You will be able to set an assistant to keep less about you, and explain what deleting a chat does and does not do.
Mei Ling is a nurse in Jurong (an invented example). Some months back she used an AI assistant to make sense of a hospital letter about her mother, and the assistant saved her mother's diagnosis as a memory. She had not meant it to be kept.
Try a quick test on your own assistant. If memory is switched on, ask it what it knows about you. Plenty of people find the reply surprising. It might list your job, your child's school name, the flat you are saving for, a sore back and a difficult manager. None of these was secret when you mentioned it, but collected in one place they read like a profile of you.
Four controls decide how much an assistant holds on to: chat history, memory, temporary chats and deletion. ChatGPT, Claude, Gemini and Copilot each name them differently. Menus also shift whenever an app is redesigned, so the notes below tell you what to look for in settings and leave out exact menu paths.
Most assistants save each conversation to your account unless you change the setting. That saved record is your chat history, the column of past chats down the side of the screen. It helps when you want to carry on with a long piece of work or dig out a reply you liked last week. Lesson 1.1, "What happens to the words you type into an assistant", explains the downside. Each saved chat is a copy of your words sitting on the provider's servers, and depending on your plan and settings it may be used for training.
Many assistants have a mode that keeps a conversation out of your history. In ChatGPT this is a temporary chat, and other products call it incognito or something similar. Make it your default for a one-off question you won't come back to, or for anything you would prefer not to keep. Before you rely on it, read the short description beside the switch. Some providers hold temporary chats for a short time for safety checks, even though they never show up in your list.
Memory works differently from history. History stores whole conversations. Memory picks out facts about you and brings them into later conversations, so the answers fit you without you explaining yourself again. On some assistants memories are saved automatically from what you say. On others they are saved only when you ask the assistant to remember something.
Settings usually let you view the saved memories, remove single items, or switch memory off completely. Read through that list at least once. Having memory on is fine in itself. What matters is what the list contains and whether you would mind someone reading it over your shoulder.
When Mei Ling opened her list, she found the note about her mother's diagnosis and deleted just that entry. She kept memory on for things like her preferred writing style. Leaving memory on and removing the sensitive items is a sensible middle path.
Deleting a chat takes it off your list and stops you from opening it. That is all you can confirm from your end. Providers usually hold deleted data for a further period before it leaves their systems, and the privacy policy says how long. They may keep it longer where the law requires, such as during a legal case or an abuse investigation. If the conversation had already been used for training before you deleted it, the trained model keeps whatever it learned from it.
Deleting is still worth doing, but it does not undo having typed something. The safer habit from lesson 1.1, "What happens to the words you type into an assistant", still applies: choose what goes in before you paste it.
Closing your whole account is a separate step from deleting chats. It usually clears history and memory together, and the privacy policy sets out what stays and for how long.
Uploaded files and connectors often sit outside the history and memory settings. A PDF or spreadsheet you uploaded may be stored apart from the conversation, in a files area or in a project folder you have since forgotten about. Check whether your assistant lists your uploads anywhere, and whether removing a chat also removes the files attached to it.
Connectors are links that let an assistant read your email, calendar, cloud drive or other apps. Each one is a standing permission. As long as it is on, the assistant can go into that service whenever a task needs it. Some connectors are run by third parties under their own privacy terms. If you switched on a connector once to test something and never used it again, it is still open.
Set aside about ten minutes. Start with your saved memories: read every item and delete anything you would mind someone seeing. Next, check whether chat history is on, and decide whether to use temporary chats for personal questions. Then look for an uploaded files area and clear out documents you no longer need. Finally, open the list of connected apps and switch off any you don't actively use.
You'll do the first and last of those for real in a moment: going through your saved memories, and switching off connected apps you don't use. Keep a note of anything that surprised you, because lesson 1.4 asks you to look at every AI tool you've used this month, and the habits you build here are the ones you'll repeat there.
Go through your assistant's memory, delete anything you would not want stored, and switch off any connector you do not use.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).