ChatGPT Memory — How It Works & How to Manage It
ChatGPT's Memory feature lets it remember facts about you and your preferences across separate conversations — your job, projects, writing style, dietary restrictions, the names of your kids. It works invisibly: ChatGPT decides what to remember during normal conversation, and surfaces those memories in future chats when relevant. This guide covers what it actually stores, how to control it, and the privacy and behavior gotchas most users miss.
Memory = ChatGPT writes short notes about you into a per-account store. Those notes are loaded into every conversation's context. You can view, edit, and delete any entry. Available on Plus, Pro, Business, Enterprise. Free tier has a limited version. Memory is account-scoped — Custom GPTs you build do not get your personal memories.
What Memory stores
When you mention something durable about yourself in conversation, ChatGPT writes a short fact to its memory store. Examples of what typically gets stored:
- Personal facts: "User lives in Boston." "User has two children, Mia and Leo." "User is allergic to shellfish."
- Work context: "User is a product manager at a mid-stage SaaS company." "User's main project is a Python data pipeline."
- Preferences: "User prefers concise responses." "User likes code examples in TypeScript, not JavaScript." "User does not like emojis in replies."
- Ongoing projects: "User is writing a novel set in 1920s Cairo." "User is renovating a kitchen and currently choosing tiles."
- Explicit "remember" requests: "Remember that my project deadline is October 15." ChatGPT adds it verbatim or paraphrased.
What Memory does not store: full conversation history (that's separate, in your chat list), file uploads, images, or anything from incognito (temporary) chats.
How to view, edit, and delete memories
- Open Settings (click your name → Settings).
- Click Personalization in the left sidebar.
- Click Memory.
- You'll see two lists: Saved memories (what ChatGPT has stored) and a toggle to disable Memory entirely.
- To delete a single memory: hover the entry, click the trash icon.
- To delete all memories: click Clear all at the bottom.
- To turn Memory off entirely: toggle the master switch off. Existing memories are kept (in case you re-enable) but not used in conversations.
After deleting a memory, it's gone immediately from new conversations. Old conversations that referenced that memory remain — you'd need to delete those chats separately.
How Memory shows up in conversations
When you start a new chat with Memory enabled, ChatGPT silently reads your memory store and uses anything relevant. You don't see the memories injected, but they shape the response. To check what Memory is actually using:
- Look for the "Memory updated" indicator that flashes near a response. Click it to see what was just added.
- Ask directly: "What do you remember about me?" ChatGPT will summarize. Caveat: the summary may be incomplete or paraphrased — verify against the Settings page if precision matters.
- Use a Temporary Chat to test "no-memory" behavior. Anything you tell ChatGPT in a Temporary Chat is not stored, and existing memories are not loaded.
Tier access
| Tier | Memory | Notes |
|---|---|---|
| Free | Limited | Smaller memory store, slower update cadence |
| Plus ($23/mo) | Full | Standard memory feature |
| Pro (~$200/mo) | Full + Projects scoping | Memories can be scoped to a Project |
| Business | Full, admin-controllable | Workspace admins can disable Memory org-wide |
| Enterprise | Full, audit-logged | Memory writes and reads can be audit-logged for compliance |
Privacy considerations
- Memories live on OpenAI's servers. They're tied to your OpenAI account and subject to OpenAI's privacy policy and data-handling practices.
- Free and Plus memories may be used for model training unless you turn off "Improve the model for everyone" in Settings → Data controls. Business and Enterprise have stricter defaults.
- Memories are not encrypted at rest with your key — they're encrypted with OpenAI's infrastructure keys, like other account data. Don't put genuinely sensitive secrets (passwords, SSNs, medical PHI) in Memory.
- Workspace admins on Business/Enterprise can disable Memory org-wide if compliance requires it.
- Custom GPTs you build do not have access to your personal memories — Memory is account-scoped, not GPT-scoped. A Custom GPT can have its own "knowledge files" but those are configured separately.
Common gotchas
- "It forgot something I told it." Memory is selective — ChatGPT doesn't store everything. If something is important, say "Please remember this:" explicitly. Then verify in Settings.
- "It remembers something I never told it." Memory inference happens during conversations. A throwaway phrase can become a memory. Audit the list periodically.
- "My responses got weird after I worked on a sensitive topic once." If a one-time topic shaped a memory ("User is exploring divorce options"), it persists into unrelated future chats. Delete the stale memory.
- "Memory is supposed to be off but ChatGPT still 'knows' things." Disabling Memory only stops it from using memories — your old memories are kept. Click Clear all if you want a fresh start.
- Memory across devices: works automatically — memories are account-scoped, not device-scoped. Login on a new device, your memories follow.
- Memory in mobile vs desktop: identical — same store, same behavior.
Memory vs other "persistent agent context" features
| Platform | Equivalent feature | How it differs from ChatGPT Memory |
|---|---|---|
| Claude Cowork | Projects + System Prompts | Project-scoped, not account-wide. Explicit (you author the prompt), not inferred. |
| OpenClaw | SOUL.md | Single file you edit directly. Full transparency. Versioned in git. |
| Hermes | Three-tier persistent memory | Database-backed (SQLite/PostgreSQL). Far more powerful but also far more involved. See the Hermes memory guide. |
| Kilo Code | Session-scoped only | No persistent cross-session memory — each session starts fresh. |
← Back to ChatGPT hub · Previous: Agent Mode · Next: Teams & Business →