ChatGPT Memory
ChatGPT memory lets the assistant build a persistent understanding of who you are — your name, preferences, recurring goals, and working style — so that each conversation feels like a continuation rather than a cold start. Understanding how the two memory systems work, and how to manage them, gives you meaningful control over your experience.
Two Memory Systems
ChatGPT uses two distinct memory mechanisms that work together:
Saved Memories
Explicit facts stored as discrete, labelled entries. ChatGPT creates these automatically when it notices something worth remembering — your job title, a project you're working on, a preference you expressed. You can also trigger them manually: "Remember that I prefer concise bullet points" or "Remember my company is called Larconnectado." Saved memories persist indefinitely until you delete them, and are injected into the system context of every new conversation.
Conversation History Reference
Rolled out in 2025 across all plans, this allows ChatGPT to reference information from your recent past conversations — not just the current session. If you discussed a project in a conversation three days ago, ChatGPT can recall relevant context from it without you repeating yourself. Unlike Saved Memories, this is not a labelled fact store — it is a broader retrieval of relevant prior context.
Auto-Prioritisation
On Plus and Pro plans, ChatGPT applies an auto-prioritisation layer to Saved Memories. Memories that are recent, frequently referenced, or actively relevant to current conversations are weighted higher and surfaced more prominently. Stale memories — things that were noted months ago and never referenced since — gradually fade in priority. This prevents older, potentially outdated context from overriding more current information about you.
You cannot directly control the prioritisation weighting, but you can influence it by deleting outdated memories and explicitly asking ChatGPT to remember updated versions of facts.
Managing Your Memories
You can view, edit, and delete your saved memories at any time:
- Open ChatGPT and go to Settings
- Select Personalization
- Click Manage memories
From there you can delete individual memory entries or clear all memories at once. You can also ask ChatGPT directly within any conversation: "What do you know about me?" or "Forget that I work at X."
An important caveat: if you delete a memory but the conversation where it was originally mentioned still exists in your history, ChatGPT may re-learn that fact from the conversation history reference system. To fully and permanently remove a piece of information, you should delete both the memory entry and the originating conversation.
Temporary Chat
Temporary Chat is a privacy mode that runs a session completely isolated from your memory systems. When you start a Temporary Chat:
- ChatGPT does not read your saved memories for that session
- ChatGPT does not write new memories from that session
- The conversation is not stored in your history (by default)
- The conversation history reference system cannot access it
This is useful for sensitive topics you do not want persisted, one-off tasks with a different persona, or testing how ChatGPT behaves without your personal context loaded. Access it via the model selector dropdown or the settings menu.
Plan Availability
Saved Memories are available on all plans, including Free. Conversation History Reference was rolled out to all plans in 2025. The auto-prioritisation feature (smarter memory weighting) is available on Plus and Pro. All users can access Temporary Chat regardless of plan.
Checklist
- What is the difference between Saved Memories and Conversation History Reference?
- How do you manually trigger ChatGPT to save a specific fact?
- Why might deleting a memory entry not fully remove that information from ChatGPT's context?
- What does Temporary Chat prevent in terms of memory read and write?
- On which plans does auto-prioritisation of saved memories apply?