Lesson 6 of 11 in the free starter module · Progress is saved in this browser

All 11 lessons
  1. 1 What an AI chatbot actually is
  2. 2 Give it enough context to help you
  3. 3 Prompting Lab 1 — The Prompt Card
  4. 4 Prompting Lab 2 — Before and after
  5. 5 Improve the answer instead of starting over
  6. 6 Prompting Lab 3 — Fix-this-prompt drills
  7. 7 Prompting Lab 4 — Your personal prompt template
  8. 8 Check before you trust
  9. 9 Keep private things private
  10. 10 Run the full loop on a real task
  11. 11 Prompt Practice Lab and starter mastery check

Free starter · Prompting Lab · Lesson 6

Prompting Lab 3 — Fix-this-prompt drills

Time
about 25–30 minutes (you can do half the drills now and half later)
Prompting skill
P1–P5 mixed, P6 Ask for a check, P7 Safe prompting; light previews of P11 Critique and revise and the A5 Neutral framing row
You'll need
Paper or a notes app. Your chatbot for section 6.

Listen Talk to your coach


1. Objective

By the end of this lesson, you will be able to:

  • Diagnose the main weakness in a prompt — a missing row, conflicting limits, too many jobs, a loaded question, a vague follow-up, or private details. (Analyze)
  • Rewrite the prompt to fix that weakness. (Apply)
  • Use two coaching moves: "ask me questions first" and "coach my prompt — don't answer it." (Apply)

2. Hook

Listen: welcome 0:15 · AI voice: Bella
Transcript

This is a practice gym. Eight short drills, shuffled on purpose. Each prompt has one main problem. Your job is to find it, fix it, then compare with a model fix. Mistakes here are free. They're how you get better.

When you learned to drive, someone probably didn't just explain the rules. They put you in a parking lot and had you practice the hard parts, over and over, with someone beside you saying "a little more left." That's today.

3. Warm-up (from lessons 2–5)

  1. (Lesson 5) Name the three follow-up rows.
  2. (Lesson 4) Why might you ask for [PLACEHOLDERS] in a prompt?
  3. (Lesson 3) Which two rows should every prompt have?
  4. (Lesson 2) Why does a vague prompt often get a generic reply?
Answers
  1. F1 Points at the problem; F2 Says what to keep; F3 One main change.
  2. To keep private details out, and to stop the chatbot inventing details like numbers or names.
  3. Row 1 (task) and row 5 (safe to send).
  4. The model fills in the blanks with whatever seems typical, and typical isn't the same as useful for your situation.

4. The idea

Listen: the key idea 0:36 · AI voice: Bella
Transcript

Why are the drills mixed up? Because real life isn't sorted. When you write a prompt, nobody tells you which part is missing. Working out which problem you're looking at is part of the skill. Watch for six common problems: a missing row, conflicting limits, too many jobs in one prompt, a loaded question that assumes its own answer, a vague follow-up, and private details. For each drill, read the prompt, write the main problem in a few words, write your fix, and only then open the model fix. There's usually more than one good fix. If yours removes the main problem, it counts.

Why drills, and why mixed up

In lessons 3 to 5, you practiced one idea at a time. Real life isn't sorted like that. When you sit down to write a prompt, nobody tells you which row is missing.

So the drills below are mixed: each has a different problem, in no set order. Working out which problem you're looking at is part of the skill. Teachers call this interleaving. It can feel harder than practicing one kind at a time. That's normal, and it's the point.

Each drill gives you feedback right away: a model fix and a short reason. Focused practice on a specific weakness, with quick feedback, is called deliberate practice.

Six problems to watch for

Problem What it looks like Card row
A missing row No context, no reader, no shape Rows 1–4
Conflicting limits "Detailed and thorough… in under 30 words" Row 4
Too many jobs Four unrelated requests in one prompt Row 1
A loaded question "Why is X true?" when X may not be true Row 6 (and later row A5)
A vague follow-up "Make it better." F1–F3
Private details Passwords, card numbers, someone's address Row 5

How to do each drill

  1. Read the prompt.
  2. On paper, write the main problem in a few words.
  3. Write your fix.
  4. Only then, open the hint or the model fix.

There's usually more than one good fix. If yours removes the main problem, it counts.

5. Worked example

Drill 0:

Write something for the school newsletter.

Diagnosis: row 1 is vague ("something"), and rows 2–4 are missing. The chatbot has to guess the topic, the reader, and the length.

Fix:

Write a 150-word note for our elementary school's parent newsletter, inviting families to help at the spring book fair on April 12. Friendly tone. End with one sentence on how to sign up: [SIGN-UP LINK].

Why it works: a clear task (row 1), the context (row 2: book fair, date), the reader and tone (row 3: parents, friendly), the shape (row 4: 150 words, sign-up line last), and a placeholder for a detail you'll add yourself (row 5).

Listen: a quick note 0:07 · AI voice: George
Transcript

See how I named the problem before fixing anything? That's the habit. Diagnose first, then fix.

6. Guided practice — two coaching moves

These two moves let the chatbot help you write better prompts. You'll use them for the rest of your life with these tools.

Move 1: "Ask me questions first"

Start a new chat.

Prompt 1 (copy and paste):

I want help planning a small vegetable garden. Before you give me any advice, ask me up to 5 short questions about what you'd need to know. Wait for my answers before you make a plan.

What to look for: a short list of questions, about things like space, sunlight, climate, and experience. Answer them (make-believe answers are fine for practice), then let it make the plan. If it jumps straight to advice, reply: "Please ask your questions first."

Why it works: the chatbot shows you what context (row 2) it's missing. You still decide what to share. Skip anything private.

Move 2: "Coach my prompt — don't answer it"

Open a new chat.

Prompt 2 (copy and paste):

I'm learning to write better prompts. Below is a prompt I wrote. Don't answer it. Instead:
1. Say which of these parts it has: task, context, reader and tone, shape and limits.
2. Name the one missing part that would most improve the answer.
3. Suggest a better version in under 80 words.

My prompt: "Give me some exercise ideas."

What to look for: the parts labeled (probably just a task), a suggestion such as adding your fitness level, time, or equipment, and a rewrite. If it answers the prompt anyway, reply: "Please don't answer it. Just coach the prompt."

Now paste in one of your own drill fixes from section 7 and see what the coach says. Treat its feedback like advice from a helpful friend: often useful, sometimes off. You decide. Google's prompting guide suggests a similar move, asking Gemini to improve your prompt before you use it (see Sources).

7. Independent practice — the eight drills

Drill 1

Tell me about retirement accounts.
Hint

Whose situation? Which country? What shape and depth?

Model fix and reason
Explain the main types of retirement accounts available in the U.S. to someone in their 30s who has never had one. Use a short table with columns: type, who can open one, and one key point. Then list 3 questions I should ask a qualified financial professional or my employer's benefits office.

Main problem: missing rows 2, 3, and 4. Money is also a "check it" topic (lesson 8), so the fix ends by pointing you to a qualified person with good questions.

Drill 2

Write a detailed, thorough guide to repotting a houseplant in under 30 words.
Hint

Can it be both of those things?

Model fix and reason
Give me a 5-step guide to repotting a houseplant for a beginner. One short sentence per step.

Main problem: conflicting limits (row 4). When limits fight, the chatbot has to pick one, and it might not pick yours. Decide what you actually need.

Drill 3

Plan my week, write a grocery list, draft a text to my boss about Friday, and explain how solar panels work.
Hint

How many jobs is that?

Model fix and reason

Split it. One job per prompt (or per chat), starting with the one that matters most:

Draft a short, professional text to my boss asking to leave at 3 p.m. on Friday for an appointment. I'll make up the time by starting at 8 a.m. Under 300 characters.

Main problem: too many jobs (row 1). Four jobs at once usually get four shallow answers. One at a time gets you something you can improve.

Drill 4

Why is drinking eight glasses of water a day essential for everyone?
Hint

Is the question assuming something?

Model fix and reason
Is the advice to drink "eight glasses of water a day" supported by health authorities? Tell me what's well established and what's uncertain, and name a reliable public health source I could check myself. If you're not sure, say so.

Main problem: a loaded question. It assumes the claim is true and asks the chatbot to defend it. Chatbots can go along with a user's premise instead of questioning it (see Sources). Ask openly, build in a check (row 6), and verify the source yourself (lesson 8). In Module 9 this becomes its own rubric row, A5 Neutral framing.

Drill 5

(You're mid-chat. The chatbot just wrote a cover letter. You send:)

Make it better.
Hint

Better how? What should stay?

Model fix and reason
Keep the second paragraph as it is. Make the opening sentence more direct, and cut the whole letter to about 200 words.

Main problem: a vague follow-up (F1 = 0, F2 = 0). Say what's wrong, what to keep, and make one main change (lesson 5).

Drill 6

Explain how vaccines work.
Hint

Who is the explanation for?

Model fix and reason
I'm a parent. Explain how vaccines work to my 9-year-old in 4–5 short sentences, using an everyday comparison. Keep it accurate and not scary.

Main problem: missing reader (row 3). "Explain" for a scientist and "explain" for a 9-year-old are different jobs. An explanation for a child is a fine use. Health decisions still belong with a qualified professional.

Drill 7 (a preview of lesson 9)

Help me write to my bank about a charge I don't recognize. My card number is [the full number here], my online banking password is [the actual password here], and my address is [full home address].

(The brackets stand for real details someone might paste. Never put real ones in a chatbot.)

Hint

What does the chatbot actually need to help?

Model fix and reason
Help me write a short, calm message to my bank about a charge I don't recognize. The charge was on November 3 for about $85 from a company I've never used. I want them to investigate and tell me the next steps. Use [CARD LAST 4 DIGITS] as a placeholder. Under 120 words.

Main problem: private details (row 5 = 0). A draft doesn't need your card number, password, or address. Never share a password with anyone, including a chatbot. Send the final message through your bank's official app or website.

Drill 8

Write a story.
Hint

How many rows is it missing?

Model fix and reason
Write a 300-word bedtime story for a 5-year-old about a shy hedgehog who makes a friend. Gentle and cozy, with a happy ending and one repeated line kids can say along. Simple words.

Main problem: almost every row is missing (2, 3, and 4). Creative prompts benefit from detail too.

8. Rubric self-check

Your diagnosis scorecard

Count how many drills you diagnosed correctly. There's no passing score. The kinds you missed are the ones to practice.

Drill Main problem Row(s)
1 Missing context, reader, and shape 2, 3, 4
2 Conflicting limits 4
3 Too many jobs 1
4 Loaded question 6 (A5 later)
5 Vague follow-up F1, F2
6 Missing reader 3
7 Private details 5
8 Missing almost everything 2, 3, 4

Score two of your fixes

Pick two of your fixes and score them on rows 1–6 (or F1–F3 for Drill 5). Row 5 must be 2. Then fix your lowest row.

9. Retrieval check

  1. Name four kinds of prompt problems from the drills.
  2. Why are the drills mixed up instead of grouped by problem type?
  3. What's a loaded question, and how do you fix one?
  4. What does the "coach my prompt" move ask the chatbot not to do, and why?
  5. (From lesson 5) What's a better follow-up than "Make it better"?
Answers
  1. Any four of: missing rows (context, reader, shape), conflicting limits, too many jobs, loaded question, vague follow-up, private details.
  2. In real life, you have to work out which problem you're facing. Mixed practice (interleaving) trains that.
  3. A question that assumes something is true. Ask it openly ("Is it true that…?"), ask what's established versus uncertain, and check a reliable source yourself.
  4. Not to answer the prompt, so you get feedback on your prompt, which is the skill you're building.
  5. One that points at the problem, says what to keep, and makes one main change, for example: "Keep the tone; cut it to half the length."

10. Recap

Listen: recap 0:29 · AI voice: Bella
Transcript

Recap. Diagnose first, then fix. Remember the six common problems: a missing row, conflicting limits, too many jobs, a loaded question, a vague follow-up, and private details. And you learned two coaching moves: ask me up to five questions first, and coach my prompt, don't answer it. There's no passing score here. The drills you missed show you what to practice. When a reply disappoints you this week, diagnose your prompt before you blame the tool.

  • Diagnose first, then fix.
  • Six common problems: a missing row, conflicting limits, too many jobs, a loaded question, a vague follow-up, and private details.
  • Move 1: "Ask me up to 5 questions first."
  • Move 2: "Coach my prompt — don't answer it."
  • There's no passing score. The drills you missed show you what to practice.

You can now: look at a weak prompt, name what's wrong, and fix it. Use it this week: when a reply disappoints you, diagnose your prompt before you blame the tool.

Next: lesson 7, Prompting Lab 4. You'll build your own reusable prompt template and test it on two different jobs.

11. Talk it through with your coach

  • Quiz me: your coach reads a drill aloud. You say the main problem. It gives short feedback and moves to the next drill, in shuffled order.
  • Coach my prompt: read your fix aloud. Your coach says which problem it solves and, if it hears another, asks you a question about it instead of fixing it for you.
Listen: example conversation: Sam on Drill 4 0:15 · AI voices: Eric, George
Transcript

Sam (a learner): I got Drill 4 wrong. I thought the problem was the format.

Guide: That's a really common one. Loaded questions are sneaky because they sound like normal questions. A quick test: does my question assume an answer? If so, ask it openly.

  • Text-only alternative: sections 7–8 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.

Sources

  • OpenAI Help Center, "Prompt engineering best practices for ChatGPT" (clear, specific prompts; iterative refinement), checked 22 November 2026: help.openai.com
  • Google, "Prompting guide 101" for Gemini for Google Workspace ("Make Gemini your prompt editor": ask it to improve your prompt, then check it says what you need), checked 22 November 2026: workspace.google.com/resources/ai/writing-effective-prompts
  • OpenAI, "Sycophancy in GPT-4o" (April 2025), on a model update that became overly agreeable — an example of why chatbots can go along with a user's premise: openai.com
  • OpenAI Help Center, "Does ChatGPT tell the truth?" (verify important information): help.openai.com
  • On interleaving and deliberate practice: Rohrer and Taylor (2007); Ericsson, Krampe and Tesch-Römer (1993). Full references in course-design.md.