Chapter 3 — The Art of the Prompt¶

Part I — Generative AI & Responsible AI foundations

In 30 seconds¶

The core idea: a strong prompt gives Copilot four things — a clear Goal, useful Context, the right Source, and your Expectations for the output. This "GCSE" framework is Microsoft's own.
Why it matters: the quality of the answer depends heavily on the quality of the prompt. Prompting is tested conceptually on AB-731 and practically on AB-730.
The exam angle: expect questions on the impact of prompt engineering, the four elements of a good prompt, iteration, and choosing which files or data to reference.
Remember: Goal · Context · Source · Expectations. If an answer disappoints, a missing element is usually why.

Exam map¶

Exam map — AB-731 · Domain 1: prompt engineering (impact and techniques) · AB-730 · Domain 2: create effective prompts

1. Key concepts¶

A prompt is simply the natural-language instruction you give Copilot. But how you phrase it — and what context and sources you attach — changes the result dramatically. That deliberate craft is prompt engineering.

📖 Definition — Prompt engineering: the practice of designing and refining prompts (instructions, context, and referenced sources) to get more accurate, relevant, and useful AI output.

Microsoft teaches a simple, exam-relevant model: an effective prompt contains four elements.

📌 Key concept — the four elements (Goal · Context · Source · Expectations):

Goal — what you want Copilot to do or produce ("Draft a customer email…").

Context — why you need it and any background ("…for a client who missed a deadline, keep it warm").

Source — the specific data Copilot should use ("…based on the attached thread and this proposal").

Expectations — the desired format, tone, length, or audience ("…three short paragraphs, friendly, under 150 words").

You don't always need all four, but the more high-stakes the task, the more each one matters.

An example, built up element by element¶

Version	Prompt	Why it's better
Weak	"Write an email."	No goal detail, no context, no source, no format
+Goal	"Write an email declining a vendor's proposal."	Clear task
+Context	"…they're a long-term partner we want to keep."	Steers tone
+Source	"…based on the attached proposal and my notes."	Grounds in real content
+Expectations	"…polite, 2 short paragraphs, offer to revisit next quarter."	Shapes the output

💡 Tip: reference your Source explicitly. In Microsoft 365 Copilot you can point at a file, an email, a meeting, or a person using / (for example, typing / to reference a document). Grounding the prompt in the right content (Chapter 2) is often the single biggest quality lever.

2. How it works¶

Core techniques¶

Beyond the four elements, a handful of techniques reliably improve results:

Be specific and use action verbs — "summarize", "compare", "rewrite", "list" beat vague asks.
Give the model a role or audience — "You are my finance analyst… explain this to a non-technical executive."
Provide examples (few-shot) — show one or two examples of the format or style you want.
Break big tasks into steps — ask for an outline, then expand; or chain prompts across a conversation.
Iterate — treat the first response as a draft. Refine: "make it shorter", "add a data table", "use a more formal tone."
Set constraints — length, tone, format (table, bullet list, email), and what to exclude.

flowchart LR
    A["Draft prompt<br/>(Goal · Context · Source · Expectations)"] --> B["Review response"]
    B --> C{"Good enough?"}
    C -->|No| D["Refine: adjust tone,<br/>length, add source, clarify"]
    D --> B
    C -->|Yes| E["Use / save the prompt"]

🔍 How it works: prompting is a loop, not a single shot. Because generation is probabilistic (Chapter 1), iterating and adding context is how you steer the model toward what you actually want.

Selecting the right sources¶

Choosing which content to reference is itself a tested skill. Good source selection means pointing Copilot at material that is relevant, current, authoritative, and permission-appropriate.

🎯 Exam tip: when a question asks which resource to reference, pick the most relevant and authoritative source for the task — the specific project file, the latest report, the actual meeting — not "all files" or an unrelated document. Referencing the right source is more effective than a longer prompt.

3. In the real world¶

Scenario — a status update that writes itself. A project lead needs a weekly status email. A vague "write my status update" yields generic filler. Instead she prompts:

"Goal: Draft my weekly status email to the steering committee. Context: we slipped one milestone but recovered two others; keep it confident, not defensive. Source: use the attached project plan and this week's Teams channel. Expectations: 4 short bullets under a one-line summary, executive tone, under 120 words."

Copilot grounds in the referenced plan and channel, and returns a tight, on-message draft she edits in under a minute. The four elements did the work.

4. Exam tips¶

🎯 Exam tip: memorize the four elements — Goal, Context, Source, Expectations. Questions often show a weak prompt and ask what would most improve it; the answer is usually "add the missing element" (most often Source or Expectations).

🎯 Exam tip: prompt engineering ≠ coding. On these business exams it means writing better natural- language instructions and choosing good sources — not programming.

🎯 Exam tip: iteration is a feature. "Refine the previous response" is often the best next action rather than starting a brand-new prompt, because the conversation carries context.

5. Common pitfalls¶

⚠️ Pitfall: the "one perfect prompt" myth. Expecting a flawless answer from a single vague prompt leads to disappointment; strong results come from specificity and iteration.

Vagueness: "make this better" gives the model nothing to optimize for — better how?
No source: asking about "the project" without referencing the project file forces a generic answer.
Overstuffing: a rambling prompt buries the goal; be complete and concise.
Wrong source: referencing an outdated or unrelated document degrades the answer — relevance beats volume.
Forgetting expectations: if you don't specify format/length/tone, you get the model's default, which may not fit.

6. Practice questions¶

1. Which set best describes the four elements of an effective Microsoft 365 Copilot prompt?

A. Goal, Context, Source, Expectations
B. Question, Answer, Format, Length
C. Input, Output, Model, Token
D. Who, What, When, Where

Answer

Correct: A. Microsoft's framework for effective prompts is Goal, Context, Source, Expectations. The other options aren't the taught model.

2. A user prompts "Summarize the project" and gets a vague answer. What would most improve the result?

A. Repeat the exact same prompt
B. Reference the specific project file or meeting as the Source
C. Switch to a different language
D. Ask Copilot to use more tokens

Answer

Correct: B. The prompt lacks a Source; grounding it in the relevant file or meeting is the biggest improvement. A changes nothing; C is irrelevant; D isn't a user control and doesn't address the missing context.

3. In the context of these certifications, "prompt engineering" primarily means:

A. Writing code to fine-tune a model
B. Crafting clear natural-language instructions and choosing good sources to improve output
C. Configuring server infrastructure
D. Training a new large language model

Answer

Correct: B. For business users and leaders, prompt engineering is about better instructions and source selection — no coding or model training involved. A, C, and D describe technical activities outside the prompt itself.

4. A first Copilot response is close but too long and too formal. What is the best next step?

A. Start over with a completely new chat
B. Iterate: ask Copilot to shorten it and use a warmer tone
C. Accept it as-is
D. Report the model as broken

Answer

Correct: B. Iterating within the conversation preserves context and refines the draft efficiently. A discards useful context; C ignores the quality gap; D is unwarranted.

5. Which prompt best applies the four elements?

A. "Write something about sales."
B. "Draft a 3-bullet summary of Q3 sales for the leadership review, using the attached sales report, in a concise executive tone."
C. "Sales report?"
D. "Make a great sales document, you decide everything."

Answer

Correct: B. It states the Goal (3-bullet summary), Context (leadership review), Source (attached sales report), and Expectations (concise executive tone). The others are vague and missing most elements.

Chapter 3 — The Art of the Prompt¶

In 30 seconds¶

Exam map¶

1. Key concepts¶

An example, built up element by element¶

2. How it works¶

Core techniques¶

Selecting the right sources¶

3. In the real world¶

4. Exam tips¶

5. Common pitfalls¶

6. Practice questions¶

Further reading¶