# Rewrite a Vague Request into a Clear Prompt

> LOs: LO-S1-2 — Rewrite a vague work request into a clear prompt an AI agent can act on.

## Introduction

You have probably typed something like "Help me with the defect report" or "Look at the sprint data" into an AI tool and gotten back a generic, unhelpful answer (see Slide 1). The fix is not a longer prompt — it is a **structured** one. In this class you will learn a four-step checklist that turns any vague work request into a prompt an AI agent can actually act on, in under a minute.

## Core Concept

Vague prompts produce vague answers because the AI is missing four specific ingredients. The 4-Step Rewrite Checklist gives you each one in order (see Slide 2):

| # | Step | What you add | Why it matters |
|---|------|--------------|----------------|
| 1 | Spot the vague verb | Replace "help," "look at," "summarize," "check" with a concrete task verb (prepare, draft, classify, flag) | Vague verbs hide the real job. The AI cannot pick a strategy if the verb does not commit to an output. |
| 2 | Add role + audience | "Acting as a QC engineer… for the plant manager" | Role sets the voice and depth. Audience sets the vocabulary and the level of detail. |
| 3 | Add context | The data source, the timeframe, the background facts | Without context the AI invents plausible-sounding but wrong specifics. |
| 4 | Add constraints | Length, format, tone, threshold rules, what to leave out | Constraints turn the output into something you can actually check and use. |

Treat these four steps as a **repeatable recipe**, not a one-off. After a week or two of running every prompt through the checklist, the structure becomes automatic and your hit rate climbs sharply.

## Worked Example

Let's run our Manufacturing Engineer's prompt through the recipe (see Slide 3).

**Before:** *"Help me with the defect report."*

**After:** *"Acting as a QC engineer, prepare a defect summary for the plant manager, using last week's injection-mold inspection log. Focus on the top three recurring defects. Keep it under 200 words, use a bulleted format, and flag any defect with more than 5 occurrences."*

Each ingredient is doing real work in that after-prompt:

- **Role** — *"Acting as a QC engineer"* — sets the technical voice and the level of domain assumption.
- **Task verb** — *"prepare a defect summary"* — replaces the vague *"help me"* with a concrete output.
- **Audience** — *"for the plant manager"* — tells the AI to skip jargon a manager would not need and lead with impact.
- **Context** — *"last week's injection-mold inspection log… top three recurring defects"* — pins down the data source, timeframe, and scope.
- **Constraints** — *"under 200 words, bulleted, flag defects with more than 5 occurrences"* — makes the output checkable. You can literally count words and verify the threshold rule.

Notice the prompt is roughly four sentences, not four paragraphs. Clear is not the same as long.

## Common Pitfalls

- **Stopping after Step 1.** Many learners catch the vague verb, swap in "prepare" or "draft," and call it done. Without role, context, and constraints the AI is still guessing. Always run all four steps.
- **Adding context but no constraints.** A prompt with rich context but no length, format, or threshold rules tends to produce a wall of text you cannot quickly verify. Constraints are what make the output reviewable.
- **Over-specifying constraints.** The opposite trap: piling on ten format rules, a word count, a tone, a banned-words list, and three thresholds. The model starts juggling rules instead of solving the problem. Two or three sharp constraints beat ten fuzzy ones.

## Recap

You now have a 4-step checklist — spot the vague verb, add role and audience, add context, add constraints — that rewrites any vague request into a prompt an AI agent can act on (see Slide 5). That is LO-S1-2, done. Coming up in Section 2, you will layer **prompt patterns** on top of this checklist — reusable templates like role-play, few-shot, and chain-of-thought that make these clear prompts even more powerful.