6 min read
Beginner
Draft: this article is under editorial review and may change.Walk up to the counter at your favourite mamak and say “teh”. You will get a teh: hot, sweet, in a glass. Nothing is wrong with it. But if you wanted teh tarik, less sugar, to take away, it is not your drink.
Asking an AI works the same way. Many disappointing AI answers come from the request, not the AI. Ask for “a rubric” and it has to guess the subject, the level, the scale and the layout. So it gives you the most common answer, and the most common answer rarely fits your class.
AIM fixes this with three parts you can remember on the way to class. It works in any ChatbotAn AI you talk to in a chat window. You ask, it answers, and you decide what to do next. or AI agentAn AI that does a multi-step job for you (planning, opening files, using tools and checking its work) instead of only replying once., and you need no technical background.
The idea in 30 seconds
Actor: who answers
Pick the expert you need, such as "an assessment designer".
- Actor: who answers Pick the expert you need, such as "an assessment designer".
- Input: what it can't guess Your class, your rules, your material.
- Mission: exactly what to produce Start with an action verb. Say how much and in what form.
- Check against your Mission Tick off each thing you asked for. AI can sound sure and still be wrong.
- Vague in, average out. Clear in, useful out.
Key takeaways
If you remember nothing else, remember these.
- Actor: who the AI should be while it works.
- Input: everything it needs to know but cannot guess.
- Mission: exactly what to produce, and in what form.
- Then check the answer against your Mission before you use it.
Order your AI like you order teh tarik
A regular at the mamak never just says “teh”. They give the full order, and the tea maker gets it right first time. Tap between the two orders below.
Teh satu.
What you typically get back
The house default: hot, sweet, in a glass. Fine for somebody, but not for you.
Boss, teh tarik satu, kurang manis, ikat tepi.What you typically get back
Your drink, first time. You chose who to ask (the tea maker), told them what they could not guess (less sugar), and said exactly what to hand over (teh tarik, packed to take away).
The AI is the tea maker. It is skilled and fast, but it has never met you. AIM is your full order: Actor is who you ask, Input is what they cannot guess, and Mission is exactly what you want handed back.
Why a clear order beats clever wording
Every request starts with a huge number of possible answers. A vague PromptThe message or instructions you type to an AI. A better prompt usually means a better answer. leaves all of them open, so the AI drifts towards a safe, middle-of-the-road reply. Each part of AIM closes off answers you did not want.
Switch A, I and M on and off to watch the answers narrow.
You do not need fancy phrasing or secret words. A plain, complete order works better than a clever, incomplete one.
Anatomy of an AIM prompt
Here is a real request from a lecturer on a diploma accounting course. Tap each coloured part to see what it does.
New words in that example: OBE (outcome-based education)The approach used in Malaysian higher education where courses are designed around what students should be able to do at the end., CLO (course learning outcome)A statement of what a student will be able to do after a course, starting with a measurable verb such as "construct" or "analyse"., RubricA marking guide that lists the criteria and describes what each level of performance looks like. and Analytic rubricA rubric that scores each criterion separately.. Tap any of them for a one-line meaning.
A for Actor: choose who answers
The Actor tells the AI which knowledge, priorities and vocabulary to bring forward. Think of it as choosing which colleague to ask. “You are a helpful assistant” adds nothing, because that is already the default.
A useful Actor has three ingredients:
- A role: “assessment designer”, “senior accounts executive”, “safety officer”.
- A setting: “in a Malaysian polytechnic”, “at a small trading company”, “preparing for an MQA audit”.
- An audience: “who writes for first-year students” changes the tone and difficulty for you.
You can also choose a critic on purpose. “You are a sceptical external examiner” is a good way to get honest criticism of your own draft instead of praise.
I for Input: give it what it cannot guess
The AI has never met your students, seen your faculty template or read your notes. Input is everything it needs to know but cannot work out alone. Before you send a request, run through five quick checks:
- Context: who, where, what level, and why the task exists.
- Material: the text, figures or notes to work on. Paste them or attach them.
- Rules: scales, word limits, templates, policies and deadlines.
- Example: one sample of the style you want often says more than a paragraph of description.
- Limits: what it must not do, such as “use only the notes I give you” or “do not add references”.
M for Mission: say exactly what you want back
The Mission is where most prompts go wrong. Usually it is one vague verb: “help”, “look at”, “improve”. A strong Mission starts with a clear action verb and answers four questions:
| Question | Weak | Strong |
|---|---|---|
| What action? | help me with | construct, rank, rewrite, summarise |
| What exactly? | a rubric | an analytic rubric for CLO2 |
| How much? | — | 4 criteria × 4 levels, under 300 words |
| What form? | — | a table with a weighting column totalling 100% |
The verb leads the Mission, and it matters more than people expect. Our guide on psychomotor verbs, the precise action verbs lecturers already use, shows how a sharper verb makes the answer noticeably sharper and easier to check.
Here is the same idea in the office. Tap between the vague and the precise version.
Write a notice about the claims deadline.What you typically get back
A long, formal notice that never states the actual date clearly, says nothing about late claims, and needs a full rewrite before you can send it.
You are the admin officer in a polytechnic finance unit who writes short, friendly notices for academic staff.
From 1 November, travel claims must reach the finance unit by the 5th of the following month. Late claims move to the next payment cycle. Many staff missed this rule last semester. Contact: the finance unit counter, extension 2104.
Write a notice of no more than 120 words in simple English. Give it a subject line, put the rule in one bold sentence, and add three bullet points: the deadline, what happens if a claim is late, and who to contact.What you typically get back
A short notice you can check in ten seconds: subject line, one bold rule, three bullets, under 120 words. It usually needs a light edit, not a rewrite.
Build your own AIM prompt
Your turn. Choose a scenario, pick an Actor, tick the Inputs you would include, then choose a Mission. Watch the prompt build itself and the score change. Try the weak options too: seeing the score drop is the fastest way to feel why each part matters.
AIM for AI agents
A chat app answers once. An AI agent plans, opens files, uses tools and takes several steps on its own, like a new staff member working through a job sheet. That makes AIM even more important, because a misunderstanding at step one repeats at every step after it.
When you brief an agent, add three things to the Mission:
- Done criteria: “you are finished when all 40 files appear in the table”.
- CheckpointA planned pause where the AI shows its work and waits for your OK.: “show me the first 5 rows before you continue”.
- Limits: “do not email anyone; save drafts only”.
You are a careful teaching assistant who checks tutorial submissions against a checklist.
The folder "Tutorial 5" contains 40 anonymised submissions named S01 to S40. Student names have been removed. Question 3 asks students to prepare a bank reconciliation statement. Use only these files.
Open every submission and save a table with the columns File, Q3 attempted (Yes / No / Partly) and Evidence (the first line of their Q3 answer, or "none"). Stop after the first 5 files and show me the table so I can check you are reading them correctly. If a file will not open, write "Could not read" instead of guessing. Do not mark or comment on the answers.The agent can save that table as a CSV fileA simple spreadsheet file Excel or Google Sheets can open. that opens in Excel. Our guide to briefing an AI agent covers the full agent loop.
For developers: technical detailsOptional: skip this if you only use the chat app
If you call a model through an APIA way for software to talk to an AI directly, without the chat app. Developers use it to build AI into their own systems., AIM maps neatly onto the request:
- Actor usually goes in the System promptStanding instructions set up before the conversation starts, such as "always answer in Bahasa Melayu". They apply to every reply., so it applies to every turn.
- Input goes in the user message. Wrap pasted material in clear labels such as
<notes>…</notes>so the model can tell instructions from material. - Mission goes at the end of the user message, after the material, so it is the last thing the model reads.
- For agents, put done criteria, checkpoints and limits in the task description or the project’s instruction file, and give the agent the narrowest file and tool access that does the job.
Common mistakes
- Stacking roles. “You are a lecturer, an accountant, a data scientist and a poet” blurs the view. Pick one, or ask for separate reviews from separate roles.
- Burying the Mission. Put what you want at the start or the end, never in the middle of a long paragraph.
- Forgetting the format. If you need a table, a list or 150 words, say so. Reformatting is wasted time.
- Treating AIM as one message only. AIM is your opening order. Reply, correct and refine. The Chris Voss guide shows how to turn it into a conversation.
Check your understanding
Nice work!
Lesson complete
You made it to the end. Take the quiz above if you have not yet, for bonus XP.
- 0 XP
0 day streak