9 min read
Beginner
Draft: this article is under editorial review and may change.
It is week 12. You have 140 scripts to mark, a CQI (continuous quality improvement)Recording what you will improve next time, based on results. report due, and next semester’s 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". still need revising. AI can take a real share of this load.
It can also produce a RubricA marking guide that lists the criteria and describes what each level of performance looks like. that measures the wrong thing, an answer key with wrong numbers, and feedback that breaks your university’s data rules. This guide shows you how to get the help without the harm.
The idea in 30 seconds
Start from the CLO
Quote it word for word in every teaching and assessment prompt.
- Start from the CLO Quote it word for word in every teaching and assessment prompt.
- AI drafts Rubrics, questions and comments: fast first versions.
- You decide Marks, levels and every claim stay with you.
- Anonymise first Names, matric and IC numbers never go into the AI.
- Test on real scripts Try every rubric and question before students see it.
- AI drafts; you decide.
Key takeaways
If you remember nothing else, remember these.
- Start from the CLO. Quote it word for word in every teaching and assessment prompt.
- Build each prompt in five layers: AIM, Action verbs, Named methods, Calibrated dialogue, then Verify.
- AI drafts; you decide. Final grades, misconduct decisions and accreditation claims stay with you.
- Anonymise student work before it touches any AI tool, and check your institution’s AI policy first.
One course, followed from start to finish
To keep things concrete, this guide follows one fictional course all the way through: Financial Accounting 2, in a Diploma in Accounting. Its key outcome is CLO3: “Analyse the cash-flow position of a small Malaysian business.”
Engineering examples appear along the way too, because the same steps work in any subject.
The alignment chain, with AI in the right places
Constructive alignmentOutcome, teaching and assessment all aim at the same skill. means your teaching and your assessment both point at the same outcome. Think of a recipe: the CLO is the dish you promised, and every activity and test is a step towards it. It is the backbone of 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. practice in Malaysian universities and polytechnics.
It is also the best way to organise AI work. If the CLO is right, every later prompt has a precise target. If the CLO is vague, AI will happily build a beautiful rubric for the wrong target.
Tap a step to see who does what, or press Play.
AI shows up at every step as a fast first drafter. You show up at every step too, but in a different role: you decide, test, adapt and sign off. That split makes the workflow defensible when a student or a panel asks, “Who decided this?”
Five layers in every prompt
Every prompt below is built from the same five layers. Think of them as the boxes on a job sheet you would hand a new tutor:
- AIM: who the AI should be (Actor), what it needs to know (Input) and exactly what to produce (Mission).
- Action verbs: one precise verb, such as “diagnose” or “construct”, instead of “help me with”.
- Named methods: a known approach called up by name, so you don’t have to spell out every step.
- Calibrated dialogue: an open question that makes the AI state its assumptions before it acts.
- Verify: you check the output against the CLO and your own expertise.
Step 1: Fix the CLO first
Most Malaysian programmes map each CLO to a domain and a level. Thinking skills run C1 to C6, from Bloom's taxonomyA ladder of thinking skills, from remembering to creating. as revised by Anderson and Krathwohl. Doing skills run P1 to P7, commonly cited to Simpson. Attitudes run A1 to A5, from Krathwohl and colleagues.
A good CLO has one measurable verb at the right level and a clear object. Take a draft from our accounting course: “Understand and prepare a statement of cash flows.” You can’t observe “understand”, and there are two verbs in one outcome.
AI is a good second pair of eyes for this check. It is a diagnostic task: you are asking “what is wrong here, and why?” That is exactly where the Chris Sanders method fits, so this prompt uses it.
You are an OBE curriculum specialist who reviews course learning outcomes for Malaysian diploma programmes.
Course: Financial Accounting 2, Diploma in Accounting, Semester 3. My four draft CLOs and the programme outcomes (PLOs) they should support are pasted below. We map CLOs to C (revised Bloom), P (Simpson) and A (Krathwohl) levels.
Diagnose each draft CLO, then return a table: original CLO | domain and level | problem, if any | rewritten CLO starting with one measurable verb | PLO supported. Keep each rewrite under 25 words.
Use the Chris Sanders method to complete this task. First, restate in three bullets how you will apply it here.
Before you rewrite anything, what would you need to know about my students and our assessment plan to get the levels right?The question at the end matters more than it looks. It invites the AI to ask about your students or how the CLO is assessed before it commits to a level. Answer it, then let the AI continue.
The final wording, the level and the PLO (programme learning outcome)What graduates of the whole programme should be able to do. mapping are still yours. Check them against your programme documents and your faculty’s approved verb list.
Step 2: Plan lessons that produce evidence
Once a CLO is fixed, ask AI to plan activities that let students practise it. Analysing cash flow (C4) needs students to compare and explain real figures, not just copy a format. A lab skill at P3, such as setting up a pump test in Fluid Mechanics, needs guided hands-on practice.
Give the AI your real limits: class size, room, equipment, contact hours and the SLT (student learning time)Total hours a student is expected to spend on a course. breakdown in your course pro forma. Ask for a table with minute ranges, so you can see at once whether the plan is realistic.
Robert Gagné's nine steps for a lesson, from gaining attention, through practice and feedback, to helping students remember and use what they learned. is a useful named method here. It tends to give each session a clear hook, practice and feedback.
Then walk through the plan as if you were teaching it tomorrow. Remove anything that won’t work with one test rig per six students, or a computer lab booked for only an hour. AI has never stood in your classroom.
Step 3: Design the assessment and an analytic rubric
The most common failure in AI-drafted rubrics is drift. Criteria slide towards things that are easy to describe, such as “presentation” or “grammar”, instead of the thinking the CLO names. It is like judging a nasi lemak contest on the plate instead of the taste.
So your prompt must quote the CLO, state its level, give your faculty’s scale and ask the AI to flag any criterion that misses the outcome. Tap the toggle below to compare a vague request with a precise one.
Create a rubric for my accounting group assignment.
What you typically get back
A generic 5-level rubric with criteria such as Content, Organisation, Teamwork and Language. Only one criterion relates to cash-flow analysis, and the levels do not match your faculty scale.
You are an assessment designer who writes fair, transparent rubrics for first-year diploma students with mixed English proficiency.
CLO3: Analyse the cash-flow position of a small Malaysian business (C4). Task: group report on a fictional kuih business, 25%. Faculty scale: Excellent (4), Good (3), Satisfactory (2), Weak (1).
Construct an analytic rubric with 4 criteria and 4 levels as a table, one plain-English sentence per cell, weights totalling 100%. Then list any criterion that does not directly measure CLO3.
Apply constructive alignment (Biggs): every criterion must trace back to CLO3.
How might a weak student misread these descriptors?What you typically get back
A rubric whose criteria all measure cash-flow analysis, on your scale, with a short list of risks. You test it on two anonymised scripts from last semester and adjust two descriptors.
Testing the rubric is the step people skip. Before you release it, mark two or three real, anonymised scripts with it. If two strong scripts land in different bands for reasons you can’t defend, fix the descriptors.
Then carry out ModerationA colleague checks your marking for fairness and consistency. with a colleague, as your department normally does. An Analytic rubricA rubric that scores each criterion separately. makes this easier, because you can compare scores criterion by criterion.
Step 4: Build a question bank across levels
AI can write many practice questions very quickly, at any Bloom level you ask for, with answer keys. That saves real time for tutorials and parallel exam versions. It is also where the most embarrassing mistakes happen, because generated answer keys can contain slips and confident wrong reasoning.
You are an experienced examiner for a diploma course in Financial Accounting at a Malaysian polytechnic.
Topics: errors that affect the trial balance, and bank reconciliation. CLO2 is C3 (apply). Use Malaysian small-business settings, such as a kedai runcit or a catering business, with clearly invented figures.
Generate 9 questions: 3 at C2, 3 at C3 and 3 at C4. Include one C4 item where the trial balance is out by RM1,240 and students must find the likely causes. For each question, give the level, the question, a worked answer and one common student error, as a numbered list.
Align each item to its revised Bloom level (Anderson and Krathwohl), and name the verb that sets that level.
Which of these questions would you trust least, and why?The rule is simple: never release a question you haven’t solved yourself. Accounting gives a neat test. A single transposed figure, such as 540 keyed in as 450, always leaves a difference divisible by 9, and RM1,240 is not.
So an answer key that blames one transposition is wrong. A RM620 entry posted to the wrong side, which doubles the difference, is a better candidate. Catching that kind of slip is your job, not the AI’s.
Also check that each question really sits at its claimed level. A “C4” item that only asks students to fill in a format is a C3 item with a longer stem. Keep the bank as securely as any exam paper.
Step 5: Create local case studies, clearly fictional
Local cases often feel more relevant to students: a kuih maker in Johor, a logistics start-up in Klang, an electronics plant in Penang. AI is good at drafting these, with a decision point, a data table and discussion questions.
Two cautions. First, ask for an explicitly fictional case with invented names and figures, and label it as fictional on the handout. Models sometimes produce names that match real companies or people.
Second, models can invent regulations, grant schemes or statistics that sound official. Check every real law, agency or market figure against an official source. Then ask the AI directly: “What assumptions about Malaysian SMEs did you make that I should check?”
Explore the whole workflow
Tap through the seven steps below. Each one shows what AI can do, what you must do, a prompt to copy and the main risk to watch.
Tap a step (or use the arrow keys). Each one shows what AI can do, what stays with you, a prompt you can copy and a risk to watch.
Step 1 of 7
Align course learning outcomes
What AI can do
- Diagnose vague or unmeasurable verbs in your draft CLOs.
- Suggest a domain and level (C, P or A) for each outcome.
- Check that each CLO maps to at least one PLO.
What you must do
- Decide the final wording and level.
- Confirm the mapping against your programme documents.
- Get sign-off through your faculty process.
You are an OBE curriculum specialist who writes course learning outcomes for Malaysian diploma and degree programmes.
Course: Financial Accounting 2, Diploma in Accounting, Semester 3. My draft CLOs and our programme outcomes (PLOs) are pasted below. We map CLOs to C (revised Bloom), P (Simpson) and A (Krathwohl) levels.
Diagnose each draft CLO. Return a table: original CLO | domain and level, e.g. C3 | problem, if any | rewritten CLO that starts with one measurable verb | PLO it supports.
Use the Chris Sanders method to complete this task.
Before you rewrite anything, what would you need to know about my students and PLOs to get the levels right?
Risk to watch
The AI may inflate levels (calling a recall task "C4") or pick verbs your faculty does not accept. Check every level against your own verb list.
Step 6: Give feedback at scale, safely
Feedback is where AI can save the most hours, and where you need the most care. Think of a clinic: the doctor examines the patient first, then an assistant helps write up the notes. You mark first; then AI helps you write.
Safe feedback in four steps
Anonymise
Replace names, matric and IC numbers with codes such as S01. Keep the key on your own computer.
- Anonymise Replace names, matric and IC numbers with codes such as S01. Keep the key on your own computer.
- You mark Decide every mark with your rubric. Tell the AI the marks are final.
- AI drafts Short comments tied to rubric criteria, in a set structure.
- You edit and sign Fix anything wrong, generic or unkind. Make it sound like you.
You are a supportive lecturer who writes specific, actionable feedback for diploma students.
My rubric for CLO3 and three anonymised cash-flow reports (S01–S03) are pasted below. I have already decided the marks. Do not suggest or change any mark.
Draft feedback of no more than 120 words per student, tied to rubric criteria, as a table: ID | one strength | one priority to improve | a concrete next step.
Structure each comment using Hattie and Timperley's feed up, feed back, feed forward.
What patterns do you see across these reports that I should address in class?The named method here comes from A well-known feedback model: feed up (where am I going?), feed back (how am I doing?) and feed forward (what should I do next?).. The question at the end turns single comments into teaching insight.
If several reports mix up cash from operations with cash from investing, fix that in next week’s lecture. Then record it for your CQI report.
Step 7: Prepare accreditation and CQI documents
Accreditation under the MQAThe Malaysian Qualifications Agency, which accredits higher-education programmes in Malaysia. needs evidence that your course does what it says: CLO attainment data, assessment samples, moderation records and CQI actions. AI can summarise attainment tables you provide, draft CQI paragraphs linked to the evidence, and check your course file against a checklist you paste in.
Plan, Do, Check, Act: a four-step improvement cycle. Plan a change, try it, check the results, then keep or adjust it. (Plan-Do-Check-Act) is a natural named method for structuring improvement actions. If you want the AI to investigate why CLO3 attainment fell, that is a diagnostic task, and the Chris Sanders method fits.
Two rules keep this honest. Use only numbers you have checked from your own records. And treat any “cause” the AI suggests as a guess to test, not a finding: it doesn’t know the computer lab was closed for three weeks. Always fill in the official MQA and institutional templates.
What AI can draft, and what only you decide
Some decisions should never be handed to AI, however good the output looks.
| AI can draft | Only you decide |
|---|---|
| Practice questions and answer keys | Which questions reach students, after you solve them |
| Comments that explain a mark | Every mark and final grade |
| A summary of attainment data you supplied | Any figure or claim in an accreditation document |
| A tidy template for routine notices | Replies about a student’s health, money or personal crisis |
Academic misconduct needs special care. Don’t use AI output as evidence that a student cheated, and don’t ask a chatbot whether a script was AI-written. Follow your institution’s procedure, which relies on human judgement and due process.
Check your institution’s AI policy first
Policies differ between universities, polytechnics and private colleges, and they change quickly. Before you build any of this into your teaching, find out:
- which AI tools your institution has approved, especially for student data;
- whether you must tell students when AI helped create assessments or feedback;
- what students themselves may use, so your rubric and your rules agree;
- who to ask when a case is unclear, such as your teaching and learning unit.
Being open with students also models the honesty you expect from them. A short note in your course information is often enough: “AI tools helped draft some practice questions; your lecturer checked every one.” The AI for research guide covers disclosure for researchers.
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