ChatGPT Memory vs Chat History vs Model Training: What Each Setting Actually Does
Chat history, Memory and model training are separate ChatGPT controls. Turning one off does not automatically delete or disable the others—and Temporary Chat follows different rules.
Information checked on:
Key takeaways
- Chat history and Memory are separate: turning Memory off does not delete past chats. [1][3][5]
- Turning Improve the model for everyone off does not disable Memory or remove conversations from history. [1][3][4]
- Deleting a chat does not necessarily delete a separate saved memory or a file retained in Library. [1][5][6]
- Personalized Temporary Chat can use existing personalization; neither temporary mode creates or updates memories while the chat remains temporary. [1–3]
- Complete removal can require deleting information from Memory, chats, Library and connected sources separately. [1][6][7]
ChatGPT Memory vs chat history vs model training at a glance
These controls answer different questions. Chat history determines whether a regular conversation remains available in the account. Memory determines whether supported personal context can shape later responses. Improve the model for everyone determines whether eligible new content may be used to improve OpenAI models. [1][3–5]
| Setting or action | Chat remains in history? | Can it personalize future chats? | Eligible for model improvement? | Deletes existing data? |
|---|---|---|---|---|
| Memory on | Yes, unless separately deleted | Yes, depending on available sources and settings | Depends on the model-improvement setting | No |
| Memory off | Yes | Memory-based personalization stops or changes | Depends on the model-improvement setting | No |
| Improve the model for everyone off | Yes | Memory can still work | New eligible chats are excluded under the opt-out rule | No |
| Delete a chat | No | Removes that chat as a source, but separate memories or files can remain | A separate question from deletion | Chat is scheduled for deletion under OpenAI’s retention rules |
| Personalized Temporary Chat | No while temporary | Can use existing personalization but creates no new memories | No while temporary | No immediate guaranteed deletion |
| Unpersonalized Temporary Chat | No while temporary | Does not use normal personalization | No while temporary | No immediate guaranteed deletion |
Feature names and availability can differ by plan, region, platform and workspace policy. [1–3]
What is ChatGPT chat history?
Chat history is the account’s collection of regular saved conversations. Regular and archived chats remain saved until the user deletes them or an applicable workspace retention policy removes them. Archiving changes where a chat appears; it does not delete it or shorten its retention period. [3][5]
Deleting a saved chat removes it from the account view immediately. OpenAI then schedules it for permanent deletion within 30 days, unless it was already de-identified and disassociated from the account or must be retained longer for security or legal obligations. Deleted chats cannot normally be restored. [5]
That storage and retention function is not the same as Memory.
What is ChatGPT Memory in 2026?
Memory is the personalization layer that can use relevant preferences and context in later responses. Depending on the account and available features, its sources can include past chats, saved memories, custom instructions, Library files and connected apps such as Gmail. It does not retain every detail from every conversation, and the context it considers useful can change. [1]
The current route is Settings → Personalization → Memory. An account may show controls such as Memory, Reference saved memories, Reference chat history, Memory summary, Manage, or Saved memories. Not every account exposes every control. [1]
A Memory summary is a high-level view, not a complete audit log. Don’t mention this again, where available, changes future personalization but does not delete the underlying source. Delete and turn off memory removes the remembered information represented in the summary and turns Memory off; it does not delete past chats. [1]
Improved Memory vs legacy saved memories
OpenAI describes Improved Memory as an evolving broader summary of relevant context. The legacy saved memories experience uses a more explicit list of individual items. Where the choice is available, users can switch between these experiences under Memory settings. [1]
The distinction is about how personalization context is maintained—not whether chat history exists. Changing the Memory experience does not delete conversations. Availability varies by plan, platform and workspace. [1]
What is Reference chat history?
If the account includes Reference chat history, turning it on lets ChatGPT use relevant information from earlier conversations to personalize later responses. Information derived this way can change as ChatGPT updates what is useful to remember. [1]
Turning Reference chat history off schedules information remembered from past chats for deletion from OpenAI systems within 30 days. The original conversations remain in chat history unless they are deleted separately, and separate saved memories remain. If Reference saved memories is available, turning it off also turns off Reference chat history. [1]
Does turning Memory off delete your chats?
No. Turning Memory off changes personalization; it does not delete regular or archived conversations. If Memory is later enabled again, chats that still remain in history may contribute to newly remembered information under the settings then in effect. [1]
That does not mean a previously deleted saved memory automatically returns. It means the surviving source chats can again be available as inputs to Memory where the account supports that behavior.
Does deleting a chat delete what ChatGPT remembers?
Not necessarily. Saved memories are stored separately from chat history, so deleting the original conversation does not automatically delete a separate saved memory. A file retained in Library can also remain after its chat is deleted, and an enabled connected app may continue to provide relevant information. [1][6][7]
Deleting one chat therefore removes one source. It does not prove that the same information no longer exists in Memory, another conversation, a Library file or a connected service.
How do you actually remove something ChatGPT remembers?
The required action depends on where the information exists. For fuller removal, review each relevant source: [1]
- Delete or correct the item in the Memory summary or saved-memory controls.
- Delete regular or archived chats containing the information.
- Delete related files from Library.
- Disconnect an app if ChatGPT should no longer retrieve information from it.
Don’t mention this again is a personalization instruction, not source deletion. Similarly, deleting a saved memory stops that memory from being used in future personalization but does not remove mentions from old chats. OpenAI says Memory changes can take time to propagate and that logs of deleted saved memories may be retained for up to 30 days for safety and debugging. [1]
What does “Improve the model for everyone” actually do?
This is a model-improvement control under Settings → Data controls. For personal ChatGPT services, OpenAI may use eligible content to improve models when the setting is on. After the user turns it off, OpenAI says new conversations are not used for model improvement under the opt-out rule. [3][4]
The switch does not delete or hide saved chats, Library files or Memory. Those sources and personalization controls must be managed separately. [1][3][6]
Can Memory stay on while training is off?
Yes. Under current personal-account controls, Memory can continue personalizing responses while Improve the model for everyone is off. The model-improvement opt-out does not itself disable Memory. [1][3]
Can training stay on while Memory is off?
Yes. Turning Memory off stops or changes personalization, but it is not a training opt-out. If model improvement remains enabled, eligible content may still be used under OpenAI’s current data-use rules. [1][3][4]
Does training opt-out apply retroactively?
OpenAI’s current wording is forward-looking: after the user opts out, new conversations are not used to improve models. The setting is not documented as erasing or reversing previous model-improvement use, and it is not a historical exposure checker. [3][4]
Feedback exception
Even after opting out, deliberately submitting product feedback—such as thumbs-up or thumbs-down—can make the conversation associated with that feedback eligible for model improvement. [3][4]
Temporary Chat in 2026: Personalized vs Unpersonalized
Before starting a Temporary Chat, current ChatGPT can offer two choices: [2]
- Personalized can use existing memories, custom instructions and plugins.
- Unpersonalized does not use those normal personalization sources.
Neither mode creates or updates memories while the conversation remains temporary. Neither appears in normal chat history or is used to improve OpenAI models while temporary. OpenAI may still retain a copy for up to 30 days for safety, and limited safety-relevant context can be used in rare, high-risk situations. The personalization choice cannot currently be changed after the conversation begins. [1–3]
Temporary Chat therefore does not mean zero retention, and the old blanket description “Temporary Chat never uses Memory” is no longer complete: a Personalized Temporary Chat can use existing Memory without writing new Memory. [2]
What happens if you save a Temporary Chat?
Saving converts it into a regular chat. It then appears in history and follows the account’s normal personalization and model-improvement settings. If Memory is enabled, the saved conversation may help personalize future responses. Eligible uploaded files may also be saved to Library, subject to availability and storage limits. [2][6]
OpenAI does not say that saving makes the earlier temporary period retroactively become training data. The documented rule is that the saved chat follows regular settings from that point forward. [2]
Which setting should you use?
| Goal | Best control or action |
|---|---|
| Keep chats but stop Memory personalization | Turn Memory off |
| Keep Memory but stop new chats from model improvement | Turn off Improve the model for everyone |
| Remove one conversation | Delete the chat |
| Thoroughly remove remembered information | Remove it from Memory and every relevant source |
| One-off chat with no history, no new memory and no training, while keeping existing personalization | Personalized Temporary Chat |
| One-off chat without normal personalization | Unpersonalized Temporary Chat |
| Stop future connected-app access | Disconnect that app |
| Delete a saved uploaded file | Delete it from Library separately |
These actions remain subject to account availability, workspace controls and the retention rules for each source. [1–7]
Library files and Memory
Library is a separate file store. Where supported and enabled, relevant Library files may contribute to Memory. Deleting the conversation does not delete a file that remains saved in Library, so deleting a ChatGPT chat and deleting its Library file are separate actions. [1][5][6]
Temporary Chat uploads are not saved to Library while the conversation remains temporary. If the chat is saved as a regular chat, eligible files may then be saved, subject to availability and storage limits. [2][6]
Connected apps and Memory
Depending on plan, region, permissions and enabled integrations, eligible information from a connected app can contribute to Memory. Disconnecting an app prevents future access through that connection, but it does not delete past conversations or saved memories that already contain information from the app. [1][7]
OpenAI says it does not train generalized models directly on connected Google-app data or derivations of that data, subject to listed exceptions for feedback, information manually copied, pasted or uploaded, and information included in ChatGPT responses. [7]
Personal vs Business, Enterprise, Edu and Healthcare
OpenAI says content from ChatGPT Business, Enterprise, Edu and ChatGPT for Healthcare workspaces is not used to train models by default. Memory availability and retention can still depend on workspace settings and administrator policy. [1][3]
FAQ
Is ChatGPT Memory the same as chat history?
No. Chat history stores conversations. Memory is a separate personalization system that can use relevant context from available sources. [1][5]
Does turning Memory off delete my chats?
No. Regular and archived chats remain until separately deleted or removed by an applicable workspace retention policy. [1][5]
Does deleting a chat delete saved Memory?
Not automatically. A separate saved memory can remain after its source conversation is deleted. [1]
Does turning off model training disable Memory?
No. Improve the model for everyone and Memory are separate settings. [1][3]
Can Memory stay on when training is off?
Yes. Memory can continue personalizing responses while new eligible conversations are excluded from model improvement under the opt-out rule. [1][3]
Can training stay on when Memory is off?
Yes. Turning Memory off changes personalization; it does not change the separate model-improvement preference. [1][3][4]
What happens when I turn Reference chat history off?
If the control is available, information remembered from past chats is scheduled for deletion within 30 days. The original chats and separate saved memories remain unless removed separately. [1]
What is the difference between Improved Memory and saved memories?
Improved Memory maintains a broader evolving summary. The legacy saved-memories experience exposes more explicit individual items. Availability varies, and switching does not delete chat history. [1]
Does Temporary Chat use Memory?
It depends. Personalized Temporary Chat can use existing Memory; Unpersonalized Temporary Chat does not. Neither mode creates or updates memories while temporary. [2]
What is Personalized Temporary Chat?
It is a temporary conversation that can use existing memories, custom instructions and plugins. It stays outside normal history and model improvement while temporary and does not write new memories. [2]
Does Temporary Chat train OpenAI models?
Not while it remains temporary, whether Personalized or Unpersonalized. A saved Temporary Chat becomes regular and follows the account’s model-improvement setting. [2–4]
What happens if I save a Temporary Chat?
It becomes a regular chat, appears in history and follows normal personalization and model-improvement settings. [2]
How do I make ChatGPT fully forget something?
Remove it from Memory and from every relevant source where it remains, including chats, Library files and connected apps. Updates may take time to propagate. [1]
Can Library files affect Memory?
Yes, where Library and Memory are supported and enabled. Delete a retained Library copy separately from its conversation. [1][6]
Can connected apps affect Memory?
Yes, eligible connected-app information can contribute where supported. Disconnecting the app stops future access but does not erase old chats or saved memories. [1][7]
Does a training opt-out delete past data?
No. The opt-out controls model-improvement use of new eligible conversations. Delete stored content through the controls for the chat, Memory, file or connected source. [3][4]
Sources & Verification
Information checked against current official OpenAI documentation on October 9, 2026. This article is documentation-based and does not claim hands-on account testing. Feature names and availability can vary by plan, region, platform, workspace policy and administrator settings.
- OpenAI — Memory in ChatGPT — Memory sources and controls, Memory summary, Reference chat history, saved memories, Improved Memory, removal, Temporary Chat interaction and workspace defaults.
- OpenAI — Temporary chat in ChatGPT — Personalized and Unpersonalized modes, history, Memory, training, retention and conversion to a saved chat.
- OpenAI — Data controls in ChatGPT — Improve the model for everyone, saved-chat behavior, Temporary Chat, feedback and managed-workspace defaults.
- OpenAI — How your data is used to improve model performance — personal-service opt-out, feedback exception and business-service defaults.
- OpenAI — Chat and file retention in ChatGPT — regular, archived, deleted and temporary-chat retention plus separately saved Library files.
- OpenAI — Using Library to manage files in ChatGPT — Library separation, file deletion, Temporary Chat files and use under Memory/data controls.
- OpenAI — Google app data controls FAQ — connected Google-app Memory, disconnection effects and Google-specific model-training limitations.
