How to make ChatGPT, Claude and Gemini remember you (and forget what you want)

This is where most of the confusion comes from. Each assistant has two separate things it calls memory.

The first is saved facts: short notes like "works in marketing in Pune" or "prefers short answers". You can read these in a list and delete them one by one. The second is searching your old chats: the model looks back through past conversations when it thinks they are relevant. You cannot edit that one, only switch it on or off.

Deleting a chat removes it from the second kind. It does not always remove a fact already saved into the first kind. That is the mistake people make, and it is why "I deleted the chat, why does it still know?" is one of the most common complaints I see.

Open your profile picture, then Settings, then Personalization. Under Memory there are two switches: Reference saved memories and Reference chat history. Turn on both for the full effect. Turning off saved memories switches off chat history too.

Press Manage memories to read the list. Delete anything you would not want in there. OpenAI's help pages say a newer memory system is rolling out, so the labels may move again. If they have, the two switches will still be under Personalization.

As of October 2026 we could not load OpenAI's help pages from here to confirm the exact wording, so if a label differs by a word, look for the one that mentions saved memories.

In Claude, open Settings, then Your Claude (older accounts still see it as Memory). Memory is on by default on Free, Pro and Max. On Team and Enterprise an owner switches it on and then each person turns it on for themselves.

The switch called Memory from chats is the saved-facts kind. Search and reference chats is the old-chats kind, and it needs a paid plan. There is also Include sensitive topics in memory, which I leave off: it decides whether health, money and similar topics get stored.

Press Manage next to Your memory to see what it holds, as a list of topics. You can edit or delete any topic, or just tell Claude in a chat to remember, change or forget something. Pause memory stops it without deleting anything. Reset memory deletes everything, including project memories, and cannot be undone.

Google renamed this in 2025 and it still trips people up. The setting is Keep Activity, and it lives on your Google activity page at myactivity.google.com/product/gemini, not inside the chat window.

With Keep Activity on, your chats are saved to your account and Gemini can draw on them. With it off, new chats stay for 72 hours so Gemini can respond and process feedback, then go, and they are not used to train Google's models. You can review and delete activity on the same page at any time.

Memory fills up with trivia if you leave it to chance. The fix is to give it the five things that actually change how it should answer you, once, in plain words. Paste the prompt below into a fresh chat in each assistant, fill the brackets, and send. Then open the memory list and check it saved what you meant.

Do not put passwords, account numbers or anything about other people in here. It is a note the model reads before every answer, not a vault.

Sometimes you want the opposite. All three have a no-trace mode. In ChatGPT it is Temporary chat, from the model menu at the top of a new chat. In Claude it is the ghost icon in the top right, which opens an incognito chat, on every plan. In Gemini it is Temporary chat in the new chat menu; those chats are kept for 72 hours and not used for training.

In these modes nothing is saved to memory and the chat does not appear in your history. It is the right place for a medical question, a salary negotiation, or anything about someone else.