AI Assistant Memory and What It Really Remembers About You
You mention your daughter's allergy once while planning a birthday menu, and three weeks later the assistant brings it up in a recipe for a work lunch.

You mention your daughter's allergy once while planning a birthday menu, and three weeks later the assistant brings it up in a recipe for a work lunch. That small surprise is how most people discover that AI assistant memory exists, and it raises fair questions. What does it keep, where does it keep it, and how do you make it useful without handing over more than you meant to?
This guide explains the main kinds of memory in today's assistants, how to check what has been saved, and a simple routine for keeping memory helpful rather than unsettling.
What AI assistant memory actually means
The word "memory" covers three different things, and most confusion comes from mixing them up.
- The conversation window. Within a single chat, the assistant can see everything above your latest message, up to a limit. This is not stored as memory; it disappears from view when you start a new chat.
- Saved memories. Many assistants can now write short notes about you, such as "prefers metric units" or "runs a bakery in Leeds," and pull them into future chats. Some save these automatically; others only when you ask.
- Chat history reference. Some assistants can search or summarise your past conversations to answer a new question, even if nothing was saved as a formal note.
There is also a fourth layer people forget: custom instructions or a profile you write yourself. That is not memory in the automatic sense, but it gets added to every chat, so it behaves like memory you control fully.
Separately, your chats may be kept on the provider's servers for safety review or, depending on your settings, used to improve future models. That is a data retention question, not a memory feature, and it has its own switch.
How to check what your assistant remembers
Every major assistant that offers memory also offers a way to view it. The exact menu names change often, but the path is usually similar: open Settings, look for a section called Personalisation, Memory or Saved info, and open the list.
When you open it for the first time, set aside ten minutes. A few months of normal use can easily produce somewhere between 15 and 60 saved items, many of which you will not remember sharing. Typical entries look like this:
| Type of memory | Example | Usually worth keeping? |
|---|---|---|
| Preferences | Writes in British English, likes short answers | Yes |
| Work context | Manages a team of six in customer support | Yes, if still true |
| One-off tasks | Planning a trip to Lisbon in May | No, once the trip is over |
| Health or family details | Child has a nut allergy | Only if you chose to share it |
| Wrong guesses | Lives in Toronto (you visited once) | Delete |
You can delete individual items, clear everything, or switch memory off. You can also ask in a chat, "What do you remember about me?" The answer is a useful starting point, but treat the settings page as the source of truth, because the assistant's spoken summary can miss items.
A simple routine for keeping memory useful
Memory is most valuable when it holds stable facts that save you typing. It becomes a problem when it fills with stale details or things you never meant to share. A short routine keeps it in shape.
- Decide your baseline. Write three to five facts you want every chat to know: your role, your preferred tone, your units, any tools you use daily. Put these in custom instructions rather than relying on automatic memory.
- Review monthly. Put a 10-minute calendar reminder on the first of each month. Delete finished projects, wrong guesses and anything you feel uneasy about.
- Use temporary chats for sensitive topics. Most assistants offer a mode that skips memory and history. Use it for medical questions, money worries or anything involving another person's private details.
- Correct it out loud. If an answer leans on a wrong memory, say "That's out of date, please forget it." Then check the settings list to confirm it was removed.
The same thinking applies to what you type in the first place. Our guide on what not to paste into a chatbot covers the kinds of details that should never reach a saved note at all.
Why switching memory off is not always the safer choice
The common advice in privacy circles is simple: turn memory off and be done with it. That advice is not wrong, but it misses how people actually behave.
When memory is off, most people do not stop sharing context. They paste the same background into chat after chat, often in longer and less careful form. A support manager who retypes "I manage six people, two are new, one is on a performance plan" every week has shared more about a colleague than a single, carefully worded saved note would.
A curated memory, reviewed monthly and limited to facts you chose, can actually reduce how much personal detail flows to the assistant. The risk is not memory itself. The risk is unreviewed memory that quietly collects whatever you happened to mention.
So the better rule is this: switch memory off if you will never review it. Keep it on, and trim it, if you will. Either way, check the separate setting that controls whether your chats help train future models, because that is where your data goes beyond your own account.
Getting better answers from memory you trust
Once your memory holds a clean set of facts, you can lean on it. Instead of opening with a paragraph of background, you can write "Draft the weekly update in my usual style" and get something close on the first try. That saving adds up. If you write ten prompts a day and each one skips 40 words of context, that is around 2,000 words a week you no longer type.
Memory works best alongside good prompts, not instead of them. For one-off tasks, the context still belongs in the chat itself, and our piece on giving an AI assistant context it can actually use shows how to structure it. You can find more practical guides in our AI assistants section.
One last tip: if you share a device or account, memory is shared too. A family laptop logged into one assistant account will mix everyone's preferences, and the results get strange quickly. Separate accounts, or at least separate browser profiles, keep each person's memory their own.
Frequently asked questions
Can an AI assistant remember previous conversations?
Many can, if memory or chat history reference is switched on. Within a single chat, the assistant sees earlier messages automatically. Across chats, it relies on saved memory notes or a feature that searches your history, both of which you can view and switch off in settings.
How do I make an AI assistant forget something?
Open the memory or personalisation section in settings and delete the specific item. You can also ask the assistant to forget it in a chat, but confirm in settings afterwards. Deleting a whole conversation may not remove a memory that was already saved from it.
Is AI assistant memory used to train the model?
Memory and training are usually controlled by separate settings. Memory personalises your own account, while the training setting decides whether your chats may help improve future models. Check both, since switching one off does not automatically switch off the other.
More in AI Assistants
AI Assistants
How to Get an AI Assistant to Cite Sources You Can Check
I asked an assistant for the Swedish rules on invoice retention periods and it gave me a clean answer with a reference: a named...
AI Assistants
AI Assistant Custom Instructions That Are Worth Writing
Most people start every chat with an AI assistant by retyping the same three facts.
AI Assistants
AI Assistant for Spreadsheet Formulas That Actually Work
You know what the spreadsheet should do, you just cannot remember whether it is XLOOKUP or INDEX MATCH, and the forum answer...