Lesson 10 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 · Lesson 10

Run the full loop on a real task

Time
about 35–40 minutes
Prompting skill
P1–P7 together (all six rubric rows plus follow-up rows)
You'll need
Your free AI chatbot, a new chat, and a browser tab for one official source.

Listen Talk to your coach


1. Objective

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

  • Apply all four parts of the loop (ask clearly, improve, check, protect) to one everyday task.
  • Self-grade your own prompts with the full Prompt Check and fix your lowest row.
  • Choose which skill a disappointing reply calls for.

2. Hook

Listen: welcome 0:09 · AI voice: Bella
Transcript

This is the capstone of the free starter. You already have every piece. Today you'll use them together, on purpose, on one real-life job.

So far you've practiced each skill on its own, like learning to steer, brake, and check your mirrors in an empty parking lot. Today is your first drive around the block: one real-feeling job, every skill, in order. Most people find the pieces click into place here.

3. Warm-up (from lessons 1–9)

A quick mixed round. Answer from memory.

  1. (Lesson 2) Name the four questions to glance over before sending.
  2. (Lesson 5) What should a strong follow-up always say besides what's wrong?
  3. (Lesson 8) Where does real checking happen: inside the chat, or outside it?
  4. (Lesson 9) Which rubric row must always be 2 before you send?
  5. (Lessons 3 and 7) Which three lines of your prompt template do most of the work?
Answers
  1. Who is this for? Why am I asking? What tone? What limits?
  2. What to keep.
  3. Outside, on a reliable source the chatbot didn't write.
  4. Row 5, Safe to send.
  5. Task, Context, and Shape and limits.

4. The idea

Listen: the key idea 0:38 · AI voice: Bella
Transcript

Here's the whole loop. One: ask clearly, with task, context, reader and tone, and shape and limits. Two: improve the answer with specific follow-ups: what's wrong, what to keep, one main change. Three: check anything that matters. Ask for checkable claims, then verify outside the chat. Four: protect privacy, and keep the decision yours. Share the minimum, and you make the call. One habit pulls it all together: score your prompt before you send it. Once you're used to it, that takes about twenty seconds, and it catches most problems before the chatbot ever sees them.

The loop:

  1. Ask clearly. Task, context, reader and tone, shape and limits. (Rows 1–4)
  2. Improve the answer. Specific follow-ups: what's wrong, what to keep, one main change. (Rows F1–F3)
  3. Check anything that matters. Ask for checkable claims, then verify outside the chat. (Row 6)
  4. Protect privacy, and keep the decision yours. Share the minimum; you make the call. (Row 5)

One habit pulls it together: score your prompt before you send it. It takes about twenty seconds once you're used to it, and it catches most problems before the chatbot ever sees them.

Which skill does this need?

When a reply disappoints, the fix is usually one of a few moves. Matching the problem to the move is a skill of its own:

What you see Likely missing Move
Generic, could be for anyone Context (row 2) Add the situation and facts
Wrong voice: too stiff, too jokey Reader and tone (row 3) Name the reader and the tone
Too long, wrong format Shape and limits (row 4) Give a number and a format
Almost right Iteration (F1–F3) Follow up: fix one thing, keep the rest
A number or rule you'd act on Check (row 6) Ask for the source, then open it yourself
You're tempted to paste a document with personal details Safe to send (row 5) Describe it; use placeholders

5. Worked example

Task: plan a casual dinner at home for four this Saturday. Small kitchen, moderate budget, one guest doesn't eat gluten. (That's the only dietary fact needed, and it isn't anyone's medical history.)

Before:

Dinner ideas for Saturday?

Rubric: Task 1, Context 0, Reader/tone n/a, Shape 0, Safe 2, Check 0 (gluten makes it matter).

After: see Prompt 1 below. Rubric: Task 2, Context 2, Reader/tone n/a, Shape 2, Safe 2, Check 2.

What changed: the "after" names the situation (four adults, small kitchen, budget, gluten), the shape (menu, grouped list, steps, time limit), a privacy line, and a built-in check ("if an ingredient sometimes contains gluten, say so"). It's longer, but every line earns its place.

6. Guided practice

This is a practice scenario. If you use the result for a real dinner, you still check what matters and cook safely for your guests.

Start a new chat and work through the four steps in order.

Step 1: Ask clearly

Tip: if you built a personal template in lesson 7, you can write this step with it instead of the prompt below. Keep the same facts and limits, then compare the replies.

Before sending, score Prompt 1 yourself with rows 1–6. (Spoiler: it should score well. Notice why.)

Prompt 1 (copy and paste):

Help me plan a casual dinner at home for four adults this Saturday.

Constraints:
- Small kitchen, no special equipment
- Moderate budget
- One guest does not eat gluten
- I want one main dish, one side, and one simple dessert
- Total hands-on cooking time under 60 minutes
- Give me: (1) the menu, (2) a shopping list grouped by produce, pantry, and other, (3) a short sequence of steps for the day

Do not include anyone's personal information. If an ingredient sometimes contains gluten, say so and suggest a gluten-free option.

What to look for: a three-part menu, a grouped shopping list, and short steps, with some attention to gluten. Dishes will vary: one chatbot might suggest a chicken main, another a gluten-free pasta, another a rice or potato dish. There's no single correct menu. You want a plan that respects the constraints and is clear enough to follow.

If it doesn't go as described: if the reply ignores gluten, runs long, or reads like a restaurant menu you'd never cook, don't start over. Go to Step 2.

Step 2: Improve the answer

Pick the follow-up that matches what you see, or write your own in the same spirit. Same chat.

Prompt 2a, if the plan feels too ambitious (copy and paste):

Please simplify. I want fewer pans, common ingredients I can find at a regular grocery store, and a dessert that does not need a mixer. Keep the gluten-free need in mind. Keep the same three-part structure: menu, shopping list, steps.

Prompt 2b, if the plan is simple but a bit vague (copy and paste):

This is a good direction. Please tighten it: make the shopping list specific (for example, "2 heads of broccoli," not just "broccoli") and add a rough timing note to each step. Flag any ingredient that I should double-check as gluten-free, such as soy sauce, broth, or oats.

Score your chosen follow-up on F1–F3. Prompt 2a points at the problem ("fewer pans… no mixer"), keeps the structure, and makes one main change (simpler).

What to look for: a clearer, more cookable version of the same plan. One more specific turn is fine: "Cut the dessert; I'll buy fruit instead" is a perfectly good third follow-up. If you wouldn't cook a dish, swap it. You're in charge of the menu.

Step 3: Check anything that matters

Pick one factual claim and verify it yourself. Use this prompt to find good candidates, then leave the chat to check.

Prompt 3 (copy and paste):

From the plan you just gave me, list two claims I should verify before I cook, such as a safe cooking temperature, a timing rule, or an ingredient that may contain gluten (for example soy sauce, broth, spice mixes, or oats). For each claim, name a type of official or manufacturer source I should check. If you are not sure, say so.

The chatbot may point you to a temperature, a "this is gluten-free" assumption, or an ingredient that often contains gluten. Regular soy sauce, for example, is often brewed with wheat, which is easy to miss on a "gluten-free" menu. The chatbot may or may not have been careful. That's why you check.

Then check outside the chat:

  • If the plan includes poultry (chicken or turkey), look up the safe minimum internal temperature on the U.S. Department of Agriculture's page: fsis.usda.gov safe temperature chart. As of November 2026, USDA FSIS says all poultry should reach 165°F (73.9°C) on a food thermometer. If the chatbot said something different, trust the USDA page.
  • If the plan includes an ingredient that might contain gluten, check the package label when you shop, or the manufacturer's site. Don't treat the menu as a guarantee.
  • If the plan is vegetarian with no risky ingredient, check any cooking time or temperature the chatbot gave against a trusted cookbook or an official food-safety page.

Write one sentence for yourself: "I checked ____, and it was ____." That sentence is the skill.

This is educational practice, not personal dietary or food-safety advice for your household. When you cook for others, follow official guidance and your guests' needs.

Step 4: Protect privacy, and keep the decision yours

Look back at your prompts. You shouldn't have pasted a guest's real name, address, health history, or anything else private. "One guest does not eat gluten" is enough.

For a real dinner, you'd also:

  • leave guest contact details out of the chat;
  • not upload a photo of a pantry or fridge if it shows mail, names, or personal papers;
  • make the final call yourself, especially for an allergy or serious sensitivity. When stakes are high, confirm with the guest and with labels, not with the chatbot alone.

If anything still feels off, change it. A good chatbot draft is a starting point, not a promise.

7. Independent practice

Which skill does this need?

For each situation, name the missing row and write a one-line fix. Use the table in section 4.

  1. You asked for "a birthday message for my coworker" and got something so generic it could be for anyone.
  2. The chatbot's party budget says a venue costs "$300 on average in most cities."
  3. The reply is great except it's twice as long as you need.
  4. You're about to paste your whole lease into the chat to ask about the pet policy.
Answers (one strong fix each)
  1. Row 2, Context. "She's retiring after 12 years on our team, loves gardening; keep it warm and two sentences."
  2. Row 6, Check. Ask "Where does that figure come from? Name a source," then check real local prices yourself. Don't repeat an "on average" figure you haven't checked.
  3. F1–F3, Follow-up. "Keep everything, but cut it to half the length."
  4. Row 5, Safe to send. Don't paste the lease with names, addresses, or account details. Copy only the pet clause, with names removed, or describe it: "My lease has a clause about pets that says [paraphrase]. What questions should I ask my landlord?"

Fix-it drill (no hints)

Rewrite this prompt so it scores at least 1 on every row and 2 on row 5:

Plan my mom's 70th. Her address is 18 Elm Street and she has diabetes. Budget whatever.
One strong version (not the only one)
Help me plan a small 70th birthday lunch at home for about 12 guests. Budget about [AMOUNT]. One guest needs low-sugar options. Give me: a simple menu, a timeline for the day, and a short shopping list grouped by store section. Flag any food-safety or dietary claim I should double-check, and say if you're unsure about anything.

Why it's stronger: the address and the specific medical condition are gone (row 5), "low-sugar options" gives the chatbot what it needs without a diagnosis, the shape is clear (row 4), and a check is built in (row 6). For a real event, confirm dietary needs with the guest, not the chatbot.

8. Rubric self-check (full Prompt Check)

Score your fix-it drill and Prompt 1 with all rows:

Row Question 0 1 2
1. Task What do I want done and back? Neither One Both
2. Context Situation and facts? None Some Enough for a stranger
3. Reader and tone Who's it for, how should it sound? Neither One Both (or n/a)
4. Shape and limits Length, format, include, avoid? None Vague Specific
5. Safe to send Fine for a stranger at the company to read? Private details One unneeded detail Only what's needed
6. Check built in Sources or uncertainty requested, and a plan to check outside? Stakes real, no check Partial Yes (or n/a)
F1–F3 (follow-ups) Points at the problem? Says what to keep? One main change? — — —

Reading your score: row 5 must be 2. Then fix your lowest row. A 2 in every row doesn't guarantee a perfect reply; it means you gave the tool a fair chance and kept yourself safe.

9. Retrieval check

  1. List the four parts of the loop, in order.
  2. Why was "one guest does not eat gluten" enough, without adding a medical history?
  3. The menu says to bake chicken "until the juices run clear." What should you do before treating that as the safety test?
  4. True or false: once you can run this loop, you no longer need to check short, everyday answers.
  5. A reply is generic. Which rubric row is most likely low?
Answers
  1. Ask clearly. Improve the answer. Check anything that matters. Protect privacy, and keep the decision yours.
  2. The chatbot only needed the constraint that affects the menu. More medical detail would be unnecessary private information.
  3. Check an official food-safety source. As of November 2026, USDA FSIS says to use a food thermometer and cook all poultry to 165°F (73.9°C). "Until the juices run clear" isn't a substitute.
  4. False, but with a nuance: you don't verify every casual sentence. You check when being wrong would matter: food safety, allergies, money, health, or anything you'll repeat as fact.
  5. Row 2, Context.

10. Recap

Listen: recap 0:27 · AI voice: Bella
Transcript

You now have your first AI skill: a loop you can take into any chatbot. Ask clearly. Improve the answer. Check anything that matters. Protect privacy, and keep the decision yours. And score before you send: row five must be two, then fix your lowest row. The wording of replies will change from tool to tool and day to day. The loop doesn't. This week, run the loop once on a real task you actually need.

You now have your first AI skill: a loop you can take into any chatbot.

  • Ask clearly: task, context, reader and tone, shape and limits.
  • Improve the answer: point at the problem, say what to keep, one main change.
  • Check anything that matters: ask for checkable claims; verify outside the chat.
  • Protect privacy, and keep the decision yours: share the minimum; you decide.
  • Score before you send. Row 5 must be 2; then fix your lowest row.

The wording of replies will change from tool to tool and day to day. The loop doesn't.

You can now: run the whole loop on an everyday job and grade your own prompts. Use it this week: run the loop once on a real task: a message, a plan, or an explanation you actually need.

Next: lesson 11, the Prompt Practice Lab and mastery check, where you show yourself you've got it. After that, the full path continues with how these tools work, everyday jobs, and deeper prompt craft. A paid full course is planned; details will be on this site when they're ready.

11. Talk it through with your coach

  • Coach my prompt: before Step 1, say your own version of the dinner prompt out loud. Your coach walks you through the six rows as questions ("Who's eating? Any limits? Anything private in there? What would you check?") and you revise it aloud. It doesn't write the prompt for you.
  • Text-only alternative: the rubric in section 8.

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

  • U.S. Department of Agriculture, Food Safety and Inspection Service, "Safe Minimum Internal Temperature Chart," checked November 2026: fsis.usda.gov
  • OpenAI Help Center, "Does ChatGPT tell the truth?" (verify important information from reliable sources): help.openai.com
  • OpenAI Help Center, "Prompt engineering best practices for ChatGPT," checked 22 November 2026: help.openai.com