You ask Claude something it should already know about you, and you want to see where the answer came from. Good news: you can. Claude memory runs on two layers, an editable summary of standing facts about you and an automatic synthesis of recent chats that rebuilds every 24 hours, and you can open both, read them line by line, and edit them directly.
What sets Claude apart is the moment of retrieval. When it reaches into your past chats, it runs a search you watch happen as a tool call inside the conversation, in real time, before the answer lands. This guide explains how the two layers work, where the controls live, and the one privacy setting worth checking before you rely on any of it.
Does Claude remember past conversations?
Yes, but not by default. Claude starts each chat blank and pulls in past context only when it would help your current request. When it does, it runs a chat search you can see as a tool call inside the conversation. Anthropic's help center describes the feature as letting you ask Claude to find previous discussions, with the search reflected in your current chat as a tool call.
The timing matters. ChatGPT and Gemini now both surface memory after the answer, as source labels you can inspect. Claude surfaces it before, as a live search step you watch run. You are not told later that history was used. You see it being used.
If Claude used your past, you see it search. No search step in the chat means nothing from your history touched the answer. It came from the current chat plus your memory summary, and nothing else.
Blank-start also means relevance gates what surfaces. Claude retrieves from memory selectively, only when it judges past context useful, so a new chat about an unrelated topic will not dredge up last week's project. We cover why most assistants struggle with this in our breakdown of why ChatGPT keeps forgetting you and the real fix.
What are Claude's two memory layers?
There are two. An editable summary you control, and an automatic synthesis Claude rebuilds for you. The summary holds the standing facts about who you are, your role, your preferences, and how you like Claude to respond. Claude builds the synthesis by summarizing your conversations, and it rebuilds that synthesis every 24 hours as you create, change, or delete chats.
Think of the summary as the durable part and the synthesis as the rolling part. The summary holds what you explicitly told Claude to remember: I am a backend engineer, keep answers concise, default to Python. The synthesis is a daily digest of recent activity, refreshed without your input, so it tracks what you have been working on lately.
What belongs in memory, and what does not
Anthropic limits the scope on purpose. It designed memory for general context about you as a person and a professional: your role, preferences, and communication style. Sensitive data does not belong there. Anthropic's guidance recommends being thoughtful about highly sensitive details, naming financial information such as Social Security numbers, credit card numbers, and bank account details, along with passwords and private login credentials.
Treat the memory summary like a profile, not a vault. Standing preferences belong there. Account numbers and secrets do not. Anything you write to memory may surface in a future chat where you would not expect it.
Where do you view and edit Claude memory?
To view or edit Claude memory, open Settings, go to Capabilities, then click View and edit memory. There you see the full profile, exactly as Claude holds it. You can add a fact, correct a wrong one, or delete an entry, and edits apply to your next conversation. You can also tell Claude what to remember directly inside a chat, and it updates the summary without you leaving the conversation.
Memory is portable too. Claude added an import and export flow so you can move context in and out. To export, open the memory view and read your full profile, or ask Claude in chat to write out its memories of you verbatim. To import, use the Import memory card on the home screen or in Settings, Capabilities, then paste an export from another assistant, and Claude extracts the details into individual memory edits.
Two caveats from Anthropic's own docs. Import and export are available on Free, Pro, Max, and Team plans, on the web and Claude Desktop. And import is experimental: Anthropic warns it may not always incorporate imported memories successfully, and it prioritizes work-related content over unrelated personal details. Export is simpler, since you are just reading back the stored text.
Memory reaches every Claude tier, free included. It launched first for paid users, and Anthropic later extended it to the free plan along with the ability to import context from rival assistants in the same rollout.
How Claude memory compares to ChatGPT and Gemini
Here is what most guides won't tell you: the transparency gap has narrowed. ChatGPT now shows inline source citations for the saved facts behind a response, and Gemini labels when it draws on Your saved info. So the line that all three remember you, and all three let you read and edit the record, is true today. The remaining difference is when you see retrieval, not whether.
| Capability | Claude | ChatGPT | Gemini |
|---|---|---|---|
| Read the raw memory text | Yes, full profile in Settings | Yes, saved memories list | Yes, saved info list |
| When retrieval is shown | Live tool call during the search | Source citations after the answer | Saved-info label after the answer |
| Edit or delete individual entries | Yes | Yes | Yes |
| Auto synthesis rebuild cadence | Every 24 hours | Continuous, not user-visible | Continuous, not user-visible |
| Import memory from rivals | Yes, experimental | No native import | No native import |
The takeaway: Claude is the only one of the three that shows retrieval live, as a search step you watch run before the answer, rather than a label attached after it. It is also the only one with a native import path from rivals. If portability is your concern, the real question is whether this memory is yours to take. We dig into that in can you export your AI memory, and into the ownership angle in who owns your AI memory.
Does Claude train on your conversations?
No, not by default. Anthropic's privacy documentation states that it uses your chats to improve its models only if you choose to allow it, alongside narrow cases like safety reviews or opting into a trusted tester program. Anthropic never trains on incognito chats. If you have enabled model improvement, you can turn it off again in Privacy Settings.
One number to know. If you allow training, Anthropic retains conversations for up to five years. If you decline, the standard 30-day retention applies. The choice covers Free, Pro, and Max accounts and Claude Code from those accounts. The consumer training policy excludes commercial offerings: Claude for Work, Claude for Government, Claude for Education, and the API.
Action item: open Privacy Settings once and confirm the model improvement toggle reads the way you want. It governs the five-year versus 30-day retention difference, independent of whatever lives in your memory profile.
Claude memory vs a dedicated memory app
Claude's memory is built for one assistant. It remembers how you like Claude to talk and what you have discussed inside Claude. That is the right tool for shaping AI conversations. It is not where your photos, receipts, PDFs, voice notes, and messages should pile up.
That gap is why we built MemX (memx.app), a personal second brain for your own material. You save photos, PDFs, documents, voice notes, and WhatsApp messages, then ask in plain English to get them back, with MemX citing the exact source memory it answered from. It reads text inside photos and scans with OCR, so a snapshot of a receipt becomes searchable, not just a filename. Say a reminder naturally and MemX writes it to your calendar, recurring ones included. We encrypt your data, never train AI on it, and lock out our own staff. There is a free tier with no credit card on Android, iOS via TestFlight, and WhatsApp.
Use Claude memory to make the assistant know you. Use a dedicated memory app for the documents and details you need to retrieve for years. Different jobs, different tools.
01Can I see exactly what Claude remembers about me?
Yes. Open Settings, go to Capabilities, and click View and edit memory. You see the full profile in plain text, and you can edit or delete any entry. Claude also shows retrieval live in the chat flow rather than hiding it behind silent personalization.
02How often does Claude update its memory of me?
Claude rebuilds its automatic synthesis of your conversations every 24 hours as you create, change, or delete chats. Edits you make directly to your memory summary apply to your next conversation right away, without waiting for the daily rebuild.
03How do I turn off Claude's memory training?
Open Privacy Settings and switch off the model improvement toggle. Training is off by default and only runs if you allow it. Turning it off keeps the standard 30-day retention; leaving it on extends retention to up to five years. Incognito chats are never trained on.
04Is Claude memory available on the free plan?
Yes. Memory reaches all Claude tiers, with free users getting both memory and the ability to import context from rival assistants. Import and export work on Free, Pro, Max, and Team plans across web and Claude Desktop.
05Can I move my memory from ChatGPT or Gemini into Claude?
Yes, through Claude's import flow. Export your memory from the other assistant, then paste it into Claude's import card or Settings. Anthropic notes the feature is experimental, may not capture everything, and prioritizes work-related context over unrelated personal details.
Written by Arpit Tripathi, who works on AI memory at MemX. Sources: Anthropic Help Center on chat search and memory, on importing and exporting memory, and on sharing sensitive data; Anthropic Privacy Center on data and training; Anthropic's consumer terms update; 9to5Mac on the free-tier rollout; MindStudio on GPT-5.5 inline memory citations; and Google on Gemini saved info.
