A marketer no longer has to tell ChatGPT in every conversation that she writes in British English, for a business audience, with no exclamation marks. The tool remembers. That is AI memory at its best: less repeating yourself, useful answers sooner.
The same feature can also cut the other way. A support lead asks ChatGPT in January to rewrite a difficult complaint. The prompt names the customer, Aoife Walsh, and mentions an open dispute. Months later the same person asks something unrelated, about an internal process. ChatGPT casually refers to "a case like the Aoife Walsh one". The name was never typed again.
This is not a bug and not a breach. It is the same memory, doing what it was built to do. So the question is not whether to use it, but what you let it remember.
What does "Memory updated" mean?
The short notice under an answer tells you the tool has just saved something from this conversation to use in later ones. OpenAI added it to ChatGPT in September 2024, so people could see when a memory was created and go straight to the list of what ChatGPT had picked up. Microsoft 365 Copilot Chat shows a notice with the same wording when it saves something to its saved memories. (Checked 22 October 2026.)
When you see it at work, take ten seconds. Was the thing it saved about you (your role, your writing style) or about someone else (a client, a colleague, a case)? The first is what memory is for. The second is worth deleting straight away, using the steps per tool further down.
Not everything a tool remembers comes with a notice. With the newer version of ChatGPT memory, it keeps a broader summary up to date from your chats, and OpenAI says that summary does not necessarily show every detail it can draw on (checked 22 October 2026).
What is AI memory, actually?
Until recently, every AI conversation was a clean slate. That has changed. The major AI tools now carry context from one conversation into the next, in two ways that work independently.
The first is saved memories: facts you ask the tool to keep, or that it decides are worth keeping. "Remember that I work in support" or "I always write in British English." These are discrete notes you can usually read back as a list.
The second is referencing your past chats: the tool looks back at what you typed before and draws preferences and context from it, without you asking. This is the part most people miss. You do not have to ask the tool to remember anything for it to remember things anyway.
Depending on plan and region, ChatGPT can also draw on files in your Library and on connected apps such as Gmail, according to OpenAI's help page. The context derived from all of this is fuzzier than a list: not a single note, but a pattern pulled from your chat history and other sources.
What gets remembered, and when is that awkward at work?
For personal use, memory is mostly convenient. The tool remembers you are vegetarian or that you prefer short answers. At work, the balance shifts.
The data you put in a work prompt is exactly the data you would rather not see resurface months later. A client name in a rewrite request. A project number in a summary. A line about a sick-leave request in a draft letter. A supplier name and an overdue payment in an email you asked it to draft.
When the tool remembers that context, three problems follow. Context bleeds from one conversation into another, so sensitive information from an old case can surface in an unrelated task. Identifiability lasts longer than the employee expects, because the original chat is long forgotten while the derived memory is not. And nobody decides on purpose to keep that client name; it just happens.
Under the GDPR, a prompt containing personal data is a processing operation. If that same name comes back through memory in a different conversation, it is no longer an isolated event but processing that runs for longer than the organisation assessed. The question "what data is in our AI tools?" gets harder to answer as memory ages.
How do you make memory work for you?
Memory works best for things about you and how you work, not about the people you work for.
- Let it keep preferences, not case files. Your role, your writing style, the language and the format you like are ideal things for it to remember. Client names, case details and HR information are not.
- Use a temporary chat for case work. ChatGPT has temporary chats and Claude has incognito chats, and neither adds to your memory. Handy for a task involving client or employee data.
- Check what is stored now and then. Running through the memory screen once a quarter takes two minutes and keeps it useful.
How do you view, switch off and delete memory per tool?
Each tool has its own screen, and they do not behave the same way when you switch memory off. These are the paths from each supplier's own help pages, checked on 22 October 2026. Menus vary by plan, region and app version.
ChatGPT
Go to Settings > Personalization > Memory. Depending on your account you will see a Memory summary, a list of Saved memories, or both switches Reference saved memories and Reference chat history. OpenAI calls the newer version improved memory: it keeps a broader summary up to date. Where the option is available, you can switch back to the older experience with a list of separate memories.
In the memory summary you can type a correction, choose Don't mention this again (ChatGPT brings it up less often, but the source is not deleted), or choose Delete and turn off memory from the ••• menu. That last option deletes what the summary shows and switches memory off; your past chats stay.
Three things surprise people, all from OpenAI's help page on memory:
- Deleting a chat does not automatically delete a saved memory created from it. Delete the memory itself as well.
- Turning off Reference chat history schedules what ChatGPT learned from past chats for deletion within 30 days, but your saved memories stay until you delete them.
- OpenAI may keep logs of deleted saved memories for up to 30 days for safety and debugging.
For a sensitive one-off task, start a temporary chat. It does not create or update memories, and if you choose Unpersonalized before you start, it does not use your existing memories either. If you later save a temporary chat, it becomes a regular chat that follows your memory settings again.
Google Gemini
Gemini has saved info and can look at your past chats before it answers. Go to Settings & help > Personal context on gemini.google.com. There you can add, edit and delete what Gemini keeps about you, and switch the setting for info you asked it to save on or off (Google's help page). Google says it is rolling out the past-chats feature to the mobile app gradually.
The chats themselves live in your Gemini Apps activity, under Settings & help > Activity. By default Google deletes activity older than 18 months; you can set that to 3 or 36 months, or switch it off (Google). Temporary chats, and chats you have while Keep Activity is off, are still kept with your account for 72 hours, and Google says temporary chats are not used to train its models (Gemini Apps Privacy Hub).
Microsoft Copilot
In the consumer app on copilot.com, Windows and macOS, select your profile icon, then your profile name, then Memory > Personalization and memory. Microsoft says that if you switch this off, Copilot forgets its memories of your conversations; your conversation history stays. Ask "What do you know about me?" to see what it holds, tell it to forget something specific, or use Memory > Delete all Memory (Microsoft).
The work version, Microsoft 365 Copilot Chat, behaves differently. Saving memories is on by default, and Copilot asks whether you want to keep something it thinks is important. To manage it, go to ... Settings and more > Chat settings > Personalization > Saved memories > Manage saved memories. Turning the switch off does not delete what is already saved; delete memories one by one or with Delete all memories (Microsoft).
Claude
Go to Settings > Memory. Everything Claude remembers is listed under Topics: open a topic to read it, edit it, or delete it. Memory is on by default on the Free, Pro and Max plans. On Team and Enterprise, the owners decide whether memory is available, and it stays off for each member until they turn it on (Anthropic).
To stop using memory, you choose between Pause memory (Claude keeps what it has but stops using it and saves nothing new) and Reset memory (all memories, including project memories, are deleted permanently). For a single chat, turn off Memory in the "+" menu before you send the first message, or open an incognito chat via the ghost icon. As with ChatGPT, deleting a chat does not remove the memory entries made from it.
One detail that matters at work: Anthropic says Claude never saves government ID numbers, criminal history, financial account numbers or immigration status to memory, even if you ask. That covers memory, not the conversation itself.
On a work account, the organisation also has a say in what may be remembered. Our article on who decides what an AI assistant remembers covers that side.
What does managing memory not solve?
Viewing, limiting and clearing memory is sensible. But it only works after the fact. It does not decide whether the client name, the account number or the project code should have been in the prompt at all.
A switch set to off does not always erase a memory that is already there. The memory screen cannot see that the name in your next prompt is sensitive. A temporary chat stops memory from forming, but it does not stop anyone typing and sending a confidential line. And someone in the middle of a task is thinking about the task, not about the tool's settings.
What remains is the moment of typing. The few seconds when the sensitive data is still on your screen and not yet sent.
How do you keep sensitive context out of memory?
What never enters the prompt cannot be remembered. That is where BeeSensible helps: sensitive details are highlighted while you type, including in AI tools such as ChatGPT, Gemini, Copilot and Claude. By default you choose what happens next: mask it, which swaps the value for a placeholder such as [NAME1], or remove it. The tool can still help with the complaint, without the customer's name ending up in its memory. The checked text is not stored.
Outlook
