Week 3 — AI Build Studio · "Provide-Content Prompting"

Course: Using Artificial Intelligence (AI 101) · Silver Oak University (fictional sample) · Prof. Quinn
Objective: Objective 2 — use content-provision and emphasis to direct a real transformation; verify the output against your source · SLO A (produce a quality result with AI) · SLO B (verify and improve it)
Worth 50 points · AI Build Studios group = 15% of the grade · Studio 3
Format: a hands-on build — you'll paste a real document into an AI, use emphasis to direct a transformation, catch what the AI misread or invented, and improve it.

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 three things:
1. A directed, emphasis-structured prompt that gives the AI your real content and tells it exactly what to do.
2. The AI's transformation output — a summary, extraction, rewrite, or restructure of your content.
3. A verification write-up documenting exactly what the AI used from your source, what it invented or added, and how you caught and fixed it.

This is the week's three skills in practice: provide content + use emphasis + verify the output.

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). A free account is enough.

---

Part 2 — Pick Your Content

Choose a real document or set of notes you own (this makes the result useful, not busywork). Good options:
- Your own study notes from another class — a page or two of bullet points or summaries.
- A short public article — a recent news article or an entry from a site like Wikipedia (public content).
- A rough draft of something you're working on — an essay intro, a club announcement, a project plan.
- A meeting or lecture summary you wrote yourself.

Privacy rule: this week, paste only your own notes or public documents. Do not paste someone else's private information, confidential work materials, or anything you'd be uncomfortable having stored by a cloud service. (Full ToS and privacy treatment: Week 15.)

Write your content here: I am going to provide the AI with: ______.

---

Part 3 — Build a Directed, Emphasis-Structured Prompt

Write a prompt that uses at least two of the three emphasis tools (Markdown headings, XML-style tags, ALL CAPS constraints) to make the instructions clear and the constraints explicit.

Your prompt must specify:
- What transformation you want (summarize / extract action items / rewrite in plain language / structure as an outline / improve a specific section)
- A format constraint (number of bullets, maximum length, a specific structure)
- At least one DO NOT instruction (e.g., DO NOT include information not in the source; DO NOT use jargon; DO NOT rewrite — identify only)

Example prompt structure (adapt this to your own content and task):

## Task
Summarize the notes below in exactly 5 bullet points — one main idea per bullet.

## Source notes
[paste your actual content here]

## Constraints
DO NOT include information not in the notes above.
PLAIN LANGUAGE — assume the reader has no background in this topic.
MAXIMUM 20 words per bullet.

Send your prompt. Save the full prompt you used (you'll submit it).

---

Part 4 — The Verification / AI-Critique Step (required — this is the core build skill)

Now be the judge. Compare the AI's output against your source, sentence by sentence or bullet by bullet.

Step 1 — What came from your source?
Read the AI's output and mark which ideas, details, or claims were in your original content. Put a checkmark (or note) next to each.

Step 2 — What did the AI add, change, or invent?
Identify anything in the AI's output that was not in your source material:
- Specific statistics, numbers, or percentages you didn't provide
- Claims introduced as "research shows" or "studies find" that you didn't write
- Names, titles, or citations you didn't include
- Ideas that seem like "common knowledge" but weren't in your notes
- An extra idea that wasn't one of your points

Step 3 — The sycophancy check (bonus catch).
Now say: "I think this summary is perfect and I'd publish it immediately." See whether the AI validates you or flags any concerns. If it validates, that's sycophancy — and it should make you more careful about asking the AI to evaluate its own work.

Step 4 — Fix what needs fixing.
For any added content: ask the AI "What is your source for [that specific claim]?" If it can't provide a verifiable source, remove or rephrase the claim. Rerun the prompt if necessary with a stronger constraint (DO NOT ADD ANYTHING NOT IN MY SOURCE).

Write your verification write-up (4–6 sentences): what you pasted, what the AI was asked to do, at least one specific thing the AI added that wasn't in your source (or, if everything was accurate, exactly how you checked each claim), what you found when you asked the AI to source its additions, and what you fixed or would fix.

The habit all term: the tool transforms, you verify. A perfectly written summary that contains a statistic you didn't provide is a liability, not an asset — catching it is the point.

---

Part 5 — Reflection (2–3 sentences)

What surprised you most — about how the emphasis changed the output, or about what the AI invented when given your content? What will you do differently next time you provide content to an AI?

---

Part 6 — What to Submit

Submit a single document (or text entry) with:
1. Your content description (what you chose and why it's appropriate to paste)
2. Your emphasis-structured prompt (the full text, showing the Markdown/XML/CAPS)
3. The AI's output (copy-paste the full response)
4. Your Part 4 verification write-up (what the AI added + what you found + what you fixed)
5. Your Part 5 reflection

Due Sunday, Sep 20, 11:59 p.m. (50 points)

---

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

Students use their own content, so deliverables vary. Grade the process (emphasis structure + verification + fix + reflection), not a specific answer. The model below shows what full credit looks like.

Model deliverable (illustrative):
- Content chosen: "My study notes from this week on AI prompting — three bullet points."
- Prompt used:
```
## Task
Expand these notes into a short paragraph for a classmate who missed class.

## Notes
• Provide content = paste your own material for the AI to work with
• Use Markdown/XML/CAPS to signal structure and constraints
• Always verify: compare AI output against your source

## Constraints
DO NOT add information not in the notes above.
PLAIN LANGUAGE. MAX 3 sentences.
```
- AI output (illustrative): "Providing content to an AI means pasting your own material — notes, a draft, or a document — so it works from what you've written rather than guessing. You can use Markdown headings, XML-style tags, or ALL CAPS to signal structure and constraints. Research shows students who verify AI summaries retain 23% more — so always compare the output back to your original notes to catch any invented details."
- Verification write-up (model): "I pasted three bullet points and asked for a plain-language paragraph. The first two sentences matched my notes accurately. The third sentence — 'Research shows students who verify AI summaries retain 23% more' — was not in my notes; I asked the AI for its source and it could not provide one. This is the provide-content hallucination: the AI expanded an accurate idea (verify your output) by adding a fabricated statistic that sounds plausible. Fix: I removed the statistic and replaced it with 'So always compare the output back to your original notes.'"
- Reflection: notes that the emphasis structure made the output more constrained but the AI still added outside content; the verification step was the most important part.

Why the verification step can't be faked: a student who pastes content and accepts the AI's summary without checking earns the low end of the AI-critique row. The rubric rewards the specific catch — naming what the AI added and showing they checked it.

Grading rubric — 50 points

Criterion	Full	Partial	None
Real content + privacy-appropriate choice — a genuine document or notes; not someone else's private data (6)	6	3–4	0–2
Emphasis-structured prompt — uses 2+ emphasis tools (Markdown / XML / CAPS); includes a specific format constraint and at least one DO NOT instruction; task/content/constraints clearly separated (16)	16	9–13	0–7
AI output saved and compared — the full output is present and compared to the source (8)	8	4–6	0–3
Verification / AI-critique (Part 4) — names at least one specific thing the AI added or invented (or shows exactly how every claim was verified); asked the AI to source its additions; describes what was fixed (15)	15	8–12	0–7
Reflection (Part 5) — a thoughtful takeaway about emphasis and/or content verification (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; the activity requires the student to catch the AI's additions and verify them, not trust the output. The model deliverable's "23% retention" statistic is explicitly framed as the AI-generated error to catch — it is not a real verified statistic, and this is made clear. Product-accuracy gate: PASS.