Claude Memory Explained: What Does It Actually Remember?
For a long time, the standard complaint about Claude was that it woke up with amnesia every morning — a brilliant working session on Tuesday, a total stranger on Wednesday. That era is over. Claude now builds real memory entries about you as you chat, organizes them by category, and carries them into future conversations automatically. Here's exactly what it keeps, what it deliberately leaves out, and how to stay in control of it.
Every memory entry Claude builds is visible, editable, and deletable from a single settings screen.
What Memory Actually Is
Claude's memory feature automatically learns about you across conversations — your role, preferences, ongoing projects, and feedback — and carries that forward without you re-explaining it each time. It builds individual memory entries in real time as you chat, organizes them into categories, and surfaces them in future conversations when relevant. You don't need to explicitly ask Claude to remember something for it to notice and store it, though you can also just tell it directly: "remember that I prefer bullet points" works as a direct instruction, and Claude updates memory immediately when you phrase it that way.
What memory is not: a transcript. It doesn't save the entirety of your messages or function as a searchable log of every conversation you've had. It stores distilled facts and preferences, not full message history. A separate feature, chat search, handles retrieving specific past conversations when you ask for one directly, and that feature is generally gated to paid plans rather than bundled with memory itself.
From Daily Batches to Real-Time Entries
A conversation from this afternoon can now inform a new chat later the same day.
Editing an inaccurate memory entry takes a few seconds from the Memory settings page.
Claude's automatic memory used to run on a roughly 24-hour cycle: it reviewed recent conversations once a day in the background and folded whatever mattered into its ongoing understanding of you, while explicit "remember that" commands updated in real time as the exception to that rule. As of a July 10, 2026 update, that daily batch process was removed. Memory is now built from individual entries Claude reads and updates while you're actually talking to it, not reviewed in bulk overnight — a conversation from earlier the same day is already available to a new chat that afternoon, rather than waiting until the next day's review cycle catches up.
The old delay existed for a reason: it stopped Claude from latching onto a single throwaway comment or one-off request and treating it as a permanent fact about you. The real-time version handles that differently, weighing entries for relevance rather than treating everything stored as equally permanent, which is also why memory and Claude's separate Preferences setting aren't the same system — Preferences is where "always" language forces something to apply in every conversation, while Memory is judged for relevance each time rather than applied unconditionally.
What Memory Deliberately Leaves Out
As of an August 25, 2026 update, Anthropic excludes a specific list of categories from memory by default: health conditions, race, ethnicity, religious beliefs, political views, and gender identity. There's a toggle if you specifically want Claude to remember these anyway, but it stays off until you turn it on yourself — the default is exclusion, not inclusion with an opt-out. Beyond that default list, Claude is also designed to never memorize credentials, passwords, API keys, or bank card numbers under any configuration.
| Setting | What It Controls |
|---|---|
| Settings > Memory | View every stored entry by category, edit or delete individual ones, or reset everything at once |
| Sensitive categories toggle | Off by default; opting in allows health, race, religion, and political entries specifically |
| Incognito mode | Prevents new entries from being created or updated, but existing memory still applies during that session |
| Per-account scoping | Memories are tied to your account only and are never shared with other users |
| Admin controls (Team/Enterprise) | Organization admins can disable memory entirely but cannot view what's stored in individual accounts |
There's also an import path worth knowing about if you're switching from another AI assistant: Anthropic has offered a tool that transfers existing memory from ChatGPT, Gemini, or Grok into Claude's own memory system, rather than making you rebuild months of accumulated context from scratch. It's opt-in and reviewable before anything gets committed, consistent with how the rest of memory is designed to work — visible and editable rather than a silent background process.
General Memory vs Project Memory vs Chat Search
- Covers conversations outside of any Project
- Builds entries automatically from what you say across regular chats
- Available on Free, Pro, and Max plans in the current rollout
- Isolated to that specific Project — doesn't draw on general account memory
- Compounds the more that specific Project is used
- Prioritizes the Project's uploaded document context over general memorized preferences
The distinction matters in practice: a preference you've established in your general chats won't automatically show up inside a Project, and vice versa. If you want a Project to consistently apply something Claude already knows about you from regular conversations, it's worth stating it explicitly inside that Project once rather than assuming it carries over. Team and Enterprise accounts, as of mid-2026, were reported to still be running on the older summary-based memory system while the newer entry-based experience continued rolling out more broadly — worth checking directly in your account settings if you're on one of those plans and the behavior described here doesn't match what you're seeing.
Using Memory Well
Give Explicit Feedback
When Claude does something well, say so directly — "remember, I liked that format" reinforces it faster than hoping it's inferred.
Correct Mistakes the Same Way
If a stored entry is wrong, either edit it directly in Settings or tell Claude to forget it in conversation.
Separate Contexts If Needed
Some people keep separate accounts for work and personal use specifically to keep memory scoped cleanly.
Check In Periodically
A quick look at Settings > Memory every so often catches anything inaccurate before it shapes too many future replies.
A Real Change, Not a Gimmick
Worth turning on, worth checking occasionally, worth trusting with the ordinary stuff.
The gap between "explaining your job and preferences at the start of every chat" and "Claude already knowing them" is a genuinely different day-to-day experience, and it's rolled out broadly enough now that older guides claiming Claude has no persistent memory are simply out of date. The privacy defaults — sensitive categories off unless you opt in, full visibility and editability of every entry, a one-click reset — are reasonable enough that most people don't need to think hard before turning it on. The one habit worth building is treating it like a helpful assistant's notes rather than a secure vault: useful, transparent, but not where anything genuinely sensitive should end up by default.
Frequently Asked Questions
Does Claude remember my entire conversation history?
No. Memory doesn't save the entirety of your messages or function as a log of every conversation. It stores individual entries about you — your role, preferences, ongoing projects, feedback — organized by category, not full transcripts. A separate chat search feature, available on paid plans, can retrieve specific past conversations.
Can I see and delete what Claude remembers about me?
Yes. Go to Settings, then Memory, and every entry is listed by category. You can edit an inaccurate entry, delete a single one, or reset memory entirely to clear everything in one click. You can also just tell Claude what to remember or forget during a normal conversation.
Does memory work differently inside a Claude Project?
Yes. Each Project keeps its own separate, isolated memory space rather than drawing on your general account memory. Global memory only covers conversations outside of Projects, so a fact you want a specific Project to know may need to be established inside that Project directly.
More AI Tool Guides, Written Clearly
Browse our full library of AI tool comparisons, updated as the category evolves.
More from the creator
Explore two other projects from the same creator.