As of March 2026, Claude remembers you. Not just within a conversation — across conversations, across weeks, permanently until you tell it to forget. Anthropic quietly rolled out persistent memory to every Claude account — free tier and Claude Pro alike — in early March 2026. If you use Claude at claude.ai or through the Claude iOS or Android app, memory is on by default right now, and Claude has probably already stored things about you from your recent conversations. This guide covers everything you need to know: what Claude is actually storing, what it is explicitly designed not to store, how to pull up your full memory list in under 60 seconds, how to delete specific items or everything at once, how this compares to ChatGPT's memory (they are meaningfully different in one key way), the privacy implications most guides skip over, and the exact prompts that make memory go from mildly useful to genuinely transformative for students, developers, writers, and professionals. Source: Anthropic memory launch announcement, March 2026.
Quick summary for readers in a hurry: Claude memory is on by default for all users as of March 2026. It stores your preferences, context about your work, and facts you share — not conversation transcripts. You can view everything Claude has stored by going to Settings > Memory in claude.ai. You can delete individual memories or clear everything. You can disable memory entirely if you prefer a fresh-start model. Temporary Chat mode lets you have private conversations that never create memories. Source: Claude settings documentation, March 2026.
What Claude Actually Stores in Memory
Claude's memory system is not a conversation log or transcript. It is a structured set of derived facts and preferences that Claude extracts from what you share and stores for future reference. The best analogy: think of it as the notes a great assistant takes between meetings — the key context that makes every future interaction more useful without needing you to re-explain your situation from scratch. Claude stores four main categories of information. Source: Anthropic memory documentation, March 2026.
- Your stated preferences: how you like responses formatted (bullet points vs prose vs code blocks), how long you want answers (concise vs comprehensive), whether you prefer formal or casual language, and specific style preferences you have expressed — for example, 'always give me the TL;DR first' or 'explain things like I'm new to this field.'
- Context about your work and study: your programming languages, frameworks, tech stack, subjects you are studying, your year and major if you're a student, your job title or field if you've mentioned it, and the specific projects or goals you're working toward.
- Facts you have shared: your name if you've used it, your location if relevant to queries, specific ongoing projects and where they stand — for example, 'I'm building a B2B SaaS product for restaurant inventory management, currently in the pricing page sprint.'
- Working style preferences: whether you want Claude to ask clarifying questions before attempting a task or just dive in, whether you want it to flag potential problems in your approach or just execute the request, and whether you want step-by-step explanations or just the final output.
- What Claude explicitly does NOT store: verbatim conversation content or transcripts. Sensitive credentials like passwords, API keys, or bank details — Claude is designed to decline storing these. Personal health details beyond what you explicitly ask it to remember. Information shared in conversations that happened before memory was activated for your account.
How Claude Memory Works: The Mechanics
When Claude detects something worth remembering — a stated preference, a key fact about your work, a working style signal — it creates a memory entry automatically. You do not need to tell Claude to remember something, though you can explicitly ask it to. When you start a new conversation, Claude reads your stored memories and incorporates the relevant context into how it responds, without you needing to re-establish that context. Claude is also designed to be transparent about when it is using memory — it will typically acknowledge when stored context is influencing a response, unlike some other AI systems that use memory silently. Source: Anthropic, March 2026.
There are two important constraints on how memory works: First, memory is account-scoped — your memories are private to your Claude account and are never shared with other users or visible to anyone at Anthropic other than for safety and abuse review purposes. Second, memory is additive, not a running transcript — Claude stores extracted facts and preferences, not a full log of what was said. This means Claude will remember 'User prefers concise, bullet-pointed answers' rather than storing every word of the conversation where you expressed that preference. Source: Anthropic privacy policy, 2026.
How to View, Edit, and Delete Your Claude Memories: Step-by-Step
Viewing and managing your memory takes under 60 seconds on any platform. Here are the exact steps for each surface Claude is available on. Source: Claude interface documentation, April 2026.
- Claude.ai (desktop browser): Navigate to claude.ai. Click your profile icon or initial in the bottom-left corner of the sidebar. Select 'Memory' from the dropdown menu. You will see a full list of every stored memory, each with a delete button. A 'Clear all memories' button is also available at the top of the list. Changes take effect immediately — the next conversation starts without the deleted memories.
- Claude mobile app — iOS: Open the Claude app. Tap the three-horizontal-line icon in the top-left corner to open the menu. Tap 'Settings'. Select 'Memory'. You will see the same list with the same individual-delete and clear-all options. The interface is identical to the web version.
- Claude mobile app — Android: Open the Claude app. Tap the profile icon or menu in the top-left. Go to Settings > Memory. Same interface, same controls.
- Ask Claude directly (fastest audit): In any conversation, type 'What do you remember about me?' Claude will list everything currently in its memory in that conversation. This is faster than navigating settings and gives you Claude's own interpretation of what it is using.
- Correct or update a memory via conversation: You do not need to go into settings to update a memory. Tell Claude directly: 'I'm no longer working on that restaurant project — I've switched to a healthcare app focused on appointment scheduling. Please update what you remember.' Claude will revise its stored context accordingly.
- Delete a specific memory from conversation: You can also say 'Please forget that I mentioned my location' or 'Remove your memory of my tech stack — I want to start fresh on that.' Claude will confirm the deletion.
How to Enable or Disable Claude Memory (and What Happens Either Way)
Memory is enabled by default for all Claude accounts as of March 2026. You have three options for controlling it: leave memory on (default), turn off memory entirely, or use Temporary Chat mode for individual conversations where you want no memory created. Source: Claude settings documentation, March 2026.
- Leaving memory on (recommended for most users): Claude progressively learns your preferences and context. Each conversation benefits from everything you've established in prior conversations. This is the default and the setting that delivers the most personalized, useful experience over time.
- Turning off memory entirely: Go to Settings > Memory in claude.ai or the mobile app. Toggle memory off. With memory disabled, every conversation starts completely fresh — Claude has no recollection of anything from prior conversations. This is the right choice for users who prefer maximum privacy, work in sensitive professional environments, or simply find the idea of persistent AI memory uncomfortable.
- Temporary Chat mode (per-conversation privacy): Start any conversation in Temporary Chat mode by clicking the 'Temporary' toggle at the top of a new conversation. Conversations in Temporary Chat mode: (1) do not create any new memories, (2) do not have access to your existing memories, and (3) are not used for AI training by default. This is the right tool for conversations about sensitive topics where you want a completely fresh, private session without disabling memory for all your other conversations.
- What happens to existing memories when you disable memory: disabling memory does not delete your existing stored memories — it only stops new ones from being created. Your memory list is preserved and will be active again if you re-enable memory. To actually erase your stored memories, use the 'Clear all memories' option in the Memory settings panel.
Claude Memory vs ChatGPT Memory: The Key Differences in 2026
Both Claude and ChatGPT now have persistent memory for all users, free and paid. The surface-level features are similar — both store preferences and context, both let you view and delete memories, both have a temporary/private chat mode. But there are meaningful differences in how they handle memory transparency, user control, and the underlying approach to what gets stored. Source: OpenAI memory documentation, 2026; Anthropic memory documentation, 2026.
| Feature | Claude Memory (2026) | ChatGPT Memory (2026) |
|---|---|---|
| Availability | All users — free and Pro — since March 2026 | All users — free and Plus — since 2024 |
| What gets stored | Preferences, stated context, working style — extracted facts, not transcripts | Preferences, stated context, working style — similar extraction approach |
| Transparency about memory use | Claude explicitly tells you when it is using a stored memory in a response | ChatGPT uses memories silently — you are not told when a memory is influencing a response |
| User control — view | Full list in Settings > Memory; also accessible by asking Claude directly in conversation | Full list in Settings > Personalization > Memory — also askable in conversation |
| User control — edit | Delete individual items or clear all; update via conversation | Edit individual memory text directly in settings; delete individual or clear all |
| ChatGPT advantage: manual editing | Cannot edit memory text directly — must delete and re-add via conversation | Can directly edit the text of any stored memory in settings — more surgical control |
| Temporary/private mode | Temporary Chat mode — bypasses memory entirely, no memories read or created | Temporary Chat — same function, same protections |
| Opt-out | Full disable available in settings; existing memories preserved until manually cleared | Full disable available in settings; same preservation behavior |
| Privacy architecture | Stored server-side by Anthropic; private to your account; subject to Anthropic privacy policy | Stored server-side by OpenAI; private to your account; subject to OpenAI privacy policy |
| API / Claude.ai-only | Memory is a Claude.ai feature — does not apply to API-accessed Claude or third-party apps | Memory is a ChatGPT.com feature — does not apply to API-accessed GPT models |
The single most meaningful functional difference: Claude tells you when it is using a memory. ChatGPT does not. If you open a conversation and Claude references something from a previous session, it will acknowledge this explicitly — for example, 'Based on what you told me about your React project last week...' ChatGPT incorporates stored context silently, with no indication that a memory is shaping the response. This transparency difference matters if you want to understand and audit how AI is personalizing its responses to you. Source: comparative user testing, LumiChats, April 2026.
The Memory Prompts That Actually Transform Your Workflow
The most valuable thing you can do with Claude memory is not let it accumulate passively — it is to invest 5 minutes intentionally setting it up in your first post-memory conversation. The prompts below are designed to give Claude the context it needs to be significantly more useful in every future session. Copy, customize, and send them as your first message after enabling memory.
- Master setup prompt (universal): 'I want to set up my memory intentionally. Please store the following: My name is [Name]. I work as [role] at [type of company or in what context]. My primary use of Claude is [1-2 sentence description]. I prefer responses that are [formal/casual], [concise/comprehensive], formatted as [prose/bullets/code when relevant]. I want you to [ask clarifying questions / just attempt the task] when I give you an ambiguous request. The ongoing projects I'll be coming back to are: [brief list]. Please confirm you've stored this.' — This single prompt sets the foundation for a much more useful relationship with Claude.
- For US students and researchers: 'Please store this about my academic context: I'm a [year] studying [major] at [type of school — large state university, liberal arts college, community college, etc.]. My strongest areas are [X and Y]. I consistently struggle with [Z]. I learn best when I see concrete examples before abstract definitions. My current academic goals are [exam/paper/project]. When helping me study, always check my understanding before moving on — don't just give me answers. Source my explanations to specific concepts I can look up.' This turns Claude into a tutor that actually knows your academic profile.
- For software developers: 'Please remember my development setup: Primary language: [Python/JavaScript/Rust/etc.]. Framework: [React/FastAPI/Rails/etc.]. Operating system: [macOS/Linux/Windows]. Current project: [1-2 sentence description of what you're building]. I prefer code explanations that cover the why behind architectural decisions, not just the what. When I ask for code, default to production-ready with error handling and type annotations unless I specify otherwise. When I share an error, walk me through diagnosing it rather than just giving me a fix.' This eliminates the context-setup tax in every coding conversation.
- For writers and content creators: 'Remember my writing voice and preferences: I write [formally/conversationally/journalistically/academically]. Sentence style: [short and punchy / longer and analytical / varied]. Things I avoid: [jargon / passive voice / specific words I overuse — list them]. My target reader is [specific audience description]. Primary content I'll be working on: [articles / scripts / fiction / reports / social content]. When editing my work, prioritize clarity over elegance, and always explain why you're suggesting a change — not just what to change.' Claude will apply this voice profile consistently in every future editing and writing session.
- For professionals and business users: 'Store this professional context for our future conversations: My role is [title] and I work in [industry]. The decisions I use AI for most are [list 2-3 types — e.g., drafting client communications, analyzing market data, preparing presentations]. I communicate with [internal team / external clients / executives / all of the above]. Tone for my work: [formal / semi-formal / startup casual]. When I ask you to help draft something, match the tone appropriate to the audience unless I specify otherwise. I'll sometimes share confidential context for analysis — do not reference it outside of the conversation where I shared it.' This gives Claude professional context that makes every work-related conversation more precise.
- Maintaining memory as projects evolve: 'I've finished the [module/phase/project] I mentioned before. Please update your memory: the current focus has shifted to [new project or phase description]. The main challenge right now is [specific problem]. The decision we've landed on is [approach]. Next steps are [list].' Use this update prompt at the end of any major project phase — Claude's memory of your work stays current rather than drifting out of date.
The most underrated memory use case: session handoffs on complex projects. Before you close a conversation, tell Claude exactly what to preserve: 'Please store a project checkpoint: We're building the recommendation engine. Current blocker: the cold-start problem for new users. Decision made: collaborative filtering with a content-based fallback for users with fewer than 10 interactions. Next session should start with implementing the item similarity matrix. Relevant file: /src/recommendations/similarity.py.' When you return tomorrow or next week, Claude picks up with full context — no re-explaining, no lost decisions, no repeated reasoning.
Privacy: What Anthropic Actually Does With Your Memory Data
This section covers the facts from Anthropic's published policies — not speculation. Memory data is stored server-side by Anthropic and is subject to Anthropic's Privacy Policy and Usage Policy. Your memories are private to your account and are not shared with other Claude users. Anthropic's standard data usage policy applies to memory content: by default, Anthropic may use conversation data (including memory-creating conversations) to improve Claude's models — but you can opt out of this in your account privacy settings. Deleting a memory removes it from your account immediately. Deleting your Claude account deletes all associated memory data. Source: Anthropic Privacy Policy, 2026.
Practical privacy guidance for US professionals: Claude's memory system carries the same sensitivity considerations as any AI tool you use for professional work. For attorneys, doctors, financial advisors, and others with confidentiality obligations: do not use Claude's memory feature to store client-identifiable information, case details, or any data covered by HIPAA, attorney-client privilege, or SEC regulations. Use Temporary Chat mode for any conversation involving sensitive client information — it creates no memories and leaves no persistent record in your account. The American Bar Association and ABA Tech Report both flag that AI tools with persistent memory carry heightened confidentiality risks for legal professionals. Source: ABA Formal Opinion 512, 2023 (AI tools and professional responsibility).
Common Memory Mistakes and How to Avoid Them
- Letting memory accumulate unchecked: most users never look at their memory list. After a few months, it can contain outdated information — old projects, former jobs, preferences that have changed. Set a calendar reminder to review your memory list quarterly. Stale memories can make Claude less accurate rather than more.
- Sharing sensitive credentials in conversations: Claude is designed not to store passwords, API keys, or financial account numbers — but it is not guaranteed to catch every case. Never share actual credentials with Claude regardless of whether memory is on or off. If you need Claude to understand an authentication flow, describe it conceptually.
- Expecting memory to work across platforms: Claude memory is specific to claude.ai and the official Claude iOS/Android apps. If you access Claude through a third-party app, an API integration, or a plugin in another tool, those conversations do not have access to your memory and do not create memories. This surprises many users who assume memory is universal.
- Using memory for genuinely confidential professional information: if you are a lawyer, doctor, therapist, or financial professional with clients, do not use Claude's memory system to store any client-identifiable information. Use Temporary Chat mode for any client-related conversations. Your professional confidentiality obligations do not pause because you are using AI.
- Not explicitly setting preferences in the first session: the biggest missed opportunity with memory is waiting for Claude to infer your preferences over time rather than stating them explicitly upfront. Claude learns faster and more accurately from explicit statements ('I prefer bullet points over prose') than from behavioral patterns. Use the setup prompts in the section above.
Frequently Asked Questions
01Does Claude memory work on the free tier?
Yes. Anthropic activated memory for all Claude accounts — free tier and Claude Pro — in March 2026. You do not need a paid subscription to use Claude's memory feature. The memory experience is identical on both tiers: the same storage, the same settings controls, the same Temporary Chat mode. The difference between free and Pro is not memory access — it is usage limits, priority access, and model tier (Claude Sonnet 4.6 on free vs Claude Opus 4.6 on Pro). Source: Anthropic memory announcement, March 2026.
02Can I see everything Claude has stored about me?
Yes, and it's easy. Go to Settings > Memory in claude.ai or the Claude mobile app. You'll see the complete list of everything Claude has stored — each memory entry is displayed as a short text snippet. You can also ask Claude directly in any conversation: 'What do you remember about me?' and it will list its current memory contents for that session. The settings view and the conversation view should match — if they don't, the settings view is the authoritative record. Source: Claude interface documentation, April 2026.
03What is Temporary Chat and when should I use it?
Temporary Chat is a per-conversation privacy mode in Claude. When you start a conversation in Temporary Chat mode, Claude does not read your existing memories (it starts completely fresh) and does not create any new memories from that conversation. It is also not used for AI training by default. Use Temporary Chat for: conversations about sensitive personal or professional topics, any work involving client-confidential information, situations where you want Claude to respond without any of your established preferences influencing the output, or simply when you want a clean-slate session. You can find the Temporary Chat toggle at the top of a new conversation in claude.ai and in the Claude mobile apps. Source: Claude support documentation, 2026.
04Does deleting a memory actually remove it from Anthropic's servers?
Deleting a memory removes it from your active memory list immediately — Claude will no longer use it in future conversations. Anthropic's privacy policy specifies data deletion procedures for account data, which memory falls under. If you want complete data deletion, deleting your Claude account initiates full data deletion per Anthropic's policy. For users with specific data retention concerns — for example, in GDPR-regulated contexts — Anthropic's privacy policy and their data deletion request process are the authoritative references. Source: Anthropic Privacy Policy, 2026.
05Does Claude memory work in Claude Code or API integrations?
No. Claude memory is a claude.ai and Claude app feature — it does not apply to API access or Claude Code. When you or a developer queries Claude through the API, there is no persistent memory layer. Each API call starts fresh. Claude Code similarly does not have access to your claude.ai memory profile. If you are building an application using the Claude API and want persistent user context, you would need to implement your own context management — either storing conversation history in your database and passing it with each API call, or using a vector database for semantic retrieval. Source: Anthropic API documentation, 2026.
06Will Claude remember things I said before memory was turned on?
No. Memory is prospective — it only stores information from conversations that happened after memory was activated for your account. If you had months of conversations with Claude before March 2026, none of that content was retroactively stored in your memory. Your pre-memory conversation history exists in Claude's conversation logs (visible in your sidebar) but was not processed into structured memories. This is why using the intentional setup prompts in this guide — explicitly telling Claude your preferences and context in a post-memory conversation — gives you better results than waiting for Claude to infer preferences from your existing history. Source: Anthropic memory documentation, March 2026.
07Can I use Claude memory for professional work?
Yes, with important limitations depending on your profession. For most professionals — engineers, marketers, consultants, business analysts, writers — Claude memory is a significant productivity asset. Store your work context, project details, communication preferences, and domain knowledge. For professionals with legal confidentiality obligations — attorneys, physicians, therapists, financial advisors — use memory for general professional context ('I work in corporate M&A law', 'I prefer formal language for client-facing content') but use Temporary Chat mode for any conversation involving actual client information. Your professional ethics obligations apply to how you use AI tools, including memory-enabled ones. Source: ABA Formal Opinion 512 on AI tools; HIPAA guidance on AI tools, 2024.
08How is Claude memory different from just starting every conversation with a long system prompt?
Functionally, memory achieves similar personalization to starting every conversation with a detailed context message — but automatically, without you needing to do it. The practical differences: memory is maintained and updated by Claude as your context evolves; a manual prompt must be updated by you. Memory also applies to mobile conversations seamlessly without pasting text. For power users who have already been using detailed context prompts, memory is a direct upgrade — it takes that same intentional setup and makes it persistent. You can also layer the two approaches: let memory handle your stable preferences, and use a brief conversation opener to add session-specific context on top.
One practical test to see memory working: start a new conversation and ask Claude 'What do you remember about me?' If memory is working correctly, Claude will list your stored preferences and context. If it says it has no memories, check that memory is enabled in Settings > Memory, or check whether you are in a Temporary Chat session. The direct-question audit is the fastest way to verify that your memory setup is working as intended.
LumiChats gives US users access to Claude Sonnet 4.6 — the memory-enabled Claude model — plus GPT-5.4, Gemini 3.1 Pro, and 37 other frontier AI models in a single session, at approximately $1 per day with no monthly commitment. Study Mode adds another layer of personalization on top of Claude memory: it pins AI responses to your uploaded course materials with page citations, so every answer is grounded in your actual textbooks and notes. For students and professionals who want AI that genuinely knows their context and their materials, this is the most personalized AI setup available in 2026.
