Layer 2 — the profile file
Delete ChatGPT's memory of you, completely.
Memory is the closest thing to a dossier an AI keeps: distilled, standing facts about your life that persist across every new chat. It's also the easiest layer to erase — five minutes, fully reversible regret.
The procedure
Full memory wipe, step by step.
- Read what it has on you first.ChatGPT → Settings → Personalization → Memory → Manage. This is the actual list — job, family, health mentions, preferences. Screenshot it if you're filing a formal request later; it's useful evidence of what was held.
- Delete the entries — selectively or all.Delete individual memories (the sensitive ones first), or hit Clear ChatGPT's memory for the full wipe.
- Toggle Memory off.Same screen. Off means no new dossier accumulates. You can also tell it in-chat: "Forget that I…" — then verify in Manage that it actually did.
- Kill chat-history referencing.Toggle off Reference chat history too — otherwise the model can still mine your old conversations for context. OpenAI deletes the associated data from its servers within ~30 days of the toggle.
- Use Temporary Chat for anything sensitive from now on.Temporary chats don't write memory, don't appear in history, and aren't used for training. The habit that prevents the next cleanup.
This is a whole subject — we have a whole site on it
Memory limits, memory for business accounts, saved memories vs custom instructions, why memory 'forgets,' temporary chat mechanics — our sister site covers ChatGPT memory wall to wall.
chatgptmemory.com: full deletion guide → Memory & privacy →Beyond ChatGPT
Memory switches at the other providers.
| Service | Where the profile lives | How to clear it |
|---|---|---|
| Gemini | Gemini Apps Activity + personalization | myactivity.google.com/product/gemini → delete activity; turn Apps Activity off. Note: human-reviewed samples are retained separately (up to 3 years). |
| Claude | Project knowledge & chat history | Delete conversations/projects; memory-style features are scoped and removable per project. Formal requests via privacy.anthropic.com. |
| Copilot | Personalization from your history | Copilot privacy settings → personalization off; delete activity via account.microsoft.com/privacy. |
| Grok | Conversation history + X account data | Grok data controls → personalization/history off; X Settings → Privacy → Grok data sharing off. |
What people miss
Clearing memory is not the same as deleting every trace.
Memory is only one layer. Use this checklist to avoid the most common false reset.
The explicit profile
This is the readable list: facts ChatGPT decided should persist. Delete these first because they are direct, inspectable, and often surprisingly personal.
The hidden context layer
Even with saved memories cleared, past chats can still influence future answers if chat-history referencing remains enabled. Turn it off for a real reset.
The profile you wrote yourself
Custom instructions can contain job, location, family or tone preferences. Review them separately because clearing memory does not rewrite them.
Uploaded material can remain separate
Files, projects, GPT knowledge and shared links may have their own deletion controls. Remove sensitive uploads as their own cleanup task.
Memory off does not equal training off
Memory controls personalization. Training controls whether conversations can improve models. For privacy, check both.
Screenshot before you wipe
If you plan a formal deletion request, capture the memory list first. It proves the provider held a profile-like record about you.
Verification
How to confirm the memory wipe worked.
- Reopen the memory manager.After deletion, close settings, reopen Settings → Personalization → Memory → Manage, and verify there are no saved entries left.
- Ask a neutral test question.Try: "What do you remember about me?" A correct answer should say it does not retain personal facts unless you provide them in this chat.
- Check custom instructions and projects.If the model still knows your role or preferences, the information may be in custom instructions, a project, or the current conversation rather than memory.
- Use temporary chat for the next sensitive topic.Do not rely on periodic cleanup. Use the mode that avoids history and memory from the start.
FAQ
Memory questions.
Where is ChatGPT's memory of me stored?
Settings → Personalization → Memory. 'Manage' shows the exact list of saved facts; each can be deleted individually, or use 'Clear ChatGPT's memory' to wipe all of them.
What's the difference between saved memories and chat-history referencing?
Saved memories are explicit distilled facts; 'Reference chat history' lets the model consult your past conversations wholesale. For a real reset, clear memories AND toggle off chat-history referencing — the referencing data is deleted from servers within about 30 days of turning it off.
Do other AIs have memory too?
Yes: Gemini personalizes from Gemini Apps Activity (delete at myactivity.google.com/product/gemini), Grok and Copilot personalize from conversation history with their own toggles, and Claude keeps project-scoped context. Same principle everywhere: find the personalization setting, read what it holds, clear it.
Will clearing memory change my answers?
Yes — responses lose the tailoring ('as a teacher, you might…'). Many people prefer it. You can keep memory on but prune specific entries: delete the sensitive ones, keep the convenient ones.