Stage 0: Ask, part 2 of 2
Stay in control
3 more lessons: fix an answer without starting over, check what it tells you, and keep private details private. About 10 minutes.
Lesson 4: Reply to fix, don’t restart
When an answer is close but not right, reply with notes, the way you would mark up a draft. Say what to keep and what to change.
First, ask for a letter
Write a short letter to my landlord asking for a repair. Two paragraphs.
Then reply with notes like these
Keep the first paragraph. Make the second paragraph shorter and firmer. Add that the heater broke on Monday. Change nothing else.
What you should see: only the parts you named have changed.
Still wrong after about three tries? Start fresh. Ask for a summary, then paste it into a new chat.
To start fresh, paste this
Sum up everything I've told you in this chat as one short block I can paste into a new chat.
If it doesn’t look right
- If it changed things you liked, name what to keep and ask again.
- If it keeps missing, use the summary and a new chat.
Lesson 5: Check before you trust
AI can sound sure and still be wrong. A long, confident answer is not always a right one. Give it room to say it isn’t sure, and ask what you should check.
Try this
Can cooked rice be reheated twice? If you're not sure, say so. Then list what I should double-check, and where.
What you should see: an answer, plus a short list of things to check and where to check them.
If it doesn’t look right
- If it names sources, open them. AI sometimes makes sources up.
- For health, money or legal questions, confirm on an official site before you act.
Lesson 6: Know what not to paste
What you type may be stored. Swap private details for placeholders in square brackets before you paste.
Leave these out
- passwords, PINs and login codes
- ID, bank account or card numbers
- health details
- other people’s private details
Try this
Tidy up this note. I've replaced the private details with [NAME], [ADDRESS] and [AMOUNT]. Keep those placeholders exactly as they are. [paste your note here]
What you should see: your note, tidied, with the placeholders still in place.
You can also stop your chats being used to train the AI.
In Claude
Settings, then Privacy. Turn off “Help improve our AI models”. Incognito chats are not used for training.
Checked on a live screen, October 2026.
In ChatGPT
Settings, then Data controls. Turn off “Improve the model for everyone”. A Temporary Chat is not used for training.
From OpenAI's help pages, October 2026. Not yet checked on a live screen.
If it doesn’t look right
- If it fills in a placeholder with a guess, reply: “Keep my placeholders exactly as written.”
- If you pasted something private by mistake, delete that chat.