Which tool to choose
Cross-conversation memory exists on the main assistants, but with different philosophies: on ChatGPT it's partly automatic (it gathers cues from past chats) and partly explicit (the notes you ask for); Claude keeps a memory in the form of readable, editable files; Gemini also personalizes based on your activity. For control, in all of them what you explicitly tell it to remember matters more than what it gathers on its own.
The interfaces change: if you can't find the memory setting, look for it in the personalization or privacy section of the settings.
How it's done
- Turn on memory in the assistant's settings (usually under personalization or data management).
- Tell it explicitly the stable facts you want it to keep. The operational syntax:
Remember for the next conversations that: I work as [your role], I write for an audience of [who reads you], and I prefer short, direct answers. Take this into account from now on.
- Check in the memory settings what it has actually saved, and correct or delete the wrong notes.
- If memory isn't available or isn't enough, keep a short "context sheet" ready to paste when you open a new chat.
A concrete example
Elena used to repeat at every chat that she was an accountant writing for clients without tax expertise, and that she wanted simple answers. Tired of doing it, she turned on memory and told it once explicitly. She then opened the settings and checked: the note was there, written as she wanted. Since then every new conversation already starts with the right tone, and when a detail changed — a new type of client — she simply told it, updating the note.
When it does NOT work (and how to fix it)
If it doesn't remember something you thought it had understood
The automatic part of memory is opaque: it decides what to keep, and it doesn't always coincide with what you expect. The fix is not to hope it guesses: tell it explicitly "remember that...", so the information ends up among the certain notes instead of those it maybe picked up.
If it remembers wrong or now-outdated things
It happens that it drags along an old detail. Go into the memory settings, find the note and correct or delete it. It's material you control: treat it like a notebook to keep tidy, not like a black box.
If memory is off or unavailable
No panic: paste at the start of the chat a three- or four-line summary of who you are and what you need. It's a manual memory, but it works everywhere and doesn't depend on the tool's settings.
A tip from someone who really uses it
Treat memory like a notebook to reread from time to time: open it, check what it has noted, delete what's old or what you don't want kept. This periodic cleanup does two things: it keeps its notes accurate, so the advice stays relevant, and it protects you, because you don't let information about you you'd rather not keep pile up.
Frequently asked questions
Where do I see what the AI knows about me?
In the settings, in the section dedicated to memory or personalization. There you find the list of notes, you can edit them one by one or delete them in bulk. It's the first place to check if the answers seem to be based on something wrong.
Does it remember everything I've told it?
No: it remembers what it has noted, not the entire history of conversations word for word. That's why the explicit notes are more reliable: what you ask it to remember clearly has a much higher chance of staying than what it hopes to have caught on the fly.
Is memory private?
The notes are tied to your account and you control them, but they remain data on a server: treat them as such. Don't have it memorize passwords, sensitive data or other people's information, and clean up from time to time.
With memory, does the AI know me like a friend?
No, and it's the misunderstanding that leads to trusting it too much. Memory is a notebook of notes that the AI rereads, not an understanding of who you are: it doesn't understand you, it remembers data about you. It works much better if you're the one telling it explicitly what to keep, instead of expecting it to "know" you by guessing. It's a tool you govern, not a bond that grows.