← Back to the Using Artificial Intelligence outline The Course Maker
← Using Artificial Intelligence outline
Week 4 · AI Build Studio

Week 4 — AI Build Studio · "Build a Reusable Structured-Prompt Template"

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: Objective 2 — use meta-prompting and the nine structured-prompt components to build a reusable template; catch the template's weaknesses and improve it · SLO A (produce quality results through excellent prompting) · SLO B (evaluate and verify AI-generated work)
Worth 50 points · AI Build Studios group = 15% of the grade · Studio 4
Format: a hands-on build — you'll use meta-prompting to draft a template for a real recurring task, test it with real input, catch its weaknesses, and document what you improved.

This is the course's signature weekly component. Every instructional week has one Studio — a real thing to build, a required step where you verify and improve the AI's work, and a short reflection. All tools are free; everything is links to external sites.

Part 1 — The Build Goal

By the end of this Studio you'll have produced four things for a real recurring task from your own life or major:
1. A structured-prompt template (built with at least five of the nine components, using meta-prompting to draft it).
2. A test run — you ran the template with real input and got a real output.
3. An AI-critique write-up — you identified at least one weakness in the template or the output (something the AI got wrong, over-assumed, or left out) and documented how you caught it and fixed it.
4. A short reflection on what the template-building process taught you about prompts.

This Studio practices the course's central discipline: build deliberately, test honestly, and catch what breaks.

Open one approved assistant to build in: ChatGPT (https://chatgpt.com), Claude (https://claude.com), Gemini (https://gemini.google.com), or Copilot (https://copilot.microsoft.com). Free accounts are sufficient for every step.

Part 2 — Pick a Real Recurring Task

Choose something you actually do repeatedly — the more real the task, the more useful the template will be. Examples:

  • Summarizing readings or lecture notes before class.
  • Drafting weekly updates (for a study group, a job, a club, a team).
  • Writing emails of a specific type (following up after meetings, requesting extensions, applying for positions).
  • Generating practice quiz questions from your notes.
  • Rewriting your own paragraphs for clarity.
  • Preparing a brief for a project or presentation.

Write your task here: I need a reusable template for ______.

This task should be something you expect to do again — a one-time task doesn't need a reusable template.

Part 3 — Meta-Prompt to Draft the Template

Step 3a — Start the meta-prompting move. Open your approved assistant and type exactly this structure (adapting the task to yours):

"I need a reusable prompt template for [your recurring task]. Ask me clarifying questions one at a time — what do you need to know to write a great reusable prompt for this? When you have what you need, return the template in Markdown format, clearly labeled with each prompt component."

Answer each of the AI's questions honestly. If a question seems off, say so — that's information too.

Step 3b — Receive and annotate the draft. When the AI returns the Markdown template:
- Read it all the way through before using it.
- For each section, note which of the nine components it represents (Context, Role, Goal, Audience, Constraints, Voice/Format, Data/Logic, Examples, Evaluation).
- Note which components are present and which are absent.
- Note any instructions you didn't ask for that the AI added.

Step 3c — Ask the AI to audit its own draft. After receiving the template, send this follow-up:

"Before I use this template, what did you assume that I didn't explicitly state? What are the most likely ways this template could fail or produce a disappointing output?"

Save what the AI says. This is your pre-test weakness list.

Part 4 — Test the Template with Real Input

Take your template and run it with actual, real content — not example data, but something from your real life right now.

What to capture:
- The template you ran (paste it or the key parts).
- The real input you provided (the actual reading you're summarizing, the actual update you're drafting, the actual notes you're working with).
- The output the AI produced.
- Your assessment: does the output do what you actually needed? Is it the right length, tone, format? Does it stay within the task's scope?

If the output isn't right, that's a finding, not a failure. Write down exactly what's wrong — the template will be better for it.

Part 5 — The AI-Critique / Verification Step (required — this is the BYOAI step)

Now catch what went wrong or what was assumed. Every template built by meta-prompting has at least one weakness — an assumption the AI made, a component that's redundant, a constraint that's missing, or an instruction that's contradictory in practice. Find it.

Work through these in order:

  1. Compare the AI's pre-test weakness list (from Step 3c) against what actually happened when you ran the template. Did the weakness it predicted show up? Did something else go wrong that it didn't anticipate?

  2. Test at least one assumption. Pick something the AI assumed that you didn't explicitly state. Rerun the template with a deliberately different input that violates the assumption and see whether the template still works.

  3. Check for over-engineering. Look at the template. Is there a component that doesn't change the output when you remove it? Is there an instruction that's contradictory with another? Remove or resolve it.

  4. Check for a missing piece. Is there a situation where you'd run this recurring task and the template would produce something wrong? What component would prevent that?

Write 3–5 sentences reporting: (a) the specific weakness you found (or, if everything worked, exactly how you verified it — don't say "it worked great" without naming what you tested); (b) what the AI got wrong, assumed, or over-engineered; (c) the fix you made.

This step cannot be skipped or faked. A student who submits the template without a genuine critique and fix earns the low end of the AI-critique rubric row. The rubric rewards your judgment about the template — not the template's prose.

Part 6 — Refine and Lock the Template

Based on your critique, write the final, refined version of your template — the version you'd actually save and reuse.

It should:
- Have at least five of the nine components with real content (not just labels).
- Be free of contradictory instructions.
- Be shorter than the first draft if you found redundancies.
- Include a note to yourself (in a comment or a separate line) about the most likely failure mode to watch for.

Part 7 — Reflection (2–3 sentences)

What did the meta-prompting process teach you about what makes a prompt work? What surprised you — about what the AI assumed, what the template got wrong, or what you had to add after testing?

Part 8 — What to Submit

Submit a single document (or text entry) containing:
1. Your recurring task (one sentence).
2. The meta-prompting request you used to start the process (paste it).
3. The annotated draft template (with component labels noted).
4. The AI's pre-test weakness list (from Step 3c).
5. Your test run — real input + output (or the relevant part).
6. Your Part 5 AI-critique write-up (the weakness you found + the fix).
7. Your final, refined template (the version you'd actually reuse).
8. Your Part 7 reflection (2–3 sentences).

Due Sunday, Sep 27, 11:59 p.m. (50 points).

Instructor answer key & model deliverable — REMOVE BEFORE PUBLISHING TO STUDENTS

Students use their own tasks, so deliverables vary. Grade the process (meta-prompting + testing + genuine critique + refinement), not a specific template. The model below shows what full credit looks like.

Model deliverable (illustrative):

  • Task: "I need a reusable template for summarizing weekly readings before class."

  • Meta-prompting request used: "I need a reusable prompt template for summarizing weekly readings before class. Ask me clarifying questions one at a time — what do you need to know to write a great reusable template for this? When you have what you need, return the template in Markdown format, clearly labeled with each prompt component."

  • AI's clarifying questions (illustrative): "What subject are these readings usually in? / How long are the readings? / Who is the summary for — just you, or do you share it? / What do you want the summary to include — main argument, key terms, examples? / Is there a length you prefer for the summary? / Should I include a 'questions to bring to class' section?"

  • Draft template (annotated):
    **[ROLE]** You are a supportive study tutor helping me prepare for class. **[GOAL]** Summarize the assigned reading I'll paste below so I can walk into class ready to discuss. **[AUDIENCE]** This summary is for me only — I'll use it to prepare questions and track the argument. **[CONSTRAINTS]** Do not add information from outside the reading. Keep the summary to 150–200 words. Use plain language; avoid unnecessary jargon. **[VOICE/FORMAT]** Structure: (1) Main argument in one sentence. (2) Three key supporting ideas, one sentence each. (3) Two questions I could raise in class. (4) One term I should look up. **[EVALUATION]** Before returning: does the summary stay within 200 words? Are all three key ideas genuinely from the reading, not added? [DATA: paste the reading or key excerpt here]
    Component notes: Role ✓, Goal ✓, Audience ✓, Constraints ✓, Voice/Format ✓, Evaluation ✓, Data/Logic ✓. Missing: Context (no background about my course or class level) and Examples (no sample summary).

  • AI's pre-test weakness list: "I assumed your readings are typical academic prose — if they're very technical or quantitative, the 'plain language' constraint may create tension. I also assumed you want the summary to be ready in one run; if your readings are very long, you may need to paste in sections. The '3 key ideas' structure may not fit readings with a different organizational shape."

  • Test run: pasted a 1,200-word sociology article. Output was 198 words, correctly structured. Found: the "one term to look up" section produced a term that was already defined in the reading — the AI didn't catch that it was already explained.

  • AI-critique write-up: "The specific weakness I found was in the Evaluation component — I told it to check word count and sourcing, but I didn't ask it to check whether the 'term to look up' was already defined in the reading. The AI surfaced this itself when I ran Step 3c, but I confirmed it by testing: it listed 'habitus' as a term to look up, even though the reading defined it in the second paragraph. Fix: I added to the Evaluation component: 'Does the term in section 4 appear undefined in the reading? If it's defined in the reading, choose a different term instead.'"

  • Final, refined template: [same as above with the added Evaluation line]

  • Reflection: "The meta-prompting process forced me to articulate requirements I'd never have written cold — I hadn't thought about 'what if the term is already defined' until the AI asked about edge cases. The most surprising thing was that the template's pre-test weakness list was more useful than I expected: one of the three weaknesses it named was exactly what broke on the first test."

Why the verification step can't be faked: a student who submits a template with no test run, no pre-test weakness audit, and no named fix earns the low end of the AI-critique row — the rubric rewards judgment and iteration, not the AI's prose.

Grading rubric — 50 points

Criterion Full Partial None
Real recurring task + meta-prompting request — a genuine task and an honest meta-prompting move (includes one-at-a-time + Markdown request) (8) 8 4–6 0–3
Draft template annotated with components — at least 5 components present with real content; component labels noted (10) 10 5–8 0–4
AI pre-test weakness list (Step 3c) — the "what did you assume / what could break this?" follow-up was run and the AI's answer is documented (6) 6 3–4 0–2
Test with real input + honest output assessment — ran the template with real (not fabricated) content; assessed the actual output honestly (8) 8 4–6 0–3
AI-critique / verification step — names a specific weakness, tests an assumption, makes a real fix (not "it worked great") (13) 13 7–10 0–6
Reflection — a thoughtful, specific takeaway about the template-building process (5) 5 3 0–2

Quality gate (self-checked): the tools and links named (ChatGPT/Claude/Gemini/Copilot homepages) are real and current (verified live); no fabricated product features; no AI-generated content is presented as a ground-truth "right answer" — the Studio grades the process and the student's judgment, not the template's prose. The "model deliverable" above is illustrative; students use their own real tasks. Product-accuracy gate: PASS.

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