Which tool to choose
In the assistant's settings, look for the memory section: where it exists, it shows the list of things it has saved about you, and next to each one you'll find a way to edit or delete it. Alternatively, act through words inside the chat, asking it to forget or update a piece of information. The best tool is the one that lets you see the list: being able to read what the AI thinks it knows about you is the first step to truly controlling it.
How to do it
- Open the settings and look for the memory or recollections item.
- Scroll through the list of what the AI has saved: read it all, at least once.
- Delete the outdated or wrong entries; correct the imprecise ones.
- For on-the-fly changes, say it in the chat: "forget the old address, I now live at..." and check that the entry has updated.
A concrete example
Months earlier, Anna had told the AI she worked as a freelance graphic designer, and the assistant had memorized it. Having changed jobs, she kept receiving suggestions tuned to freelancing. She went into the memory section, found the entry "works as a freelance graphic designer", corrected it to "works as a graphic designer at an agency", and from that moment the advice became relevant again. It only took reading the list once to discover an old truth that had become false and was polluting her answers.
When it DOESN'T work (and how to fix it)
If you can't find the memory section
Not all assistants have one, and where it exists the button often moves around. Try asking directly in the box: "what have you memorized about me?" and then "forget this". If the assistant doesn't handle a memory at all, it means it isn't saving anything between one chat and another, and there's nothing to clean up.
If you delete but the AI still seems to remember
Within the same conversation the AI keeps seeing what has already been said there, even if you've cleaned the long-term memory. Open a new chat to start fresh: the deleted memory won't come back, but an old open conversation still contains its text.
If you want to wipe everything
Many assistants let you empty the entire memory in one go. Use it when you feel it has become a confused pile-up: starting over from an empty memory and rewriting only what matters is often better than chasing one wrong entry at a time.
A tip from someone who really uses it
Reread the memory every now and then, the way you'd empty a drawer that fills up by itself. The assistant tosses in things said in passing, and over time old or misunderstood details accumulate and steer the answers without your noticing. Five minutes of cleanup let you discover what the AI believes about you, sometimes with surprises. A well-tended memory works for you; a memory left to itself becomes a voice behind your back, repeating things that are no longer true.
Frequently asked questions
Does the AI memorize everything I tell it?
No, in general it saves only what seems to be a stable fact about you, and it doesn't always guess what matters. That's why the list should be read: sometimes it has noted down something irrelevant and ignored the important one.
Can I prevent it from memorizing at all?
Often yes: where the memory exists, there's usually also the switch to turn it off. With memory off, every chat starts fresh and nothing is saved from one conversation to the next.
Is deleting an entry immediate?
As a rule, yes: once removed, the AI won't use it in new conversations. Keep in mind that chats already open still contain what was written in them.
If I delete the conversation, does it disappear from the memory too?
No, and this is the misunderstanding that leaves old data lying around that you thought was removed. The conversation and the long-term memory are two different things: you can delete a chat and still have, in the memory, a fact the AI drew from it; and you can empty the memory and still find that fact written in an old open chat. To truly remove a piece of information, go to the memory section and remove it there: counting on deleting the chat to clean the memory is the most common way to end up with suggestions based on something you thought you'd eliminated.