Claude Memory (2026): How It Works, What It Stores, How to Turn It Off, and Why Your Team Plan Has It Off

Claude memory: the short answer

People turn on Claude memory and assume Claude now knows their business. It doesn't. It knows you. As of September 2026, Claude memory saves individual topics from your own chats as you talk, and it is on by default for Free, Pro, and Max on the web, Claude Desktop, and Claude Mobile. On Team and Enterprise plans it is off until an owner turns it on, and each person still switches it on for themselves. Everything lives at Settings > Memory, where you can read, edit, or delete each saved topic, pause memory, or reset it. Below: memory versus chat search, what Claude will and won't store, the plan-by-plan defaults, pause versus reset, and the part the launch posts skip: what belongs in memory and what belongs somewhere your whole team can see it. Every fact here was read off Anthropic's help center and blog on September 14, 2026.

Does Claude have memory?

Yes, and it's actually two features that get lumped together. Knowing which one you're talking about saves a lot of confused Slack threads.

• Chat search. Claude can search your previous conversations to build on earlier context. It searches all of your chats outside projects, and conversations inside a project only from within that same project. Anthropic says chat search is on by default once it has rolled out to your account. One exception: if an Enterprise organization uses customer-managed encryption keys, past chats can't be searched.
• Memory. Claude saves what it learns about you as individual topics while you chat, and uses them in later conversations. This is the part people mean when they say "Claude remembers me."

If you used Claude's older memory feature, Anthropic has moved everyone off that legacy experience. The option to export legacy memory ran until September 9, 2026, so that window has closed. The help center has a separate article, "Import and export your memory from Claude," if you're moving memory around.

How Claude memory works

Claude doesn't wait for a conversation to end and write a summary. It saves memory as a set of individual topics as you go. You can also just tell it "remember this," which is the fastest way to fix something it keeps getting wrong.

The part I like: it isn't a black box. Open Settings > Memory and look under Topics. Everything Claude remembers is there as a list of files, one topic each, and you can read, edit, or delete any of them. The files are short, so a correction pays off everywhere. Anthropic's own example is your company's old name. Fix it once and it stops showing up in every draft.

Projects get their own wall. Each project has a separate memory space and a dedicated project summary, kept apart from your other projects and from chats outside projects. So what Claude picks up inside your Acme proposal project stays in the Acme proposal project. How to set projects up so that wall works for you is on how to use Claude Projects.

Claude memory and Cowork

The memory you use in chat is the same memory Claude Cowork uses, with one catch that matters. Memory is shared between chat and Cowork when Cowork runs in the cloud. Cowork sessions that run locally on your computer don't use memory at all.

So if you told Claude in chat that your fiscal year starts in July, a cloud Cowork session can use that. A local session starts cold. If Cowork feels like it forgot everything, check where it's running before you blame the model. What Cowork is and when to use it is on what is Claude Cowork.

Claude memory by plan

Here is the default on each plan, as of September 2026:

The Team row is the one that confuses small businesses. You buy Team, people expect the Claude they use at home, and memory just isn't there. Nothing is broken. Nobody flipped the switch. Incognito chats, by the way, exist on every plan in that table. The full plan-by-plan picture is on Claude AI pricing.

Memory is personal. Your business context is not.

Here's where I'll pick the fight. Memory is built from each person's own chats, and each person turns it on for themselves. Your office manager's Claude learns what your office manager tells it. Your sales lead's Claude learns something else. Neither one knows your pricing unless somebody typed it in.

That isn't a flaw. It's a notepad, and a notepad is the wrong place to keep the company's facts. If two people on the same Team plan can get two different answers to "what do we charge for this?", you don't have an AI problem. You have a source-of-truth problem.

The split I use:

• Put in memory: things about you. Your role, how you like drafts formatted, the tone you want in your own emails, the clients you personally handle, the shorthand you use. If it would be fine for a coworker's Claude to not know it, it can live in memory.
• Put in a project: work that belongs to one job, client, or process. It gets its own memory and its own summary, walled off from everything else, so client A's details don't leak into client B's drafts.
• Put in a written business-context doc: offers, prices, policies, voice, how you sell. Written once, owned by a named person, updated the day a price changes, used by everyone. I run both of my businesses off a document like this. Memory is the notepad; the doc is the source of truth.

The rule of thumb: if it would be wrong for someone else on your team to get a different answer, it doesn't belong in anyone's memory. The rest of the team rollout is in the Claude for small business setup guide.

What Claude memory won't store

By default, Claude doesn't store sensitive topics: health, race, ethnicity, religious beliefs, politics, gender identity, and similar. You can opt in with a setting called "include sensitive topics in memory." If you do, you'll see a notice each time one is saved, and it isn't retroactive, so nothing from before gets pulled in.

Some things stay out even with that setting on. Claude won't store sensitive identification numbers like a Social Security number or government ID numbers, criminal history, immigration status, or anything that violates Anthropic's Acceptable Use Policy. When it can't save something, it tells you.

For a business owner, the practical read: memory is not a place for customer records. It will refuse the worst of it on its own, and the rest belongs in systems your company controls.

How to turn off Claude memory

The toggle is at Settings > Memory, and there are two very different ways to turn it off.

• Pause memory. Claude keeps what it already remembers but won't use it and won't make new memories. Conversations you have while paused are not added later if you turn memory back on.
• Reset memory. Permanently deletes all memories, including every project's memory. It cannot be undone.

Pause first. Reset is the nuclear option, and nobody should wipe every project's memory over one wrong topic. Go to Topics and delete the one file instead.

For a single conversation you don't want remembered, use an incognito chat: the ghost icon when you start a new chat outside a project. Incognito chats aren't saved to your chat history, and Claude won't pull from them when it searches. They're available on Free, Pro, Max, Team, and Enterprise. One thing Team and Enterprise users should know: incognito chats are still included in standard data exports and follow your organization's retention policies. Incognito means out of Claude's memory, not out of your company's records.

Should your team turn Claude memory on?

My take: yes, per person, with a habit attached. Memory is genuinely useful for the stuff that's about you, and the Topics list makes it auditable in a way most "AI remembers you" features aren't. The habit is a short review: every so often, each person opens their Topics list and deletes what's wrong or stale.

What I wouldn't do is turn it on and call that your company's AI setup. Memory makes each person's Claude a little better at being their assistant. It does nothing to make ten people's Claude agree on your prices, your policies, or how you talk to customers. That part is still a document and a project, and somebody owns it.

What to do today

Open Settings > Memory. Read your Topics list top to bottom. Find one file that's wrong or stale, an old price, a title you no longer have, a client you stopped working with, and fix it or delete it. That's five minutes, and it pays off in every chat after. If you're on Team and memory is off, that's an owner decision: ask whoever runs the workspace. And if you want Claude set up for the whole company, with your business context in one written place instead of scattered across everyone's memory, that's the Claude for small business install, from $4,500 in about two weeks; the $397 Playbook is the do-it-yourself version.