1. Objective
By the end of this lesson, you will be able to:
- Explain why an "after" prompt is likely to get a more useful reply than its "before." (Understand)
- Rewrite a weak prompt into a stronger one using the Prompt Card. (Apply)
- Compare two replies and say which fits the job better, and why. (Evaluate)
2. Hook
Transcript
Today you'll see five pairs of prompts: a "before" and an "after." The first two are fully worked for you. Then, bit by bit, you take over. By pair five, it's all you.
Cooking shows love a before-and-after: the messy counter, then the finished dish. The interesting part is what happened in between. That's what you'll study today, for prompts.
3. Warm-up (from lessons 1–3)
- (Lesson 3) Say the Prompt Card memory line.
- (Lesson 2) What's a constraint?
- (Lesson 1) What's a token?
Answers
- Say what you want, about what, for whom, in what shape — and, when it helps, as whom and like what. Then: safe to send? How will I check?
- A limit you set in a prompt: a length, a format, something to include, or something to avoid.
- A word or a piece of a word. The model writes one token at a time.
4. The idea
Transcript
Here's how to read each pair. Ask three questions. First, what would the chatbot have to guess from the before? Second, which card rows did the after add? Third, is anything in the after unnecessary? Good prompts aren't padded. The help fades as you go. Teachers call this worked examples with fading. You study finished examples first, then you do more of the work yourself each time. And watch for one useful trick: asking for placeholders, like a bracketed word for a number, so the chatbot doesn't invent facts you haven't given it.
Studying a finished example first, then doing more of the work yourself each time, is a well-tested way to learn a new skill. Teachers call this worked examples with fading: the help fades as your skill grows. So the pairs below go in order:
| Pair | How much is done for you |
|---|---|
| 1–2 | Fully worked, with every change explained |
| 3 | Partly worked: you fill in two blanks |
| 4 | You write the "after," with hints |
| 5 | You write the "after" alone |
For each pair, ask three questions:
- What would the chatbot have to guess from the "before"?
- Which card rows did the "after" add?
- Is anything in the "after" unnecessary? Good prompts aren't padded.
5. Worked example
Pair 1: A note to a landlord (fully worked)
Before:
Write an email to my landlord about the sink.
What the chatbot has to guess: what's wrong, for how long, what you want done, how formal to be, how long the email should be.
After:
Write a short, polite email to my landlord. ← ROW 1 + ROW 3 (tone)
The kitchen sink has been draining very slowly for a week. ← ROW 2 (facts)
I've tried a plunger and it didn't help.
I'd like someone to come look at it, and I'm home most weekday ← ROW 2 (what I want)
mornings.
Keep it under 120 words, with a clear subject line. ← ROW 4
Don't sound angry. We have a good relationship. ← ROW 3 / ROW 4
Rows added: 2, 3, 4. Row 5: your address, unit number, and the landlord's name aren't there. Add them yourself after the draft is done. Score 2. No role or example needed.
Transcript
Notice the after isn't fancy. It's just what you'd tell a friend who offered to write the email for you.
Pair 2: Understanding a word (fully worked)
Before:
What is inflation?
A perfectly fine question. But the chatbot has to guess how much you already know and how deep to go.
After:
Explain inflation to me like I'm a smart adult who never took an ← ROW 1 + ROW 3
economics class.
Use one everyday example from grocery shopping. ← EXAMPLE (the kind I want)
Keep it to 3 short paragraphs, then give me 2 questions I could ← ROW 4
ask myself to check I understood.
No charts or math formulas. Tell me if any part is simplified. ← ROW 4 + ROW 6
Rows added: 3, 4, 6, and a request for a specific kind of example. The "2 questions to check I understood" line turns an answer into a mini-lesson. Handy for any topic you want to learn.
A note on checking: an explanation like this is usually low-stakes. If you were about to make a money decision from it, you'd check a reliable source first (lesson 8).
6. Guided practice
Pair 3: A weekly meal plan (partly worked)
Before: Give me a meal plan.
After — fill in the two blanks on paper:
Make a 5-day weeknight dinner plan for two adults.
Context: we both work, so each dinner should take ____________________ .
Format: a table with columns for Day, Meal, and Main ingredients, then ____________________ .
Limits: no fish, and reuse ingredients across days so less goes to waste.
One possible answer
"…should take 30 minutes or less to cook" and "…then a combined shopping list." Yours may differ and still be good. Each blank removes a guess.
Pair 4: Choosing a gift (you write the "after," with hints)
Before: What should I get my dad?
Hints: Row 1: how many ideas? Row 2: the occasion, his interests, your budget. Row 4: one line per idea, with a rough price range? Anything to avoid?
One possible "after"
Suggest 6 gift ideas for my dad's 65th birthday.
He likes gardening, history podcasts, and cooking on the grill. My budget is about $50.
Format: a list, one line per idea, with a rough price range.
Limits: nothing that needs a subscription, and no gadgets that are hard to set up.
Did yours include the occasion, his interests, and a budget? Those three remove most of the guessing. Prices in the reply are rough guesses by the chatbot. Check them before you buy.
Run and compare
Pick one pair (or your own Pair 4 "after").
- Open a new chat and send the before.
- Open another new chat and send the after.
- Fill in this comparison sheet:
| Question | Before reply | After reply |
|---|---|---|
| Could I use this with few or no changes? | ||
| Did it guess anything I'd have to correct? | ||
| Was it the right length and shape? | ||
| Did it get the tone right? |
What to look for: usually, the "after" reply needs fewer corrections. Sometimes the "before" reply is fine. For simple jobs, a short prompt can be enough. That's worth noticing too.
7. Independent practice
Pair 5: Your turn, no hints
Before: Help me with my resume.
Write an "after" on your own. Don't paste your real resume for this exercise. It usually contains your address, phone number, and work history. Write the prompt as if you'd describe your experience instead.
One strong version (look after you've tried)
Act as an experienced hiring manager in retail.
I'm updating my resume to apply for assistant store manager jobs. I've been a shift supervisor at a grocery store for 4 years.
Suggest 5 bullet points I could adapt to describe my work, focusing on scheduling, training new staff, and handling customer problems.
Format: bullets that start with an action verb, each under 20 words.
Limits: don't invent numbers or achievements. Use [NUMBER] as a placeholder wherever a figure would go, so I can fill in real ones.
Notice the last line. Chatbots may fill gaps with made-up details, including numbers. Asking for placeholders keeps the facts in your hands (rows 5 and 6).
Bonus: let the chatbot show you a before-and-after
Prompt (copy and paste):
Here is a vague prompt: "Help me plan a trip."
Show me a stronger version for a weekend trip by car with two kids, ages 6 and 9. Label each line with which part it is: task, context, reader and tone, shape and limits, role, or example. Then explain in two sentences why the stronger version should get a more useful answer.
Compare its version with yours. You're the judge.
8. Rubric self-check
Score your Pair 5 "after" on rows 1–6 (0, 1, 2, or n/a).
| Row | Your score | If it's low… |
|---|---|---|
| 1 Task | Start with a verb and say what you want back. | |
| 2 Context | Add what a stranger would need to know. | |
| 3 Reader and tone | Who reads it? How should it sound? | |
| 4 Shape and limits | Give a number: items, words, or sentences. | |
| 5 Safe to send | Swap private details for [PLACEHOLDERS]. Must be 2. |
|
| 6 Check built in | Ask it not to invent facts, or to say when unsure. |
9. Retrieval check
- What are the three questions to ask about a before-and-after pair?
- In Pair 2, what did "give me 2 questions I could ask myself to check I understood" add?
- Why did the Pair 5 prompt ask for
[NUMBER]placeholders? - True or false: the "after" prompt always gets a better answer.
- (From lesson 3) Which two parts of the Prompt Card are optional extras?
Answers
- What would the chatbot have to guess? Which rows did the "after" add? Is anything unnecessary?
- It turned an explanation into a mini-lesson with a way to check yourself.
- So the chatbot wouldn't invent numbers or achievements. You fill in real figures yourself.
- False. It usually helps when details matter. For simple questions, a short prompt can be enough.
- Role and example.
10. Recap
Transcript
Recap. A weak prompt makes the chatbot guess. A stronger one removes the guesses that matter. Read each pair with three questions: what's guessed, what was added, and what's unnecessary. Placeholders keep facts you haven't given in your hands. And you judge the result. This week, before you send a quick prompt, ask yourself: what will it have to guess?
- A weak prompt makes the chatbot guess. A stronger one removes the guesses that matter.
- Read pairs with three questions: what's guessed, what was added, what's unnecessary.
- Placeholders keep facts you haven't given (numbers, names) in your hands.
- "Give me questions to check I understood" turns any explanation into a mini-lesson.
- You judge the result. The comparison sheet helps.
You can now: turn a weak "before" into a strong "after" on your own. Use it this week: before you send a quick prompt, ask yourself, "What will it have to guess?"
Next: lesson 5, improving a reply you already have with specific follow-ups, instead of starting over.
11. Talk it through with your coach
- Role-play: a short two-voice scene. A character reads a "before" prompt aloud, then the narrator reads the "after," and your coach asks which rows changed.
Transcript
Sam (a learner), reading the before: Write an email to my landlord about the sink.
Narrator, reading the after: Write a short, polite email to my landlord. The kitchen sink has been draining very slowly for a week. I've tried a plunger and it didn't help. I'd like someone to come look at it, and I'm home most weekday mornings. Keep it under 120 words, with a clear subject line. Don't sound angry. We have a good relationship.
Guide: Which rows did the second version add?
- Coach my prompt: tell your coach which pair you ran and which reply you'd use. It asks what the "after" reply still got wrong, and helps you turn that into one more line for your prompt.
- Text-only alternative: sections 6–8 above.
Talk to your coach
River is an AI voice coach, not a person. River knows this lesson and can quiz you, listen to a prompt you're working on, and give hints. River won't do the exercise for you. You can talk out loud or type.
Ideas to start:
Sources
- OpenAI Help Center, "Prompt engineering best practices for ChatGPT," checked 22 November 2026: help.openai.com
- Google, "Prompt design strategies" (examples, constraints, response format), checked 22 November 2026: ai.google.dev
- OpenAI, "Why language models hallucinate" (models may guess when unsure), on why placeholders help: openai.com
- On worked examples and fading as a teaching method: Renkl and Atkinson (2003). Full reference in
course-design.md.