What ChatGPT Memory Does
Memory gives ChatGPT a persistent understanding of who you are across conversations. Without it, every chat starts from scratch — you are a stranger every time. With memory on, ChatGPT carries relevant facts forward automatically.
What gets remembered:
- Your job title and industry
- Preferred response format (bullet points, short answers, code blocks)
- Ongoing projects and their tech stack or context
- Personal preferences ("I prefer metric units," "I am vegetarian")
- Communication style preferences ("be direct," "skip the disclaimers")
- Explicit instructions you give ("always format dates as DD/MM/YYYY")
What does not get remembered:
- The full text of past conversations
- Sensitive data like passwords, card numbers, or health details (ChatGPT actively avoids storing these)
- One-off context that is only relevant to a single session
How to Enable Memory
- Click your name or avatar in the bottom-left sidebar
- Go to Settings → Personalization
- Toggle Memory to on
- Start a new conversation — memory is now active
Memory begins saving automatically. You will not see a notification each time something is saved; check Manage memories periodically to see what has accumulated.
How to View and Manage Saved Memories
- Settings → Personalization → Manage memories
- Scroll through the list of saved facts
- To delete one: click the trash icon next to the entry
- To delete all: click Clear all at the bottom
You can also manage memories from within a conversation:
- "What do you remember about me?" — ChatGPT lists its current memories
- "Forget that I am a teacher" — removes a specific memory
- "Clear all your memories about me" — wipes everything
How to Save Specific Facts
Explicit saves work better than hoping ChatGPT infers what to keep:
- "Remember that I work in a React/Node.js stack"
- "Remember that my client's company is called Acme Corp and they are in the healthcare sector"
- "Please remember I prefer responses in British English"
- "Note that I want you to always suggest tests when reviewing code"
ChatGPT confirms each save with a short message: "Got it, I'll remember that you prefer British English."
Temporary Chat: When to Turn Memory Off for One Session
Temporary Chat is a privacy mode that disables memory and history for a single conversation.
To start one: click the pencil/new chat icon → look for Temporary chat at the top, or click your avatar → Temporary chat.
Use it for:
- Sensitive topics you do not want associated with your account
- One-off tasks where your usual preferences do not apply
- Testing prompts without polluting your memory with test data
- Sharing your screen with others without exposing your history
Nothing from a Temporary Chat is saved, even if memory is globally enabled.
Common Mistakes to Avoid
- Not checking what has been saved — Memory can accumulate inaccurate facts over time. Check Manage memories monthly and delete anything outdated.
- Expecting memory to replace conversation context — Memory stores facts, not full conversations. If you want ChatGPT to continue a specific project, paste the relevant context at the start of the new chat rather than relying on memory alone.
- Relying on memory for sensitive project details — Do not rely on memory to store confidential client information or proprietary data. Memory storage goes through OpenAI's systems. Keep sensitive context in your own notes.
- Confusing memory with conversation history — History shows past chat logs. Memory stores extracted facts. Both live in Settings but serve different purposes. Deleting history does not delete memories, and vice versa.
- Using memory-heavy chats on shared accounts — Memory is per account. If multiple people use the same ChatGPT account, memories from one person's preferences will affect another person's sessions. Use separate accounts for this reason.
Memory vs. Custom Instructions: Which to Use
Both give ChatGPT persistent context, but they work differently:
| | Memory | Custom Instructions | |---|---|---| | Set by | ChatGPT (automatic) or you (explicit) | You (manual) | | Updated | Dynamically as you chat | Only when you edit settings | | Scope | Facts and learned preferences | Format and behavior rules | | Best for | Project context, personal facts | Tone, format, language style |
Use both together for the best experience. Custom Instructions for how you want responses formatted; Memory for what you are working on and who you are.
Troubleshooting Memory Issues
Memory is on but facts are not being saved
ChatGPT saves facts on a best-effort basis — it does not save every stated preference. If a specific fact is not being remembered:
- Ask explicitly: "Please remember that I always want responses in bullet points" — ChatGPT will confirm the save with a message like "Got it, I'll remember that."
- Check that you are not in Temporary Chat mode (which never saves memories)
- Verify memory is enabled at Settings → Personalization → Memory
Memories are inaccurate or outdated
Over time, ChatGPT may save facts that have since changed. Review and delete outdated memories periodically:
- Go to Settings → Personalization → Manage memories
- Review each entry — delete anything inaccurate with the trash icon
- Tell ChatGPT the corrected fact so it saves the updated version
Memory is affecting responses in unexpected ways
If memory is causing ChatGPT to apply preferences that do not fit a specific conversation:
- Tell ChatGPT at the start of the conversation: "For this conversation, ignore any memory about [topic] and treat it fresh"
- Use Temporary Chat for sessions where your usual preferences should not apply
- Delete specific memories that are causing the interference
Why This Happens
ChatGPT Memory works by having the model identify facts that seem personally relevant and stable, then storing a summary of those facts separately from conversation history. The system is designed to be additive — it saves new facts without automatically overwriting older ones, which is why outdated facts can persist until manually deleted. Memory capacity is not unlimited; OpenAI does not publish the exact limit, but users rarely hit it in practice. Memories are stored per account and synced across all devices using that account.
Common Mistakes to Avoid
- Not checking what has been saved regularly — Memory can accumulate inaccurate facts over time; review Manage Memories monthly and delete anything outdated
- Relying on memory for sensitive project details — Memory storage goes through OpenAI's systems; keep confidential client data in your own notes rather than relying on ChatGPT Memory
- Confusing memory with conversation history — History shows past chat logs; memory stores extracted facts; deleting one does not affect the other
- Using memory-heavy chats on shared accounts — Memory is per account; multiple people sharing one account will have their preferences mixed together; use separate accounts