Claude has been writing notes about you. Here is the list.
The short version: Claude keeps a memory of you, and it is not a black box. It is a list of files at Settings › Memory › Topics, one per subject, and you can open, edit or delete any of them. As of 25 August 2026 it writes those notes while you are chatting rather than summarising afterwards, and the same memory now follows you into Claude Cowork. It leaves health, race, ethnicity, religion, politics and gender identity alone unless you switch that on, and there are four things it will not store at all. The part the announcement does not mention: deleting a conversation does not delete the memory made from it.
What this actually is, in plain terms
Claude is a chatbot made by a company called Anthropic. Like most chatbots now, it has a memory: as you talk to it, it decides certain things about you are worth keeping, and writes them down. Next time you open it, it already knows them. You do not have to reintroduce yourself.
The thing almost nobody realises is that those notes are not hidden and they are not written in code. They are a list of files, in plain English, that you can open. Anthropic calls each one a topic. One might be about how you like your answers written. One might be about where you work. One might be about a trip you were planning three weeks ago.
On 25 August 2026 Anthropic changed two things about it. First, the memory you have in normal chat is now the same memory used by Claude Cowork, the version of Claude that goes off and does longer jobs for you. Their words: wherever you work with Claude, it starts from what it already knows about you. Second, and more interesting, Claude now adds topics to memory as you chat, instead of summarising conversations after they end. The note-taking is happening live, in the middle of the conversation, not once you have closed the tab.
Where the list lives
Three taps. This is the whole thing, with nothing left out:
Open Settings
In Claude on web, desktop or mobile, open Settings. On mobile it is behind your profile; on web it is in the menu with your name on it.
Go to Memory
Inside Settings there is a Memory section. This is where the master switch lives too — it is labelled Generate memory from chats.
Open Topics
Under Memory you will find Topics. That is the list. Select any topic to read it, use the edit icon to change it, or choose Delete to remove it. When you fix something in one topic, the change applies to every conversation from then on.
There is a second way in, and it is easier to remember: just tell Claude in the chat. Say what you would like it to remember, change or forget, and the update applies to your next conversation. You do not have to go hunting through settings to correct one fact.
Anthropic’s own example: your company changes its name, and Claude keeps using the old one. Correct that in the one file, and every conversation from then on gets it right. Delete it instead and you are back to explaining yourself every time. The list is not just a privacy screen — it is the place you go to stop repeating yourself.
What to check first
You are about to open a list you have never seen. Here is the order to read it in, so you find the things that matter rather than scrolling for ten minutes.
- Anything with a name in it. Your employer, your school, your clients, people you live with. Names are the thing that goes stale fastest and the thing Claude will repeat back most confidently. This is the category Anthropic’s own example is about.
- Anything you only said once, in passing. Memory writes as you chat now, so an aside in the middle of a long conversation can become a topic. If something in there surprises you, that is usually where it came from.
- Anything that was true then and is not now. Old job, old project, old plan, old preference. Edit rather than delete, so the correction sticks.
- Whether “include sensitive topics” is on or off. It is off by default. If it is on, someone turned it on. See the next section for what that switch actually controls.
- Your projects, separately. Each project has its own memory space and its own summary, kept apart from other projects and from your ordinary chats. Checking your main memory does not tell you what a project remembers.
The things it leaves alone unless you say otherwise
By default, Claude does not store topics about personal or sensitive subject matter. Anthropic names these explicitly:
- your health
- race
- ethnicity
- religious beliefs
- politics
- gender identity
- “and other similar areas”
There is a setting called include sensitive topics in memory that turns that off. The reason you might want it on is genuinely practical: their example is Claude remembering a gluten allergy when you ask it for recipes. If you switch it on, Claude notifies you each time it saves something in one of those categories, it only applies going forward rather than retroactively combing your history, and you can switch it back off whenever you like.
The four things it will not write down at all
Even with sensitive topics switched on, four categories are refused outright:
- sensitive identification numbers — an SSN, government ID numbers and the like
- criminal history
- immigration status
- anything that violates Anthropic’s Acceptable Use Policy
And it does not fail silently: Claude tells you when it is unable to update memory to include any of that.
Deleting the chat does not delete the note
This is the trap, and it catches almost everyone, because deleting a conversation feels like undoing it. It is not. Here is Anthropic’s own help centre, word for word:
When a conversation expires or is deleted, related memory entries generated from it won’t be removed, but you can delete individual memories at any time.
Read that twice. The note Claude made about you outlives the conversation it came from. You can wipe your entire chat history and the memory built out of it is still sitting in the Topics list, still shaping every answer you get.
So deleting is not forgetting. Deleting a chat is housekeeping; removing what Claude knows is a separate job, done in the Topics list. If this feels familiar, it should — ChatGPT works the same way, and we wrote up what ChatGPT knows about you and how to delete it for exactly this reason. Two different companies, same architecture, same wrong assumption.
Use incognito before, instead of deleting after. Deleting removes a note that already exists. Incognito stops the note from ever being written. If there is something you would rather Claude did not keep, the moment to act is before you type it, not after you have read it back.
How to wipe it properly
There are four different tools here and they do genuinely different things. Picking the wrong one is how people end up thinking they have cleared something they have not.
| What you do | What it does |
|---|---|
| Delete one topic | Removes that single memory entry. The surgical option, and the only one that removes a specific thing without losing everything else. |
| Pause memory | Keeps everything Claude already knows, stops it writing anything new. Reversible. |
| Reset memory | Permanently deletes all memories including project memories. Anthropic’s own warning: once you select this and click Reset memory, this cannot be undone. |
| Incognito chat | Prevention, not cleanup. The ghost icon in the top right of a new chat, outside projects. Nothing is saved to your history and nothing goes into memory. |
Two footnotes worth having. Incognito chats are still retained for at least 30 days for safety purposes, so incognito keeps a conversation out of Claude’s memory rather than making it disappear from the world. And because every project has its own memory space, a proper clear-out means checking your projects too, not just the main list.
Whether this is switched on for you right now
Probably, and that is the bit worth knowing. Memory is on by default on Free, Pro and Max, across web, desktop and mobile. On iOS and Android you need the latest version of the app. Nobody has to opt in for it to start.
Team and Enterprise work the opposite way. Admins control whether memory is available for the organisation at all, and even then it stays off for each individual until that person turns it on. If you are on a work account and the Memory screen looks empty or missing, that is why.
One last distinction, because the two get confused constantly. Chat search is Claude looking back through your old conversations when you ask it to; it is a paid-plan feature. Memory is Claude keeping a running set of notes about you without being asked. Turning one off does not touch the other. If you want more on getting the most out of a Claude session without burning through it, we covered that in six habits that cut Claude Code token usage.
Questions people actually ask
Where do I actually find what Claude remembers about me?
Settings, then Memory, then Topics. Everything Claude has decided is worth keeping about you is there as a list of files, one per subject, and you can open any of them and read it in plain English. It is not a hidden log and there is no prompt you have to know. It is a screen, and it takes about ten seconds to get to.
Can I edit what Claude remembers, or only delete it?
Both. Open a topic and use the edit icon to change it, or choose Delete to remove it. The change applies to every conversation from then on, which is the useful part: Anthropic's own example is a company that changed its name, where correcting that one file means every later chat gets the name right instead of you repeating the correction forever. You can also just tell Claude in a chat what you want it to remember, change or forget, and the update lands on your next conversation.
If I delete the conversation, does the memory go with it?
No, and this is the single most misunderstood part. Anthropic's help centre says it plainly: when a conversation expires or is deleted, related memory entries generated from it will not be removed. The note Claude made outlives the chat it came from. To remove it you have to delete the memory entry itself, in the Topics list.
What does Claude refuse to remember?
By default it does not store topics related to personal or sensitive subject matter, including health, race, ethnicity, religious beliefs, politics and gender identity. You can turn those on with the include sensitive topics setting. Four things stay off the table even then: sensitive identification numbers such as an SSN or government ID, criminal history, immigration status, and anything that violates the Acceptable Use Policy. Claude tells you when it cannot add something.
Is Claude's memory on for me right now?
Almost certainly yes if you are on a personal plan. Memory is on by default on Free, Pro and Max, across web, desktop and mobile. Team and Enterprise are the other way round: admins control whether it is available at all, and it stays off for each individual until they turn it on.
What is an incognito chat and when should I use one?
It is a chat that is not saved to your history and is not written into memory. You start one with the ghost icon in the top right of a new chat, outside projects. The timing matters more than anything else here: incognito prevents the note from ever being written, while deleting only removes the note after it exists. If there is something you would rather Claude did not write down, the moment to act is before you type it. Incognito chats are still retained for at least 30 days for safety, so it keeps things out of memory rather than making them vanish.
What is the difference between pausing memory and resetting it?
Pause keeps everything Claude already knows and stops it adding anything new. Reset permanently deletes all memories, including project memories, and Anthropic's own warning says that once you click Reset memory this cannot be undone. Pause is the reversible one. Reset is the door that only opens one way.
Does my project have its own separate memory?
Yes. Each project has its own separate memory space and its own project summary, kept apart from other projects and from your non-project chats. So there is not one list to check, there are several: your main memory, plus one per project. A reset wipes the lot.
What changed in the 25 August 2026 announcement?
Two things. Memory is now shared between chat and Claude Cowork, so work done in one starts from what the other already knows. And Claude now adds topics to memory as you chat, rather than summarising a conversation once it has ended, which means the writing is happening live rather than afterwards.
Want this built for you?
We write these memos because we build this stuff every day. If you want it working in your business instead of sitting on your reading list, that is literally our job.