# Session 3 Handout: Using It on Your Job Tomorrow

<p class="meta">FAST-01 Applied AI Literacy at Work · 7formations · Instructor: Wadood Chaudhary · All examples fictional</p>

## Why this session matters

Knowing how AI works and what not to paste is necessary but not enough. This session turns it into something you'll use on Monday: three prompts tested on practice versions of your real tasks, a peer-reviewed check, a teach-back that proves you understand the limits, and a workstation checklist.

## Key concepts

### 1. A prompt is an assignment

Think of the AI as a capable temp worker on day one. It knows a lot in general and nothing about your company, your customer, or what "good" looks like for this task. Whatever you leave out, it will guess. A working prompt covers six things:

| Part | Question it answers | Example (Cape Harbor Inn) |
| --- | --- | --- |
| Role | Who is writing? | "I'm a front-desk associate at a small inn." |
| Reader | Who will read it? | "...replying to a guest who is booking." |
| Task | What one job? | "Explain our pet policy." |
| Source | What safe text may it use? | FAQ excerpt pasted between `---` lines |
| Format | How long, what shape, what tone? | "3 sentences, warm and clear." |
| Gap rule | What if the source doesn't say? | "If it isn't in the text, say so and offer to check." |

### 2. Fix the prompt, not just the output

When a peer finds an error, correcting the sentence fixes this one draft. Changing the prompt (adding the source, the gap rule, or the format) fixes every future draft and makes the prompt reusable by coworkers. Always rerun after a prompt change and compare.

### 3. Peer review with the four-step check

Reviewing someone else's output is easier than reviewing your own: you don't know what it was "supposed" to say, so you read what it actually says. In pairs, swap one prompt and output, run source, facts, numbers, and tone, and record a decision with a one-line reason. Feedback should be about the prompt and specific: "Add 'use only the FAQ'" beats "it was wrong."

### 4. Teach-back

If you can explain an AI limit in under a minute, in plain words, with an example, you understand it. Good teach-backs have three parts: the limit ("it writes what sounds likely"), the example ("it said 9:15; the FAQ said 8:15"), and the action ("so check the FAQ before answering").

### 5. Ownership

The person who sends it owns it. Your name is on the email, the notice, the summary. "The AI wrote it" is never an excuse, and it is a fast way to lose a customer's trust.

### 6. Starting small and measuring

Start with one low-risk task, check everything, and note two numbers: minutes saved after checking, and errors caught. If a task shows no gain once checking time is included, stop using AI for it. That is a good result, not a failure.

## Worked example: from thin prompt to tested prompt

**Thin:** "Write a reply to the guest about pets."

**Output problem:** invents "cats welcome."

**Tested prompt:**

```text
I'm a front-desk associate at Cape Harbor Inn. A guest asks whether
they can bring two dogs and a cat. Using ONLY the FAQ text between the
lines, write a 3-sentence warm reply. If the FAQ doesn't answer part of
the question, say so and offer to check with the manager.
---
Pets: two dogs up to 40 lb in garden rooms only; $35 per night. No cats.
---
```

**Check:** Facts: dogs OK in garden rooms, no cats (match). Numbers: 40 lb, $35 (match). Tone: warm, no extra promises. **Keep.**

## Worked example: a teach-back script (45 seconds)

"AI tools write what sounds likely, not what they've checked. In our class, the assistant told us the ferry shuttle leaves at 9:15. The FAQ says 8:15. It sounded completely sure. So before you tell a guest anything, find it in the FAQ or ask the manager. And never paste a guest's card number into it; that's protected under Massachusetts law."

## Checklist: is my prompt ready to share?

- [ ] Role and reader are named.
- [ ] One task, stated plainly.
- [ ] Safe source pasted between markers.
- [ ] Format, length, and tone set.
- [ ] Gap rule included ("say 'not in source'").
- [ ] A partner ran it and got a usable result without asking me questions.
- [ ] Contains no names, numbers, or confidential details.

## Do and don't

| Do | Don't |
| --- | --- |
| Start with one low-risk task and check everything | Automate a whole process in week one |
| Fix and rerun the prompt after an error | Patch only the output and reuse the same prompt |
| Measure time saved after checking | Count only the seconds the AI took |
| Keep the reviewer for customer, money, and safety items | Skip review because "it was right last time" |
| Share prompts that work for others | Keep prompts that only work with your private knowledge |

## Try this at work

1. Pick one prompt from W4. Use it three times next week on safe inputs; record minutes and errors each time.
2. Post your workstation checklist where you can see it.
3. Do a 60-second teach-back for one coworker. Note the question they ask; it tells you what to explain better.
4. After two weeks, review your AI Use Map: move one task between piles based on evidence.

## Glossary

| Term | Plain-language meaning |
| --- | --- |
| Gap rule | An instruction to say "not in source" instead of guessing |
| Practice version | A copy of a real task using fictional or public data |
| Peer review | A coworker runs the four-step check on your output |
| Teach-back | Explaining a concept to someone else to confirm you understand it |
| Final applied task | The scored task: prompt, four-step check, and decision |
| Workstation checklist | Your one-page list of daily data and checking habits (W5) |

## Key takeaways

- A prompt is an assignment: role, reader, task, source, format, gap rule.
- Fix the prompt, rerun, compare.
- You own what you send.

*Tool names are trademarks of their owners. Features as of October 2026; check your tool's current help pages.*