Week 4 · Lecture outline

Week 4 — Lecture Outline · Prompting II — Meta-Prompting & Structured Prompts

Using Artificial Intelligence · AI 101 Fall 2026 · Prof. Quinn Fictional sample

Course: Using Artificial Intelligence (AI 101) · Silver Oak University (fictional sample) · Prof. Quinn
Objective covered: Objective 2 — Use effective prompting techniques to produce high-quality, well-verified AI results, including meta-prompting and structured-prompt components.
SLOs touched: A (produce high-quality results through excellent prompting) · B (evaluate and use AI critically)
Meeting pattern: 2 sessions × 75 min = 150 min. Segments total ~150 min; scale to your pattern.

Week at a Glance


The week's big question	"If a prompt is a set of instructions, can you engineer it systematically — and can the AI help you do it?"
By the end of the week, students can…	(1) use meta-prompting to have the AI help write or refine a prompt; (2) name and apply the nine structured-prompt components (Context · Role · Goal · Audience · Constraints · Voice/Format · Data/Logic · Examples · Evaluation); (3) build and test a reusable template for a recurring task; (4) identify over-engineering and missing pieces in a prompt.
Key vocabulary	meta-prompting, structured prompt, prompt template, Context, Role, Goal, Audience, Constraints, Voice/Format, Data/Logic, Examples, Evaluation, prompt engineering, the "perfect prompt" (a moving target), over-engineering
Materials	slides (Deck 4), the week's readings + video links, one approved assistant for live demos and the AI-critique moment, the tutorial
Timing note	8 segments, ~150 min. Session 1 = Segments 1–4 (~75 min). Session 2 = Segments 5–8 (~75 min).

Segment 1 — Hook & the Scenario (8 min) · Session 1 opens

Hook. Project two AI responses to the same underlying request — one generated from a vague one-liner ("help me write a message to my team"), one from a structured prompt with role, audience, goal, and constraints. Let the room see the gap in quality before explaining why it exists. Then ask: "Which prompt did I write first? And how long did it take to go from the weak one to the strong one?" (Answer: the strong one came after asking the AI to ask clarifying questions.)

The promise: "By Friday you'll have a nine-part framework for engineering any prompt, a technique that lets the AI help you build it, and a template you'll actually reuse."

Memory hook: "The prompt is the product — build it like one."

Segment 2 — Skill 4: Meta-Prompting (25 min)

Plain language first. Meta-prompting means using the AI to help write the prompt. Instead of wrestling with the blank page, you ask the AI to interview you: "I need a prompt for [task]. Ask me clarifying questions one at a time — what do you need to know to write a great prompt for me? When you have what you need, return a reusable Markdown prompt."

Why it works. The AI knows what information makes a prompt better — it's been trained on millions of them. It will ask exactly the questions a skilled prompter would: who is this for? What's the desired format? Are there things it must or must not do? Those questions surface requirements you hadn't named yet.

The exact move — live on the projector:
1. Open an assistant (ChatGPT, Claude, Gemini, or Copilot).
2. Type: "I need a prompt for a recurring task. Ask me clarifying questions one at a time until you have enough to write a great reusable prompt. When ready, output a Markdown prompt I can copy and reuse."
3. Answer the AI's questions honestly. Notice that it asks about things you didn't think of (tone, length, audience, constraints).
4. Receive the draft prompt. Read it critically — it will be pretty good, and it will have at least one gap or overreach.
5. Ask it: "What did you assume that I didn't state explicitly? What might go wrong with this prompt?" This surfaces weaknesses.

The Markdown output rule. Asking for Markdown is part of the technique: it forces the prompt into a readable, structured format you can save and reuse, and it makes the components visible.

What to watch for: the AI may add constraints you didn't ask for, or make assumptions about your audience. Review before you trust.

Memory hook: "Ask the AI to ask you — it knows what it needs."

Segment 3 — Skill 5: The Nine Structured-Prompt Components (22 min)

The framework. Every excellent prompt — whether the writer named the parts or not — does most of these nine things. Naming them makes the technique learnable and repeatable.

Component	What it controls	Quick example
Context	The situation, background, or prior work the AI needs to know	"I'm a marketing coordinator at a small nonprofit."
Role	The persona or expertise you're asking the AI to bring	"You are an experienced science communicator."
Goal	What the output must accomplish — the primary task	"Write a 200-word summary that persuades a busy donor to read the full report."
Audience	Who will receive or use the output	"Audience: non-technical board members with no science background."
Constraints	What the output must not do; limits on scope, length, topics	"Do not use jargon. Stay under 200 words. Do not make specific donation asks."
Voice/Format	Tone, style, structure, form	"Tone: warm and urgent. Format: three short paragraphs."
Data/Logic	Specific facts, numbers, or reasoning the AI should use or draw from	"Use this data: [paste chart]."
Examples	Samples that show the style, format, or quality you want	"Here is a summary from last year that hit the right tone: [paste]."
Evaluation	The success criterion — how the AI should test its own output	"Before returning the draft, check: Does this read as non-technical? Is it under 200 words?"

Plain-language teaching: you don't need all nine every time. A grocery-list prompt needs Goal and Constraints; a ghostwritten email needs Role, Audience, Voice/Format, and Goal. The framework is a checklist, not a mandatory nine-paragraph essay.

The misconception to cure right now:
- ❌ "More words = a better prompt."
✅ Cure: A bloated, contradictory prompt is worse than a lean, clear one. Add a component when it changes the output; leave it out when it doesn't. We'll diagnose a bloated prompt together in Segment 7.

Segment 4 — Misconceptions + Quick Interaction (20 min) · Session 1 closes (~75)

Name the misconceptions out loud, then cure each:

❌ "Assigning a 'Role' makes the AI actually expert or accurate."
✅ Cure: the Role shapes style and framing, not factual accuracy. Writing "you are a doctor" doesn't give the AI medical ground truth — it still generates plausible text. You still verify.
❌ "Examples and Constraints are the same thing."
✅ Cure: Constraints say what not to do (no jargon, stay under 200 words). Examples show the format or style you want. Both sharpen output, but in completely different ways — a prompt with good examples but no constraints can still go off in the wrong direction.
❌ "The perfect prompt exists — I just have to find it."
✅ Cure: the "perfect prompt" is a moving target — it depends on the task, the model, and the moment. The goal is a reusable template that gets reliably good results and can be tuned, not one magic string.

Quick interaction — "Which component is missing?" (10 min):
Show three prompts, each missing a different component. Students identify the gap and name the fix. Example:
- "Write a blog post about climate change." → missing Role, Audience, Goal, Voice/Format, Constraints — a great example of nothing being controlled.
- "You are a marine biologist. Write a summary for a general audience." → has Role + Audience; missing Goal, Constraints, Voice/Format. What does the output look like? Length? Tone?
- "Write a weekly status email for my team. Make it short. Don't be negative." → has Goal + Constraints; missing Role, Audience, Voice/Format. Who sends this? To what kind of team?

Segment 5 — "The Medium is the Message" — How the Tutorial Itself Models the Components (18 min) · Session 2 opens

Hook back in: "We've named the nine components. But here's the meta-move: the Lecture Tutorial prompt you'll use this week — the one I wrote for your AI tutor — is itself a structured prompt. Let's reverse-engineer it."

Demonstration: pull up (or project) the Week 4 Tutorial prompt and annotate it live:
- Where is the Role? ("You are my personal tutor…")
- Where is the Context? ("Week 4 of AI 101 at Silver Oak University…")
- Where is the Goal? ("Teach me meta-prompting and the nine components…")
- Where are the Constraints? ("Never hand me the answer directly… exactly ONE question per message…")
- Where is the Evaluation? ("Pass bar: 4 of 5 on the exit check…")

The lesson: the medium is the message. A well-written tutor prompt IS a live, working demonstration of structured prompting. You're learning structured prompting by using a structured prompt that teaches structured prompting. Notice that.

Then show the students: the C- tutor file is also an example of meta-prompting in action — the instructor used the AI (and the components framework) to draft and refine it. Not the human laboring over every sentence — the AI asked questions, the instructor answered, and the template emerged.

Segment 6 — Live Demo: Meta-Prompt → Template → Test → Critique (20 min)

Full workflow, narrated:
1. Open an assistant and meta-prompt: "I run a weekly team standup. I need a reliable prompt that helps an AI draft my standup update. Ask me clarifying questions one at a time — what do you need to write a great reusable prompt for this?"
2. Answer the AI's questions live — team size, format, what "blockers" means, tone.
3. Receive the draft template. Read it out loud. Notice what the AI added (possibly an Evaluation component you didn't think of) and what it assumed.
4. Test it with a sample input: "Yesterday: finished the slide deck. Today: stakeholder meeting at 2. Blocker: waiting on approval from Sarah." — run the template and show the output.
5. The verify-the-AI beat: ask the AI: "What did you assume that I didn't say? What could break this template?" Note at least one real weakness it surfaces (e.g., "I assumed your team uses asynchronous updates — is that right?").

Key takeaway for students: even a good meta-prompted template has gaps. The AI will often surface them if you ask directly. Testing with real input is the only way to know if it works.

Misconception + cure:
- ❌ "If the AI wrote the prompt, it must be good."
✅ Cure: the AI writes a draft. You test, critique, and improve. The meta-prompt move outsources the first draft; your judgment is still the gate.

Segment 7 — Technology Workflow + AI-Critique: Audit the Bloated Prompt (20 min)

Technology workflow — the structured-prompt habit, on demand:
1. Start with a goal. What must the output accomplish?
2. Run a context audit. What does the AI need to know that it doesn't? (Background, role, audience, data)
3. Add constraints. What are the must-nots? (Length, tone, topics to avoid)
4. Decide on format. What structure or style do you want?
5. Add examples or evaluation if the stakes are high enough to warrant them.
6. Test with real input. Does the output match what you actually needed?

AI-critique moment — the week's verify-the-AI beat:

Project this over-engineered prompt (clearly framed as the error to catch — do NOT present this as a good example):

"You are a world-renowned expert in productivity, behavioral psychology, and executive communications with 30 years of experience running Fortune 500 C-suite communications for technology firms. Your writing style is precise, data-driven, warm-but-authoritative, slightly informal, never using passive voice, always using Oxford commas, referencing specific studies where possible. Write a 150–200-word update email, but only if you can confirm the accuracy of all claims first, and also make sure to check that the tone matches the organization's brand guide (attached — do not invent a brand guide), and ask for clarification on any ambiguous points before you write, and also include a subject line, and add a P.S. if appropriate, and avoid the word 'ensure.'"*

Ask the class to diagnose it: How many components are here? Which are contradictory? Which are good but buried? What would you cut? (Common findings: the Role is absurdly over-specified; "only if you can confirm accuracy" is an impossible instruction; "ask for clarification" and "write now" conflict; the P.S. instruction is noise; removing "ensure" is a micro-constraint that adds no value.)

The lesson: a bloated prompt with internal contradictions is worse than a lean one. Strip it to core components that actually change the output.

Callback + tease:
- Callback: "Meta-prompting + the nine components = the toolkit. But knowing the tools isn't the same as knowing when to use each one. That judgment is this week's Studio."
- Tease next week: "Week 5: examples, structure, and control — how to give the AI worked examples (zero, one, or few-shot) to shape voice, format, or strip PII. Once you've seen the difference one good example makes, you'll never prompt without them."

Segment 8 — Hand-off (6 min) · Session 2 closes (~75)

Due this week:
- Lecture Tutorial 4 — AI tutor on meta-prompting + the nine components; submit the share link.
- Practice exercises — floor-difficulty reps to lock in the components before the quiz.
- AI Build Studio 4 — "Build a Reusable Structured-Prompt Template" — the hands-on build; 50 pts.
- Quiz 4 (no AI) — match components to what they control; what's-the-prompting-fix scenarios.
- Discussion 4 — "Formula or Cage?" — initial post Fri Sep 25, replies Sun Sep 27.
- Assignment 4 — "Engineer a Prompt" — coached by AI, scored against the rubric; 100 pts.

Instructor FAQ — Common Stumbles

Student says / does	Quick cure
"Writing 'you are an expert' makes the AI more accurate."	Role changes style/framing, not factual accuracy. Still verify.
Confuses Examples and Constraints.	Constraints say what not to do; Examples show the style or format you want. Two different levers.
Writes an extremely long prompt and gets worse results.	Contradictory or redundant instructions cancel each other out. Strip to what actually changes the output.
"The perfect prompt just needs to be found."	The "perfect prompt" is a moving target. Build a reliable template you can tune, not a magic string.
Thinks meta-prompting means the AI writes everything for them.	Meta-prompting outsources the first draft. Your job is to test, critique, and improve it.
Uses Evaluation component but the AI ignores it.	Add it at the END of the prompt and make it specific: "Before returning, verify: is this under 200 words? Does it avoid jargon?"

Scope flag

This outline stays within Objective 2 at the structured-prompting level. It does not go into few-shot examples, PII removal, or simulation prompts — those are Weeks 5 and 6. Real products (ChatGPT, Claude, Gemini, Copilot) are named factually as the tools students use; no specific version or model-capability claims are made that would require verification. The instructor and institution are fictional. The buyer-facing verb for the product is "generates."

~ Prof. Quinn's edition · Fall 2026 · built with thecoursemaker.com