Table of Contents
Quick Answer
ChatGPT memory breaks when it's turned off, at capacity, or your account type doesn't support it. Fix: go to Settings → Personalization → Memory → ON, then delete old memories to make room.
- Memory must be enabled in Personalization settings
- Memory has capacity — delete old entries if full
- Team/Enterprise admins may disable memory globally
Why This Happens
ChatGPT memory is a Plus+ feature that stores facts across chats. It can fail because: memory is off, capacity is full (you'll see "Memory full" indicator), Temporary Chat is on (doesn't save), your plan doesn't include memory (free tier limited), or admin policy disabled it for your Workspace.
Step-by-Step Fixes
Step 1: Confirm your plan
Memory works on Free (limited), Plus, Pro, and Team (if admin enabled). Not on API.
Step 2: Turn memory ON
Settings → Personalization → Memory → toggle ON.
Step 3: Check memory storage
Click "Manage" under Memory. You'll see stored entries and capacity. Free/Plus: limited; Pro: much higher.
Step 4: Delete old memories
If capacity is full, ChatGPT stops adding new ones. Delete irrelevant entries to free space.
Step 5: Exit Temporary Chat mode
If the chat icon shows a ghost/incognito symbol, you're in Temporary Chat (no memory saved). Click New Chat for a normal chat.
Step 6: Explicitly ask to remember
"Remember that I prefer concise bullet answers." Confirmation appears: "Memory updated".
Step 7: Review what's stored
Settings → Personalization → Memory → Manage memories. Edit or delete wrong/outdated facts.
Step 8: Clear all memories (fresh start)
Settings → Personalization → Memory → Clear all memories. Rebuild from scratch.
Step 9: Check custom instructions conflict
Settings → Personalization → Custom Instructions. Conflicts between memory and instructions can cause erratic behavior.
Step 10: For Workspace accounts
If "Memory not available" appears, admin disabled it. Contact your admin.
When to Contact Support
- Memories delete themselves immediately after adding
- "Memory updated" appears but memory doesn't persist across sessions
- Bulk memory loss after a deploy (likely a bug)
Support: help.openai.com
Prevention Tips
- Prune memory monthly — stale facts pollute responses
- Use memory for preferences, not sensitive data
- Review memories before critical tasks (wrong facts affect output)
- Custom Instructions are better for constants (role, style); Memory for evolving facts
Conclusion
Memory is powerful but needs management. For multi-model AI with persistent context across providers, try Assisters AI.