The Claude Memory Feature: What It Remembers and How to Edit It
8 min read
The Claude memory feature saves facts from your chats as short topics and brings them into later conversations, on every plan including Free. This lesson shows where to read what it stored, how to correct it, and what to keep out.
Memory is switched on by default on Free, Pro and Max accounts. So if you've used Claude for a month, it holds a file on your business that you haven't read yet.
Where it lives
Open Settings > Memory. What Claude remembers is listed under Topics, in plain sentences. You can edit an entry, delete it, or switch memory off.
Claude adds to these topics while you chat. Mention twice that you close on Mondays and it will likely note that. Mention a one-off complaint about a supplier and it may note that too, which is the reason to look.
Memory inside Projects
Each Project keeps a separate memory. What Claude learns in your quotes Project stays there and doesn't show up in customer replies. Chats outside any Project feed your general memory.
The separation is useful and occasionally surprising. If you told Claude your new hourly rate in a plain chat, your quotes Project may not have heard. Stable facts belong in a knowledge file, where you can see them and date them.
Four places a fact can live
| The fact | Best home | Why |
|---|---|---|
| "I own a bakery in Guelph, use Canadian spelling" | Instructions for Claude, in Settings | True in every chat |
| "Refunds within 14 days with a receipt" | A Project knowledge file | A rule you'll update and want to see |
| "For quotes, list exclusions before terms" | Project instructions | A standing rule for one kind of work |
| "Likes bullet summaries, plans the week on Fridays" | Memory | Small preferences Claude can pick up as it goes |
Memory suits habits and preferences. Anything you'd be upset to see misremembered, a price, a policy, a legal commitment, goes where you can read it.
What to keep out of memory
Customer names and personal details shouldn't sit in a standing memory. If Claude noted "client Maria Santos is behind on payments", delete the entry. It was useful for one chat and has no business following you around.
For a conversation you'd prefer left no trace, use an incognito chat. Claude doesn't save it to memory or to your history, and it isn't used for training.
Moving from ChatGPT
If you built up context in another assistant, you don't have to start over. Claude's memory settings include an import that brings saved memory across from other AI services, and an export for going the other way. After importing, read every line before you rely on it.
On a Team plan
Memory behaves differently in a business workspace. It's off for each member until they turn it on, and the owner controls whether it's available at all. If you move your shop to Team, tell people to switch it on or they'll wonder why Claude turned forgetful.
Searching past conversations ("what did we decide about the spring promo?") is a separate feature. Anthropic lists chat search on paid plans only.
Where memory goes wrong
Memory is a summary Claude wrote, and summaries drift. It may hold your old address, a product you dropped, or a preference you mentioned once in a bad mood. It also can't tell a firm fact from a passing remark.
Two habits keep it honest. Read the Topics list on the first Monday of each month, alongside your Project files. And when Claude states something about your business that you didn't supply in the chat, ask where it got it. The answer is usually memory. Now and then it's a guess.
Try it now
Open Settings > Memory and read every topic. Delete one thing that's wrong or too personal, and move one important fact into a Project file or your account instructions. Five minutes, and you'll see what your assistant believes about you.