Why this matters
An AI draft can state a policy or a fact you never gave it, in exactly the same confident voice as everything else. Once it reaches a client, you own it: a refund you did not offer, a deadline you cannot meet, a claim you cannot back up.
The usual reaction is to delete the draft and start over, or to type "make it better" and hope. Both waste the time AI was meant to save, and neither one finds the made-up parts.
In this module you learn a short routine that fixes the draft and checks every fact: point at the exact problem, add rules, ask what it assumed, check the facts with sources you actually open, and know when to start a fresh chat.
What you’ll build
A customer email for Juniper Lane Bakery, a small made-up business. The owner, Dana, asks for an email about cake pickup, late pickups and freshness, but leaves out the facts. The first draft has a stiff paragraph, a late-pickup policy she never gave it, filler at the end, and a freshness promise she cannot back up.
You fix each problem by name, add a word limit and a must-include list, get ChatGPT to admit what it guessed, check the food-safety fact against an official source you open yourself, and write the next message in a clean new chat. You finish with a fix-it checklist in your Prompt Templates doc.
What you need
- Nothing to start the practiceChatGPT, the source page and Google Docs are practice copies. The yellow highlights are lesson hints, not part of ChatGPT.
- A ChatGPT account, to do it for realA free account at chatgpt.com works. For step 5 you need its web search, which free accounts have.
- Your Prompt Templates docThe Google Doc from Prompt Writing 101 (docs.new, named Prompt Templates). The fix-it checklist goes under your earlier templates.
- The real factsFor your own business (prices, hours, policies) you are the source, so have them to hand. For facts about the world, an official site you trust, like a government agency.
Concepts
4 ideas you’ll use in the practice.
Point at the exact problem
"Make it better" leaves the AI guessing what is wrong. "The paragraph about picking up late is too formal. Rewrite it the way I would talk to a regular client" tells it where to look and what to change.
Stay in the same chat for small fixes like this. The AI can see the draft, so you only describe the change.
Constraints turn taste into rules
A word limit, a reading level ("short sentences, everyday words"), what it must include and what to avoid. Each one is a rule it can follow, instead of a preference it has to guess.
Ask what it assumed, then check with a source you open
Made-up facts sound exactly as confident as real ones. Ask: "List every assumption you made, and every fact I should check." The guesses come into the open.
For a fact about the world, ask it to search the web, use official sources and cite them. Then open at least one source and find the line yourself. A citation you never open proves nothing.
Know when to start fresh
When the task changes, or the AI keeps bringing back something you already fixed, start a new chat with one clean prompt: your facts, your rules, and "Use only these facts. If you need something I didn’t give you, ask me instead of guessing."
Practice
Do every step yourself in a practice copy of the real tool. It checks your work, and “Do it for me” finishes any step you get stuck on.
Do it for real
Now repeat it in your own account. Tick each step as you go.
Downloads
- Fix-it checklist (.txt)The six steps you save at the end of the lesson, with fill-in prompts for each one.
Quick check
3 questions to check you’ve got it.
1. Which fix request works best?
- "Make it better"
- "The second paragraph is too formal. Rewrite it the way I talk to regular clients."
- "Try again"
Show the answer
"The second paragraph is too formal. Rewrite it the way I talk to regular clients.". Naming the spot and the change tells the AI exactly what to do.
2. ChatGPT cites a source for a safety fact. Is it checked?
- Yes, a citation means it is right
- Not until you open the source and find the same line
- Only if the source is long
Show the answer
Not until you open the source and find the same line. A citation you never open proves nothing. Open one and confirm it.
3. When should you start a new chat?
- After every message
- When the task changes, or it keeps bringing back a mistake you fixed
- Never, longer chats are always better
Show the answer
When the task changes, or it keeps bringing back a mistake you fixed. A clean prompt in a new chat drops the old mistakes and keeps only your facts.
Break Big Jobs into Steps: Prompt Chaining
You can fix one answer. Next, big jobs that are too much for one prompt: build a 30-day marketing plan in four small steps, checking each answer before the next one builds on it.
Start module 5 →