ChatGPT Memory Not Working? 9 Checks to Try (2026)

ChatGPT memory not working can mean two different things: a detail was never saved, or ChatGPT saved it but did not use it in a later reply. Those problems have different causes, so start by checking which kind of memory you expected to work before deleting chats, changing account settings or reinstalling the app.

This guide follows the current OpenAI Help Centre documentation checked on 4 November 2026. The names and availability of Memory controls can differ by plan, region, device and workspace. If a setting described below is missing from your account, that does not automatically mean your account is broken.

ChatGPT memory not working: identify what is failing

ChatGPT can use several kinds of context, and they are easy to mix up. The current conversation has its own context. Saved memories are details retained for future personalisation. Reference chat history lets ChatGPT use relevant information from previous conversations. Custom Instructions are direct guidance you provide, while a Project can have its own instructions, files and memory boundaries.

A problem with one of these features can look like a problem with all of them. For example, ChatGPT may remember a writing preference in a normal chat but not use it in a Project that has project-only memory. It may also use a saved memory without displaying every detail in the Memory summary. OpenAI explains these differences in its official Memory in ChatGPT guide.

For a broader introduction to the feature, see our guide to whether ChatGPT can remember previous conversations.

Run a clean, 60-second test before changing anything

Use a harmless preference that is easy to recognise. In a regular ChatGPT conversation, write: “Remember that I prefer concise answers in British English.” If ChatGPT confirms that it has saved the detail, open a new regular chat and ask: “What answer style did I ask you to remember?” Do not use a Temporary Chat for this test.

If the preference appears in the new chat, Memory is working at least for that simple saved detail. The original problem may be that ChatGPT did not consider another piece of information relevant, that it came from a different source, or that the chat was isolated by a Project setting. If the preference is not recalled, continue with the checks below. Avoid using medical, financial or other sensitive information as a test.

Nine checks to fix ChatGPT memory problems

1. Confirm that Memory is available and switched on

Open ChatGPT, select your account menu, then go to Settings and Personalization. Open Memory and check the controls shown for your account. Depending on the version available to you, you may see a single Memory switch, Reference saved memories, Reference chat history, a Memory summary, Manage or Saved memories.

If Memory is switched off, turn on the relevant control if you want ChatGPT to personalise future replies. If the setting is missing, check the same account in the web version and make sure the mobile app is up to date. OpenAI says Memory controls can vary by plan, region, platform and workspace settings, so a control may not have reached every account. Do not assume that every account should display identical switches.

2. Check saved memories and chat-history reference separately

Saved memories and Reference chat history are related, but they are not the same setting. A saved memory is a specific detail retained for future use, such as a preference you explicitly asked ChatGPT to remember. Reference chat history allows ChatGPT to draw on useful information from previous conversations without turning every sentence into a permanent saved memory.

If your account shows separate controls, check the one that matches the behaviour you expect. If you want a saved preference to carry into new chats, check Reference saved memories. If your settings include both controls, note that turning off Reference saved memories also turns off Reference chat history; when Reference saved memories is on, you can manage Reference chat history separately. Some accounts can switch between Improved Memory and legacy Saved memories, so follow the options actually shown in your settings.

Turning off Reference chat history does not delete the original conversations. OpenAI says information derived from chat history is scheduled for deletion from its systems within 30 days after that setting is turned off, while the original chats remain unless you delete them. The current documentation also says there is no separate storage limit for what ChatGPT can reference through chat history. Do not assume that chat-history reference has simply run out of storage.

3. Make sure you are not testing in an unpersonalised Temporary Chat

Temporary Chat is designed not to create or update memories. If you ask ChatGPT to remember something inside a Temporary Chat, it will not become a new saved memory. This is expected behaviour, not a fault with your account.

There is an important distinction: before a Temporary Chat begins, ChatGPT may let you choose whether it uses existing memories, Custom Instructions and plugins. Select Unpersonalized if you do not want those sources used. That choice cannot be changed after the conversation has started. A Temporary Chat that is saved becomes a regular chat and then follows your account's personalisation settings. See OpenAI's Temporary Chat guidance and our ChatGPT Temporary Chat explainer for the details.

4. Check whether a Project is using project-only memory

If the problem happens only inside a Project, open that Project and check its settings. Projects can use default memory or project-only memory. With project-only memory, chats can draw on other conversations in the same Project but cannot refer to conversations outside it. Chats outside the Project cannot use its conversations as memory either.

Shared Projects automatically use project-only memory and cannot access an individual member's memories or context from outside the Project. This can explain why a preference works in a normal chat but seems to disappear when you open a shared workspace. Open the Project, select its more-options menu, choose Project settings, and look at the Memory setting. If you change the setting, OpenAI says the change may take a few hours to take effect.

OpenAI's Projects guide explains the difference. You can also read our ChatGPT Projects guide.

5. Check workspace or administrator restrictions

If you use ChatGPT through an employer, school or managed workspace, your organisation may control whether Memory is available. Workspace owners and administrators can manage supported Memory settings and role permissions. A personal setting may not override a restriction applied to the whole workspace or to your role.

Check whether you are signed into a personal account or an organisation account. If the Memory option is missing only in the managed workspace, ask the workspace administrator whether Memory is enabled for your role. Do not keep changing unrelated personal settings when the restriction is being applied by the organisation.

6. Ask ChatGPT to remember one clear, reusable detail

Memory is not a recording of every sentence you have ever written. OpenAI says ChatGPT decides which available information is relevant to a response, and the information it uses can change as your context changes. A long instruction such as “remember everything about my work” is therefore a poor diagnostic test.

Use one short, stable preference instead: “Remember that I prefer step-by-step explanations with metric measurements.” Then ask ChatGPT to confirm the detail. If your account has a Memory summary or saved-memory manager, check whether the preference appears there. If it does not, try a more specific wording and check that Memory is enabled. A clear request improves the test, but it does not turn Memory into a guaranteed transcript archive.

7. Inspect the Memory summary and available Sources

When the feature is available, the Memory summary gives you a high-level view of information that may be used for personalisation. It is not a complete list of every detail or source that ChatGPT can reference. A missing item in the summary does not always prove that ChatGPT has no access to relevant context.

When ChatGPT uses personal context, a Sources area may appear below a response. It can help you identify whether a past chat, saved memory, Custom Instruction, file or connected app contributed to the reply. Sources may not show every factor. If the information is wrong or outdated, correct the memory or summary where those controls are available, and check the original source as well.

If your goal is to remove information rather than restore it, deleting one chat may not be enough. OpenAI says a separate saved memory and other sources containing the same information may also need to be removed. Follow the current Memory management instructions rather than assuming that deleting a conversation clears every copy.

8. Do not confuse Memory with Custom Instructions or model-training controls

Custom Instructions are direct guidance that you provide about what ChatGPT should know and how it should respond. They are different from Memory, which can use relevant information that develops across conversations and other available sources. If you want a consistent response format, tone or set of preferences in every chat, Custom Instructions may be the more suitable place for that guidance.

Check our Custom Instructions guide if the issue is that ChatGPT ignores a standing instruction rather than forgetting a personal detail. OpenAI also provides an official Custom Instructions help article.

The Improve the model for everyone option in Data controls is separate again. It controls whether eligible conversations can be used to improve OpenAI models; it is not the Memory switch. Turning model training off does not, by itself, disable Memory or delete your chat history. If your only change was to that training control, check Personalization and Memory separately.

9. Rule out an app or service problem after checking the settings

If the Memory page will not load, a switch will not stay on, or the feature returns an error, check the OpenAI service status page for a current incident. Then update the ChatGPT app or reload the browser and try the same simple test in a new regular chat.

If the problem appears only in one browser, try another supported browser. If it appears only in the mobile app, compare it with the web version. Signing out and back in can help when the account interface has become stuck, but clearing browser data or reinstalling the app should not be your first step for an account-level Memory problem. Do not delete conversations or saved memories as a troubleshooting shortcut.

What ChatGPT Memory can and cannot do

Memory is intended to make future conversations more useful, not to provide a perfect searchable record of your entire account. It may retain a preference, use a relevant detail from a past chat or draw on another available source, but it does not necessarily bring every fact into every answer. ChatGPT looks for context when it is likely to improve a response, rather than searching every past conversation for every prompt.

The Memory summary is also a summary, not a full audit log. It may omit details or sources that are less relevant to the summary. This is why a response can be personalised even when you cannot find the exact detail in the summary, and why a detail you mentioned once may not appear in an unrelated conversation.

For information you must keep accurate and easy to retrieve, use a Project with the relevant files and instructions, keep a separate note, or add a clear Custom Instruction. Use saved Memory for stable preferences and context that make future chats easier. Use the current conversation for details that matter only to the task in front of you. Choosing the right place is more reliable than expecting one feature to behave like a complete database.

When to contact OpenAI Support

Contact OpenAI Support if the Memory control is missing when it should be available, the setting repeatedly switches itself off, the simple test fails across the web and mobile app, or you see a persistent error after checking workspace restrictions. Include the device and operating system, browser or app version, whether you are using a personal or managed account, which Memory controls are visible, and the steps that reproduce the problem.

Add the approximate time the problem occurred and a screenshot of the error if useful, but redact email addresses, private conversation content and other personal information. Never send your password, one-time sign-in code or recovery details. A short, reproducible example is more useful than a large export of unrelated conversations.

Frequently asked questions

Why does ChatGPT remember some things but not others?

Memory is selective. ChatGPT uses information it considers relevant, and it does not retain every detail from every conversation. A clear saved preference is a better test than expecting it to recall a minor detail from an unrelated chat.

Does turning off Reference chat history turn off saved memories?

No. OpenAI documents Reference chat history and saved memories as separate sources. Turning off Reference chat history does not delete saved memories. Check the controls shown in your account because some versions combine or label the settings differently.

Does Temporary Chat use Memory?

A Temporary Chat does not create or update memories. Before it starts, you may choose whether it can use existing memories, Custom Instructions and plugins. Selecting Unpersonalized prevents those sources from being used for that temporary conversation.

Does turning off Improve the model for everyone disable Memory?

No. Model-training controls and Memory controls are separate. Improve the model for everyone is in Data controls; Memory is managed through Personalization. Changing one does not automatically change the other.

Can I make ChatGPT remember something exactly?

You can explicitly ask ChatGPT to remember a short, useful detail and review the available Memory controls, but Memory is not a verbatim archive. Keep important records in a document or Project if you need a dependable source you can check yourself.

Final checklist

If ChatGPT still seems to forget information, check these points in order: Memory is available and enabled; the relevant saved-memory and chat-history controls are on; you tested in a regular chat; the conversation is not isolated by project-only memory; your workspace administrator allows the feature; and you used one clear, harmless test detail. Then review the Memory summary or Sources and check OpenAI Status before contacting support.

The key is to identify which source of context you expected ChatGPT to use. Once you separate saved memories, chat-history reference, Temporary Chat, Projects and Custom Instructions, most apparent Memory failures become easier to diagnose without deleting useful conversations.