8 min read
Beginner
Draft: this article is under editorial review and may change.Think back to getting your driving licence. First you sat the theory test: road signs, right of way, what a double line means. Then came the practical test: the hill start, the side parking, and finally real traffic. Passing the theory did not prove you could drive. Driving well in the circuit did not prove you knew the rules.
Every piece of office work has the same two halves. Take a KRT grant claim. Someone must know the grant rules well enough to explain them and judge a claim against them. Someone must also do things: spot the missing receipt, follow the review template, write the letter to the treasurer.
The previous page showed that a clear verb makes a finding easy to trace. This page gives you two ready-made lists of verbs, one for each half. Bloom describes what the work must know and reason about. Simpson describes what it must do and adapt. Together they tell an AI helper, or a new officer, exactly what the job is.
The idea in 30 seconds
Vague verbs, vague work
"Handle", "look at" and "sort out" leave the AI to guess the job.
- Vague verbs, vague work "Handle", "look at" and "sort out" leave the AI to guess the job.
- Bloom: what it must know Thinking verbs: explain the rules, compare two cases, judge a claim.
- Simpson: what it must do Doing verbs: notice, follow, build, adapt, design.
- Climb only as high as needed Follow a template → adapt a draft → design something new.
- Higher means check more A precise verb makes the answer easier to check, not automatically right.
- Know + do = a work activity you can hand over.
Key takeaways
If you remember nothing else, remember these.
- Vague verbs such as handle, look at, sort out and check produce vague work, because the AI cannot tell what “done” means.
- Bloom’s verbs (explain, compare, evaluate) say what the work must know and reason about.
- Simpson’s verbs (identify, follow, adapt, design) say what the work must do, from noticing cues to creating something new.
- The higher you climb either ladder, the more judgement you hand over, and the more carefully you check. The officer still decides.
Why “handle this” gets you an average answer
Imagine telling a new assistant “handle the KRT claims”. Do you mean check them, sort them, approve them or file them? A good assistant asks. An AI usually does not. It does a little of everything and carries on.
Take “improve this press statement”. That could mean fix the grammar, shorten it, soften the tone or translate it. The AI usually does a little of each, and you spend ten minutes working out what changed.
There is a simple test for any verb: how would I know it was done? If you cannot see it happen or check the result, the verb is too vague.
| Vague verb | Sharper verb | What you can check |
|---|---|---|
| handle | apply the checklist to | a Met / Not met mark against every item |
| look at | categorise | every report sits in exactly one group |
| improve | revise … so that … | the condition after “so that” is met |
| check | identify | a list of the specific gaps you named |
| think about | compare | a side-by-side table with named criteria |
Two questions about every task: know and do
Teachers and trainers have long sorted learning into three kinds, each with its own ladder of verbs. We borrow two of them to describe work.
Three kinds of verb
Think and know (Bloom)
From remembering to creating. Verbs: explain, compare, evaluate.
- Think and know (Bloom) From remembering to creating. Verbs: explain, compare, evaluate.
- Do and adapt (Simpson) From noticing cues to designing new ways of working. Verbs: identify, follow, adapt, design.
- Feel (affective) Attitudes and values. Verbs: value, respect. These stay with people.
Sources: the thinking ladder is Bloom's taxonomyA ladder of thinking skills, from remembering to creating, as revised by Anderson and Krathwohl in 2001. as revised by Anderson and Krathwohl (2001). The doing ladder used here is Simpson’s (1972). The feeling levels are usually credited to Krathwohl and colleagues (1964).
An AI has no hands. It cannot walk a KRT boundary or sit at a mediation table. So we are not asking it to perform Psychomotor (domain)Skills about doing things with hands or body, such as setting up a venue or operating equipment. skills. We borrow the habit behind Simpson’s verbs: each one names an action you could observe and a result you could inspect.
Feeling verbs we leave aside. Building trust with two neighbours in a dispute, or showing respect at a community event, is an officer’s work. An AI can draft words; it cannot genuinely care.
Bloom: what the work must know and reason about
Bloom’s ladder runs from simply recalling facts to building something new from them. Each rung asks for a different kind of thinking. Here it is with KPN work on every step. All the names are fictional.
| Level | What it asks | Example task |
|---|---|---|
| Remember | recall facts you give it (list, name) | List every document a KRT grant claim must include, using only the guideline pasted below. |
| Understand | put it in plain words (explain, summarise) | Explain the KRT grant rules in plain BM for a new committee treasurer. |
| Apply | use a rule on one case (apply, calculate) | Apply the claim checklist to the claim from KRT Taman Seri Contoh. |
| Analyse | break apart and compare (compare, trace) | Compare two anonymised mediation cases: cause, parties, what was agreed. |
| Evaluate | judge against criteria (rank, justify) | Rank three venues for a community dialogue against four given criteria. |
| Create | build a new plan (propose, plan) | Propose two follow-up options after a mediation, with pros and cons. |
The lower rungs work only on what you give the AI. It does not know your department’s grant guideline or your district, so paste the rule and the case. Higher rungs ask it to judge, and a judgement can sound confident and still be wrong.
You are an experienced community mediator who writes short case notes for a district unity office.
Below are two anonymised case summaries, Case A and Case B. Both are disputes between neighbours in Taman Seri Contoh. Names, addresses and phone numbers have been removed. Use only these two summaries.
Compare the two cases in a table with the rows Cause, Parties involved, What each side wanted, What was agreed, and Follow-up needed. Then list two lessons that apply to both. If a summary does not say something, write "not stated" instead of guessing.The verb “compare” tells the AI what kind of thinking to do. The table tells you what to check. You read both summaries yourself before using the lessons.
Simpson: what the work must do
Simpson’s levels describe doing, as a ladder. At the bottom, you notice cues and get ready. In the middle, you copy a demonstration, then work alone, then work fluently. At the top, you adapt the skill to new conditions and finally design something new.
The same ladder describes how much you are asking of an AI helper.
Tap a step to see an example prompt that uses its verb.
Doing levels (Simpson, 1972): The doing-verb ladder (Simpson): P1 to P7
Lower steps: the AI follows your pattern.
Higher steps: you hand over more judgement, so you check more carefully.
In daily KPN work, four groups of rungs do most of the work:
- Perceive: “Identify every sentence in these community reports that could signal rising tension.” The AI notices cues. You decide which ones matter.
- Set and guided response: “Prepare a list of what is missing, then follow our minutes template to draft the minutes.” The AI gets ready, then copies your pattern. The result is predictable and easy to check.
- Mechanism and adaptation: “Adapt this BM press statement for 中文 or தமிழ் readers, keeping every fact unchanged.” The AI changes something that already works. You compare it with the original.
- Origination: “Design a new Bulan Perpaduan programme format that brings young people and elders together.” The AI invents. This can be the most useful and is always the riskiest, because there is nothing to compare it with.
The practical rule: climb only as high as the task needs. If you have an approved template, ask the AI to follow it. If you have something that nearly works, ask it to adapt. Ask it to design only when you truly want something new and have time to test it.
Sharpen a prompt yourself
Pick a scenario below. The highlighted words are vague verbs. Tap one, choose a sharper verb and watch the prompt rewrite itself. Each option shows whether it is a Bloom (thinking) or Simpson (doing) verb, and how high it sits.
Notice two things as you play. First, the sharper options are longer: a precise verb pulls an object and a standard in with it. Second, mixing both kinds in one prompt is normal. “Identify the gaps, then explain the rule behind each one” is a doing step followed by a thinking step, on purpose.
Examples from KPN work
Every example below uses fictional names. Before you paste real material into the AI tool your department has approved, remove names, MyKad numbers, addresses and phone numbers, and follow your department’s own data and security rules.
Perceive: early-warning signs in community reports
Spotting cues is the lowest doing rung, and one of the most useful. The AI reads a pile of reports quickly and points at lines you might miss. It does not decide what is serious.
You are a careful assistant to a district unity officer who reads community reports for early signs of tension.
Below are 12 anonymised community reports from this week, numbered R1 to R12. All names and addresses have been removed.
Identify every sentence that could signal rising tension between groups. Quote each sentence with its report number, and give a one-line reason. Do not rate how serious it is and do not suggest action. If nothing in a report qualifies, write "none".You then read the quoted lines in context and apply your department’s escalation rules. The AI points; the officer judges.
Set and guided response: minutes from a template
Minutes follow a fixed shape, so this is a “follow” job. First let the AI prepare: list what is missing, such as the attendance list or last month’s minutes. Then let it follow your template heading by heading. Because the shape is yours, any gap shows up straight away.
Adaptation: one statement, several audiences
A press statement written in BM often needs a 中文 or தமிழ் version. “Translate this” invites a word-for-word copy. “Adapt this” asks the AI to keep the facts and fit the reader.
Translate this press statement into Chinese.
What you typically get back
A literal translation. The programme name, the date format and a BM saying come across awkwardly, and nothing tells you which lines to double-check.
You are a communications officer who prepares unity-programme statements for Chinese-language newspapers in Malaysia.
Below is the approved BM press statement for a fictional Hari Jiran programme at KRT Taman Seri Contoh.
Adapt the statement for 中文 newspaper readers. Keep every fact, date, venue and contact unchanged. Flag every phrase you changed in meaning or tone, in a table: BM original, 中文 version, reason.What you typically get back
A version that reads naturally, plus a short table of every judgement call. A fluent officer checks the table and the full text before anything is released.
Origination: a new programme format
When you want something that does not exist yet, “design” is the right verb, and the checking load is highest. Ask for two or three options with trade-offs, not one answer. Then test the idea with the people who will run it, such as a KRT committee, before it goes into a plan.
From verbs to work activities
Bloom and Simpson give you the verb. The next page, Work activities, puts that verb into a full description of a job: who does it, on what, with which tool, from which input, producing which output, who decides, and how you measure it.
A useful pattern to start with is:
verb + object + standard + format
- Compare (verb) two mediation cases (object) on cause, parties and agreement, using only the summaries given (standard), in a table with a “not stated” cell for gaps (format).
Write the verb first and the rest tends to follow. Once you have typed “Compare”, you ask yourself “compare on what?”, and the standard appears.
Common traps
- Stacking verbs. “Analyse, evaluate and improve this report” asks for three jobs at once. Split them into steps and check each before moving on.
- A verb without a standard. “Rank these venues” lets the AI choose the criteria. Say what to rank by.
- Climbing too high. Asking the AI to “design” a new minutes format when you only needed it to “follow” the old one creates far more to check.
- Trusting the verb instead of the answer. A precise verb makes the answer easier to check, not automatically correct. AI drafts; the officer checks the facts, the figures and the tone, and decides.
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