saying it completely |
| Formula | A fixed structure that completes a vague sentence (role + task + context + format + constraints) |
| Multi-turn dialogue | Do not expect a final draft in one shot; each round advances one small result |
The one thing to remember from this lesson: writing a prompt is not about being polite to the AI — it is about saying it completely.
Telling apart five things
These five get mixed together all the time, and once they are mixed, nothing can be said clearly. Separate them first:
| Easily confused | Put plainly | How we handle it in class |
|---|---|---|
| Goal | What you ultimately want to get | State the result first, not the tool |
| Function | What kind of work AI should do for you | Organize, compare, break down, rewrite, or check |
| Tool | Which entry point to use | Choose it after goal and function are clear |
| Step | Which part to do first this time | One round of dialogue advances one small result |
| Acceptance | How you know the result is usable | A person can check it, not just look at something complete |
Order matters: goal → function → tool. Many people agonize over which AI to use first, but if the first two are unclear, switching tools changes nothing.
Common scenarios
These 8 kinds of work come up most often. Skim to find the one you need most; that is where the templates later will pay off.
| Scenario | What the work is | How AI helps |
|---|---|---|
| Long course material | Turn recordings or transcripts into reviewable material | Confirm the purpose first, then organize in sections, flagging highlights and gaps |
| Meeting minutes | Turn discussion into decisions, assignments, and next steps | Extract facts, decisions, owners, deadlines, and open questions |
| Weekly reports and work docs | Turn scattered notes into reportable content | Organize progress, problems, and next steps in a fixed structure |
| Content writing | Turn dictation and raw material into an editable draft | Define the reader and the intended result first, then produce a small sample |
| Task breakdown | Turn not knowing where to start into a next step | Ask about the goal, materials, constraints, and done criteria |
| Close reading of documents | Find the genuinely useful parts of a long report | Define the question in use first, then extract evidence and conclusions |
| Data analysis | Find what deserves attention in a table | Define metrics and judgment criteria first, then analyze |
| Workflow reuse | Turn a once-done task into reusable steps | Save the inputs, the process, the human judgment points, and the acceptance criteria |
Notice what the last column has in common: every row confirms first, then executes. That is exactly the problem the formula solves.
One formula (the soul of the lesson)
The universal prompt formula: role + task + context + format + constraints
This is the soul of the lesson. It completes a vague sentence so AI stops guessing:
| Part | What to make clear | Example |
|---|---|---|
| Role | What work perspective AI should assist from | Meeting-notes assistant, content editor, document analysis assistant |
| Task | What specifically it must accomplish | Extract decisions, owners, and next steps from the transcript |
| Context | What it needs to know | Who will use it, where the material comes from, existing versions, usage scenario |
| Format | What the result should look like | A table, a checklist, a one-page summary, or a spoken script |
| Constraints | What it must not do, what it must obey | Do not fabricate, do not change the meaning, make a small sample first, flag items to confirm |
Two things to be clear about
First, you do not have to write all five parts every time — but if a key part is missing, AI can only guess.
Take 'help me organize this' — with no audience, no format, and no acceptance criteria, AI can only guess what you want. Getting it right would be luck; getting it wrong is the norm.
Second, the formula is a skeleton that helps you think clearly, not a spell that produces good results the moment you recite it.
What really decides quality is whether you have thought through what must be delivered, who will use it, and what counts as usable.
Do not treat the formula as a spell. You can know the five words cold, but if you have not thought through what you are delivering, AI will still hand you something polished and unusable.
Advanced trick: ask three questions before starting
Add one line at the end of your prompt:
Before you start, ask me three questions and begin only after I confirm.This way AI does not put its head down and generate immediately — it first asks about what you left unclear, then gets to work.
In one sentence: it turns 'it guessed wrong and you rewrite' into 'it asks you first'.
The one principle for the whole lesson
AI drafts and fills the gaps; you judge and decide. Treat AI as a very smart junior colleague — you are the lead, and your job is to brief the work well and make the call.
This is the undertone of everything hands-on that follows. Next lesson, we put the formula to work on a real task.
AI Prompting Techniques and Multi-Turn Dialogue
One formula to say it completely — role + task + context + format + constraints — then a few rounds of dialogue to reach a checkable small result.
Multi-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.