ChatGPT Memory can help the assistant personalize future conversations by retaining useful details, such as preferences, goals, or information you have asked it to remember. Instead of repeating the same background in every chat, you may be able to let ChatGPT use saved details or relevant information from earlier conversations. The exact behavior depends on the memory features available to your account and the controls you have enabled.

The key distinction is that saved memories and chat history are not the same thing. A saved memory is a detail kept for personalization. Chat history is the record of conversations in your account, and some experiences can reference information from past chats without every detail becoming a separately saved memory. Deleting a chat, deleting a memory, turning memory off, and using Temporary Chat are different actions.

This guide explains the main concepts and shows how to review and manage memory. OpenAI changes these features over time, and availability can vary by plan, region, or workspace, so use the current controls and official documentation shown in your own ChatGPT account.

What is ChatGPT Memory?

Memory is a personalization feature designed to help ChatGPT use relevant information across conversations. Depending on your settings and experience, this may include explicit details you asked it to remember—such as a preferred writing style—or useful context that can make future answers more relevant.

For example, if you regularly ask for explanations aimed at beginners, a remembered preference could help ChatGPT choose a suitable level of detail. If you are working on a long-running project, relevant background may reduce the need to restate your goals in every new conversation. Memory is not a guarantee that every detail will be retained or recalled correctly, and you should correct the assistant when it gets something wrong.

Memory should not be treated as a complete, authoritative record. It can omit details, retain information that is no longer relevant, or apply a remembered preference to a situation where it does not fit. For important work, include the essential requirements in the current prompt rather than relying entirely on prior context.

Memory also differs from the model’s general learned knowledge. It is intended to personalize responses using information associated with your ChatGPT experience; it does not mean the assistant has a perfect record of every conversation or that your personal details have become universally available to other users.

Saved memories versus chat history

The distinction matters when you want to understand what ChatGPT may use later and how to remove it.

Saved memories are specific details that ChatGPT keeps for personalization. Depending on the interface, you may be able to ask ChatGPT what it remembers, review the list in settings, and remove individual items. A saved memory can remain available separately from the conversation where you originally mentioned it.

Reference chat history, where available, allows ChatGPT to use relevant information from previous conversations to personalize responses, even if a detail was not saved as a separate memory. It does not mean every message is stored as a precise memory, and it may not recall every detail from every chat.

Chat history is the collection of conversations shown in your account. Keeping a conversation in history does not automatically mean every sentence from it has become a saved memory. Conversely, deleting the conversation where you mentioned a detail may not remove a separately saved memory based on that conversation.

OpenAI’s memory experience and labels have evolved, so the available toggles may differ across accounts. Check the current Memory section in Settings and consult the official Memory FAQ for the latest distinction between saved memories and reference to past chats.

How to see what ChatGPT remembers

A simple way to inspect memory is to ask ChatGPT directly, “What do you remember about me?” The response can help you identify details that are influencing personalization, but it should not be treated as a complete audit log. For a more direct review, use the memory controls in settings if they are available.

On the web or mobile app, the usual route is:

  1. Open your profile or account menu.
  2. Select Settings.
  3. Open Personalization or the section containing Memory controls.
  4. Review the memory options and any available list of saved memories.
  5. Remove details that are incorrect, outdated, or no longer useful.

The exact labels and menu placement can change. If you cannot find a memory list or a particular control, consult the current OpenAI help article rather than assuming your account has the same interface as another user.

When checking memory, look for details that are too specific, no longer true, or inappropriate to use across different tasks. A preference such as “use concise explanations” may be useful. A temporary project detail or a one-time question may not need to influence future chats. If ChatGPT repeatedly applies a remembered detail incorrectly, remove or correct it and state the relevant context explicitly in your next prompt.

How to add or correct a memory

You can tell ChatGPT directly when a detail is useful for future conversations. For example:

  • “Remember that I prefer metric units.”
  • “Remember that this project uses Jekyll and Cloudflare Pages.”
  • “Forget the old project deadline; it has changed.”
  • “Do not use that preference for this conversation.”

Whether a specific instruction is saved and how it is represented depends on the memory features available to your account. After asking ChatGPT to remember something, you can ask what it remembers or inspect the settings to confirm the result. Do not assume that a request succeeded without checking when the detail is important.

Prefer durable, useful context over a long list of one-off facts. A good memory helps future answers without creating confusion. If a project has multiple stages, record the stable goal and current constraints, then include the latest task-specific details in the current conversation.

You should also correct mistakes promptly. If the assistant assumes an old preference still applies, say so explicitly. Memory is meant to help, not to override your current instructions. A fresh instruction in the current chat should guide the immediate task when it differs from an older preference.

How to delete a saved memory completely

Deleting a memory and deleting the conversation where it appeared are separate actions. If you want a detail removed from personalization, use the memory controls and delete that specific saved memory. If you also want the original conversation removed from your history, delete that chat separately.

OpenAI notes that fully removing information that ChatGPT remembers may require deleting the saved memory and the chats where you originally shared or discussed it. Depending on the feature, you may also need to check files, connected apps, or other places where the information was included. Consult the current Memory FAQ for the provider’s exact retention explanation.

A practical sequence is:

  1. Open Settings → Personalization → Manage memories, if that control is available.
  2. Delete the specific memory you no longer want retained for personalization.
  3. Find and delete the original conversation if you also want it removed from chat history.
  4. Review relevant shared links or connected services separately.
  5. Ask ChatGPT about the detail again in a new chat if you want to check whether it still appears to be remembered.

This final check is useful, but it is not a formal guarantee that every copy of information has been erased from every system. Follow the official data-control instructions for the exact deletion and retention behavior.

What happens when you turn Memory off?

Turning off memory features changes personalization behavior; it is not the same as deleting every saved memory or clearing your chat history. The exact effect depends on which memory control you switch off. For example, saved memories and reference to chat history may be controlled separately in some versions of the product.

Before changing a setting, read the explanation shown next to it. If you turn off a reference feature, ChatGPT may stop using that source of context in future responses, but that action should not be assumed to erase the information already stored. If your goal is to remove details, use the available deletion controls as well.

You may prefer to turn memory off if you want less personalization between conversations or if you are sharing a device. You may prefer to leave it on if recurring preferences and project context save time. Neither choice is universally correct; use the controls that fit your needs and the sensitivity of the information involved.

Memory is also distinct from model-improvement settings. Turning off personalization does not automatically mean your conversations are excluded from model improvement, and turning off model improvement does not necessarily delete saved memories. Review the separate Data controls section if your concern is how personal-account conversations may be used to improve models.

Is Memory the same as training ChatGPT?

No. Personalization and model improvement are separate concepts. Memory helps tailor responses for your ChatGPT experience by using details associated with your account. Model improvement concerns whether content may be used to improve the underlying models under the applicable product settings and terms.

For personal ChatGPT workspaces, OpenAI provides a setting called Improve the model for everyone. Its effect is separate from whether Memory is enabled. Turning off model improvement does not automatically delete conversations or saved memories, and deleting a memory does not by itself change the model-improvement setting.

If you are reviewing privacy, check both areas: Personalization/Memory for what ChatGPT may use to tailor future responses, and Data controls for model-improvement preferences and other data-management options. Temporary Chat also behaves differently from a normal conversation, but it should not be treated as a guarantee that no data is processed or retained for any purpose.

For a broader explanation, read our guide to ChatGPT conversation privacy.

How Temporary Chat interacts with Memory

Temporary Chat is designed for conversations that you do not want to appear in your normal chat history. OpenAI says Temporary Chats do not create or use memories for personalization in the standard way and are not used to improve its models, subject to the product’s stated conditions. OpenAI may retain a copy for a limited period for safety purposes, so “temporary” does not mean zero processing or instant deletion.

The available behavior can depend on personalization choices in the interface. Review the current Temporary Chat help page before using it for a particular privacy requirement. If you want to avoid having a detail carried into future conversations, do not rely only on the chat being absent from history; review your memory settings as well.

Temporary Chat is useful for one-off questions where persistent personalization is unnecessary. It is not a safe place for passwords, access tokens, confidential client records, or other information that should not be disclosed to an online service. Minimize the information you share regardless of which chat mode you choose.

What if ChatGPT remembers something incorrectly?

If ChatGPT refers to an outdated project, an incorrect preference, or a detail you do not want it to use, correct it directly. Ask it to forget the specific memory if appropriate, and then check the memory settings. If the detail came from a past conversation rather than a saved memory, review the reference-to-chat-history controls available in your account.

It can also help to provide a clear boundary in the current prompt: “For this task, ignore the old preference and use the instructions below.” Current context is especially important for time-sensitive details such as prices, deadlines, software versions, and project status.

If you are working on something consequential, keep a short, current project brief that you can paste into the conversation. That is more reliable than assuming the assistant will reconstruct every past decision perfectly. Memory can reduce repetition, but explicit context remains the best way to communicate critical requirements.

A practical Memory checklist

  • Review saved details: remove information that is outdated, incorrect, or unnecessary.
  • Separate memory from history: decide whether you want to remove a saved memory, delete a chat, or both.
  • Check each toggle: saved memories, reference to chat history, and model-improvement controls are not interchangeable.
  • Use explicit instructions for critical work: restate important constraints in the current conversation.
  • Use Temporary Chat selectively: understand its retention and personalization behavior before relying on it.
  • Check shared content separately: shared links and information sent to third-party tools have their own considerations.
  • Revisit settings after product changes: labels and features can evolve over time.

Final answer: should you use ChatGPT Memory?

Memory is useful when repeated context improves your experience: preferred units, writing style, recurring goals, or stable project information. It can reduce repetitive setup and make ongoing work feel more consistent. It is less useful when details change frequently, when you need a completely isolated conversation, or when the information is too sensitive to share with an online assistant.

The most important habit is to keep control of what is remembered. Review the available settings, correct mistakes, delete saved details you no longer need, and understand that memory, chat history, Temporary Chat, and model-improvement controls serve different purposes. For important tasks, state the essential facts and constraints directly instead of depending on memory alone.

Official sources