Five minutes, once
Settings pages are where good intentions go to die. There are dozens of switches and no indication which of them matter, so most people open the page, feel vaguely tired, and close it again.
So here is the short version. Four things are worth your attention. Everything else you can leave exactly as it is, possibly forever.
Look for settings behind your name or picture, usually in a corner of the screen. Names and positions shift as the app is redesigned, so hunt by meaning: if a setting described here is not where you expect, it has moved, not disappeared.
In plain English
- Training:
- Using conversations to help improve the AI for future versions. Whether yours are used for this is a setting you control.
- Temporary chat:
- A conversation that is not saved to your history and does not feed memory. A private mode for one-off questions.
- Data controls:
- The section of settings holding the privacy switches. Sometimes called data controls, sometimes something similar.
- Opt out:
- Choosing not to take part in something that is switched on by default. Opting out here does not cost you any features.
1. Whether your chats train the model
This is the one that matters most, and the one nobody finds.
By default, conversations on consumer accounts may be used to help improve the model. There is a switch, somewhere in the data controls area, that turns this off. Turning it off does not degrade your answers, does not cost anything and does not remove any features. It simply means your conversations are not drawn on for future training.
Should you switch it off? If anything you discuss touches work, clients, health, money or other people's private business, yes, and without much agonising. If you use it purely for recipes and crossword clues, it matters considerably less. When unsure, off is the low-cost choice, because you lose nothing by it.
Switching this off does not delete anything already sent, and it does not mean your chats vanish. They still sit in your history on the company's systems until you delete them. The setting governs training, not storage.
2. Temporary chats, for the things you would rather not keep
There is usually a way to start a conversation that is not saved to your history and does not touch memory. It goes by names like temporary chat, and it is the equivalent of a private browsing window.
Useful more often than you would think. The medical worry you want explained before deciding whether to bother the doctor. The blazing row you want help wording a reply to. The salary question. Anything you would rather not scroll past in six months.
3. What memory currently holds
Lesson 4 covered this, and here is the reminder to actually do it: open the memory list and read it. Delete anything you would not want kept. It takes two minutes and people are consistently surprised by at least one entry.
4. Your chat history
You can delete individual conversations or clear the lot. Worth knowing it exists, and worth doing occasionally, particularly if you shared something you would rather not have stored.
Same job at Claude and Gemini. Both have training controls and history controls in their own settings, in their own places, with their own wording. If you use more than one, do this exercise on each of them.
Your starter prompt pack
Here is the capstone, and it deliberately does not live on this site. LearnAI has no accounts and stores nothing about you, which means the useful thing to build is something that lives with you rather than with us.
Copy the block below into a note on your phone, a document, wherever you keep things. Then fill in the parts in square brackets so they are yours. That is your prompt pack: five openers that cover most of what beginners actually need, ready whenever the empty box makes you hesitate.
MY STARTER PROMPT PACK
-
THE STANDING BRIEF (paste into custom instructions)
About me: I am [who you are and what you do]. Most of my writing goes to [who you write to].
How to respond: Plain words, short sentences, no corporate phrasing. Answer first, reasoning after. If a question is unclear, ask me rather than guessing.
-
THE DIFFICULT MESSAGE
Help me write a message. I am [your role], writing to [recipient]. What happened: [the facts, in any order]. What I want: [the outcome]. Tone: [warm, firm, formal]. Length: under [number] words.
-
THE DOCUMENT I DO NOT WANT TO READ
I have attached [what it is]. In plain English: what does it ask of me, what are the deadlines, and what is unusual or worth being worried about?
-
THE FIX
That is not right because [what is wrong with it]. Keep [what worked] and change [what did not]. Try again.
-
THE THINKING-OUT-LOUD ONE
I am going to describe a problem and it will not be organised. Listen, then give me back the three main things I actually said and one question I have not thought about.
Two more worth keeping nearby, for the moments the tool goes quiet or goes wrong.
Before you answer, ask me the three questions that would most improve your answer. Do not answer until I have replied.
Which parts of that answer are you confident about, and which parts should I check myself before relying on them?
That last one is the single best habit in this whole course. Asking it to mark its own uncertainty will not catch everything, but it catches a surprising amount, and it keeps you in the position you should always be in: the person deciding, with a very fast assistant.
That is the course. You have earned the First Chat badge, which is a small thing on a website and also, genuinely, the hardest step in learning any of this: you have stopped being the person who has never tried.
Checkpoint
Turn off training if your chats touch anything private, use temporary chats for sensitive questions, prune memory and history, and keep a prompt pack of your own.
๐ Quiz
Question 1 of 4What does turning off the training setting actually do?