UNDERSTAND AI
Does it remember you?
Short answer: it depends, and the setting that decides it is one most people don’t know exists. By default, a lot of chatbots are “stateless” — each new conversation starts from a blank page, and the only thing the AI can “remember” is whatever is still sitting in the chat you’re in right now. But many products now also have a separate “memory” feature that carries facts about you from one chat to the next — and it’s often switched on by default. This page explains the difference, why memory is genuinely useful, why it’s the one dial where the upside and the risk are the same feature, and where to find the switch so the choice is yours.
Starting from blank: “stateless”
“Stateless” is the plain idea that the system keeps no state — no record of you — between conversations. Open a brand-new chat with a stateless system and it knows nothing about you. Not your name, not yesterday’s conversation, not the thing you told it last week. It begins from zero every single time.
Inside one conversation, it still seems to “remember” — you can refer back to something from a few messages ago and it follows. That’s not memory of you, though. It’s just that the whole conversation so far is being handed back to the model with each new message, so the earlier parts are still in front of it. Close the chat and start a fresh one, and that’s gone too. A stateless system has the memory of the current conversation, and nothing more.
Carrying things across chats: “memory”
“Memory” is a separate feature layered on top. When it’s on, the system can save facts about you — your name, your job, your preferences, things you’ve told it — and pull them into future conversations, even brand-new ones. That’s the difference: a stateless chat forgets you the moment it ends; a memory-enabled system builds up a picture of you and keeps it.
This is a real, useful thing. It’s why an assistant can stop asking the same questions, pick up where you left off, and feel less like a stranger every time. For a lot of people that’s the whole appeal — it gets more helpful, more personal, more genuinely yours the longer you use it. None of that is a trick. It’s the feature working exactly as designed.
The one dial that runs both ways
The benefit and the risk of memory aren’t two separate things you can have one without the other. They’re the same feature pointed in two directions.
Memory compounds for you: every useful fact it keeps makes the next conversation better, more personal, more tailored. And by the very same mechanism, it can compound on you. The picture it keeps is built from what you said and how the system described you back — and that picture can quietly tilt. If it once described you as brilliant, or fragile, or always right, that impression can get saved and carried forward, coloring how it treats you in every chat after — often without you ever seeing the note it wrote. A relationship that runs for months can slowly drift: the system keeps agreeing with a version of you it assembled, instead of meeting the actual you in front of it today.
Same dial. One direction makes it more useful; the other can make it a flattering or one-sided mirror you forgot you were looking into. Knowing the dial is there is what lets you enjoy the first without sleepwalking into the second.
Go deeper: the technical version
Under the hood, a language model is itself stateless — it has no built-in memory of anything between requests. What feels like memory within a chat is the “context window”: the running transcript is re-sent to the model on every turn, so prior messages are visible only because they’re physically included in the prompt again. (More on that limit on the context window page.) Cross-chat “memory” is a layer the product adds around the model: notable facts get extracted and saved to a store, then automatically inserted into the prompt of later conversations. So “the AI remembered me” is really “the application retrieved saved notes about me and pasted them in before the model ran.” That distinction matters, because it means the memory is typically editable and deletable by you — in most products it’s a list of notes you can read and remove, not something baked irreversibly into the model.
It’s probably already on
Many of the major chatbots now turn memory on by default for a lot of users. It’s what makes the product feel good out of the box — but it does mean a lot of people are building up a stored picture of themselves they’ve never looked at. You can leave it on, turn it off, or just check it now and then. What matters is knowing the choice exists.
Where to find the dial
The exact wording changes by product, and changes again whenever they redesign things, so here’s the durable version rather than a menu path that’ll be stale by next month. Look in the app’s Settings, under a section usually called something like Personalization, Memory, or Data controls. There you can typically do three things:
- Turn memory on or off. Off means it goes back to starting from blank each time — more private across chats, less personal.
- See what it has stored. Most products let you read the actual list of saved notes about you. This is the one worth doing at least once.
- Edit or delete entries. You can usually remove any single note, or wipe all of them, without deleting your account.
If you read that stored list and find a note describing who you are — not “prefers short answers” but something like “is exceptionally gifted” or “needs gentle handling” — that’s the kind of entry worth a second look. A preference is fine. An identity, stored as a fact and carried into every future chat, is the thing that drifts.
The one-line version: by default, many AIs start from blank every time — but most now have a “memory” feature, often switched on, that carries a picture of you across chats. The same feature that makes it more useful can also quietly build and keep a slanted picture of you. The dial lives in Settings, under Personalization or Memory. Look at what it’s stored, at least once.
Where to go next
See what it has stored
The memory check — a copy-and-paste prompt that makes your AI show you exactly what it’s saved about you.
Check your AI →When memory drifts
The deep end: what a long relationship drifting on its own stored picture looks like, and the research on it.
The research →What one chat can hold
The other kind of “memory” — the context window, and why even one long conversation has a limit.
The context window →Choosing for a kid
How to think about memory — and the rest of it — when the person using the AI is your teenager.
My teen & AI →Spot something here that’s out of date or could be clearer? Tell us — the products change their menus often, and this page is meant to stay accurate.