Lesson 5 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 5

Improve the answer instead of starting over

Time
about 25–30 minutes
Prompting skill
P5 Specific follow-ups (iteration), with P1–P4 review
You'll need
Your free AI chatbot, and a new chat for this exercise.

Listen Talk to your coach


1. Objective

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

  • Revise a chatbot's draft with follow-ups that name what to change and what to keep.
  • Judge when to send another follow-up and when to start a new chat with a stronger first prompt.

2. Hook

Listen: welcome 0:11 · AI voice: Bella
Transcript

The first reply is a draft, not a verdict. In this lesson you'll learn to steer it, one clear follow-up at a time, instead of accepting it as is or starting over.

A friend offers to help you pack for a trip and hands you a list of 31 items, including a hair dryer and three sweaters, for a two-night city weekend in a carry-on.

You wouldn't throw the list away and you wouldn't say "do better." You'd say, "Shorter. It's two nights, I already have toiletries, and I need one nice outfit." That one sentence is the whole skill in this lesson.

3. Warm-up (from lessons 1–4)

  1. (Lesson 2) Name the four questions to glance over before you send a prompt.
  2. (Lesson 2) A reply includes a fact you never gave it. What do you do?
  3. (Lesson 1) True or false: the chatbot's first reply is a stored answer it looked up.
  4. (Lesson 4) In Pair 5 (the resume), why did the prompt ask for [NUMBER] placeholders?
Answers
  1. Who is this for? Why am I asking? What tone? What limits?
  2. Treat it as invented, and remove or correct it.
  3. False. It's written fresh by predicting likely next words, so it's a draft.
  4. So the chatbot wouldn't invent numbers or achievements. You fill in the real figures.

4. The idea

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

Most people either accept the first reply or delete everything and start again. There's a better middle path: iterate. That means improving it in steps, with a follow-up in the same chat. A vague follow-up, like make it better, leaves the model guessing again. A strong follow-up does three things. It points at the problem. It says what to keep. And it makes one main change. For example: keep the warm tone, and cut it to under one hundred words. Two or three focused follow-ups are normal. If it keeps missing after that, start a new chat with a stronger first prompt. That isn't failure. That's you steering.

Even with good context, the first reply is a draft: a starting version, not a finished product.

Many people either accept that first draft or delete everything and begin again. There's a better middle path: iterate. To iterate means to improve something in steps, using what you already have. In a chatbot, you iterate with a follow-up, a second (or later) prompt in the same chat.

OpenAI's own guidance for ChatGPT users says the same thing: prompting "often requires an iterative approach": start, review the response, and refine (see Sources).

What a good follow-up sounds like

A vague follow-up like "Make it better" or "I don't like this" leaves the model guessing again. It may change the tone, add decorations, or rewrite everything without fixing the part that bothered you.

A useful follow-up does three things. These are the three follow-up rows of your rubric:

Row What it means Weak Strong
F1. Points at the problem Names what's wrong or missing "Make it better." "Cut this to 12 items."
F2. Says what to keep Protects what's already good (nothing) "Keep the warm tone."
F3. One main change Makes it clear what this round fixes "Shorter, funnier, more formal, totally different." One main fix, plus small details

Useful kinds of follow-ups:

  • Shorter or longer: "Keep it under 100 words."
  • A missing fact or request: "I already have toiletries. Don't list those."
  • A change in emphasis: "I need one nicer outfit for a dinner. Make that obvious."
  • A format change: "Group the list as Clothes and Other."
  • A correction: "I said two nights, not a week. Please revise."

In the same chat, the model can see its last reply, so you don't need to paste it back. If you start a new chat, you do.

When to stop following up

Two or three focused follow-ups are normal. If the reply keeps missing the point after that, start a new chat and write a stronger first prompt using what you learned. That isn't failure. That's you steering. (Long chats can also drift; you'll learn why in Module 2.)

5. Worked example

Situation: you asked for a short email to a coworker. The reply is warm and clear but twice as long as you wanted, and it buries the request at the end.

Weak follow-up:

Make it better.

Strong follow-up:

The tone is good, so keep it. Please cut it to about half the length and put my request (can we move the meeting to Thursday?) in the first sentence.

What changed:

Change Row
"Cut it to about half the length… put my request in the first sentence" F1: points at the problem
"The tone is good, so keep it" F2: says what to keep
Length and order are one fix: "make the request easy to see" F3: one main change

6. Guided practice

Start a new chat. You'll ask for a generic packing list on purpose, then revise it into something realistic.

Prompt 1 (copy and paste):

Give me a packing list for a weekend trip.

What to look for: probably a fairly generic list (clothes, toiletries, chargers, maybe extras you don't need). It may be long. That's a typical first draft when the prompt is thin.

Before you go on, score Prompt 1 on rows 1–4 from lesson 2. (It's weak on purpose. Row 2, Context, is close to 0.)

Now improve it. Stay in the same chat.

Prompt 2 (copy and paste):

Revise that list. It's a 2-night trip to a city. I'll be walking a lot, staying in a hotel, and I have a small carry-on only. I already have toiletries. I need one nicer outfit for a dinner. Give me at most 12 items, grouped as Clothes and Other. No brand names.

What to look for: a shorter list in two groups, with a nicer dinner outfit included and toiletries left out (or clearly noted as already packed). The exact items will vary. One tool might suggest a compact umbrella, another a water bottle. Your job is to judge whether the list is usable, not whether it matches one "correct" set.

Notice that Prompt 2 is really a follow-up that adds the context Prompt 1 lacked. That's often the fastest fix.

If it doesn't go as described: if the list is still long, or toiletries sneak back in, that's useful. Send one more focused follow-up.

Prompt 3 (copy and paste):

Cut two items you think I can skip, and tell me why in one sentence each. Keep the nicer dinner outfit.

What to look for: two items removed, each with a short reason, and the dinner outfit still there. Read the reasons. Keep them if they make sense for your trip and ignore them if they don't. The chatbot suggests; you decide what goes in the bag.

Check Prompt 3 against the follow-up rows: it points at the problem (F1: cut two), says what to keep (F2: dinner outfit), and makes one main change (F3).

7. Independent practice

Spot the weak follow-up

You asked for a weekly dinner plan and got seven complicated recipes. Which follow-up is weakest?

  • A. Too much. Redo.
  • B. Keep the same seven days, but swap in dinners that take 30 minutes or less and use common grocery-store ingredients.
  • C. Keep Monday and Thursday as they are. Make the other five simpler.
Answer

A. It barely points at the problem (F1 = 0 or 1), keeps nothing (F2 = 0), and the model has to guess what "too much" means. C is better: it keeps two days (F2 = 2), but "simpler" is vague (F1 = 1). B is strongest: it names the problem precisely and keeps the structure.

Fix-it drills

Rewrite each weak follow-up into a strong one. Invent any reasonable details.

Drill 1 (with a hint). You asked for a toast for a friend's retirement party. It's lovely but runs about three minutes, and you want about one minute.

Shorter.

Hint: How short? What should stay?

Drill 2 (no hint). You asked for an explanation of how a credit score works. It uses words like "utilization ratio" that you don't know.

I don't get it.
One strong version of each (not the only one)

Drill 1:

Keep the story about the fishing trip and the closing line. Cut everything else down so the whole toast takes about one minute to say out loud, roughly 130 words.

Drill 2:

Please explain that again for someone who's never studied finance. Define "utilization ratio" in everyday words, and use one simple example with round numbers. Keep it under 150 words.

(Drill 2 shows that a follow-up can also fix row 3, Reader: "someone who's never studied finance.")

8. Rubric self-check

Score your two drill follow-ups:

Row Question 0 1 2
F1. Points at the problem Does it name what's wrong? "Make it better." "Shorter." "About one minute, roughly 130 words."
F2. Says what to keep Did I protect what's good? Nothing said Implied Named ("keep the fishing story")
F3. One main change Is it clear what this round fixes? Many unrelated changes Two or three related changes One main change

Also re-score Prompt 1 from section 6 with rows 1–4. Then ask: Would a stronger first prompt have saved me a follow-up? Often it would. That's why lessons 2 and 5 work together.

Where to next: If F1 is low, add a number or a specific example of what you want. If F2 is low, start your follow-up with "Keep…".

9. Retrieval check

  1. What does it mean to iterate with a chatbot?
  2. Why is "Make it better" usually a weak follow-up?
  3. Name the three follow-up rows.
  4. True or false: if the second reply still isn't right, the only option is to switch tools.
  5. (From lesson 2) Which rubric row asks, "Did I say who it's for and how it should sound?"
Answers
  1. Improving the draft in steps, with follow-ups, instead of accepting the first reply or throwing it away.
  2. It doesn't say what "better" means, so the model has to guess the tone, length, format, or content you want.
  3. F1 Points at the problem; F2 Says what to keep; F3 One main change.
  4. False. Send another focused follow-up, or start a new chat with a stronger first prompt. Switching tools is optional.
  5. Row 3, Reader and tone.

10. Recap

Listen: recap 0:24 · AI voice: Bella
Transcript

Recap. Treat the first reply as a draft. Strong follow-ups point at the problem, say what to keep, and make one main change. Adding missing context is often the fastest follow-up. After two or three tries that miss, start fresh with a better first prompt. And you decide what to keep. Next time a reply is almost right, write a follow-up that starts with the word keep.

  • Treat the first reply as a draft.
  • Strong follow-ups point at the problem, say what to keep, and make one main change.
  • Adding missing context is often the fastest follow-up.
  • After two or three tries that miss, start fresh with a better first prompt.
  • You decide what to keep. The chatbot suggests; it doesn't pack your bag or send your message.

You can now: steer a draft to fit in two or three focused follow-ups. Use it this week: the next time a reply is "almost right," write a follow-up that starts with "Keep…".

Next: lesson 6, Prompting Lab 3: mixed fix-this-prompt drills, including weak follow-ups.

11. Talk it through with your coach

  • Coach my prompt: your coach reads you a too-long draft and you speak a follow-up. It asks, "What exactly is wrong? What should stay?" until your follow-up scores 2 on F1 and F2. It doesn't write the follow-up for you.
  • Text-only alternative: the drills and rubric in sections 7–8.

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

  • OpenAI Help Center, "Prompt engineering best practices for ChatGPT" (iterative refinement: start, review, refine), checked 22 November 2026: help.openai.com
  • For why a fluent first draft can still include guessed details: OpenAI, "Why language models hallucinate": openai.com