ChatGPT Memory Keeps Turning Off: What to Check

If ChatGPT memory keeps turning off, first check whether the visible setting actually changes in the same account and workspace. A response that forgets a detail is not proof that the switch turned off. Confirm the account, conversation type, and setting before changing anything else.

While you investigate, fenbs can hold the current project decisions and next steps outside native recall. A connected assistant can read that explicit work context, so your task does not depend on diagnosing a personalization setting first.

What should you check first?

Use a single harmless test and change one thing at a time. Keep a short note of the setting you see and the action after which it appears different. That gives you a repeatable observation instead of a collection of inconsistent conversations.

1. Confirm the active account and selected workspace. Do this again after any sign-in or account switch.
2. Open Settings, then Personalization and Memory. Record the exact controls visible in your account and their current state.
3. Close settings, reopen them, and check whether the visible state persists. Then repeat after a page refresh.
4. Start a regular conversation outside a project for a simple comparison. Avoid putting sensitive facts into a troubleshooting prompt.
5. If the same control repeatedly changes without your action, record the steps and contact Support. If the control stays enabled, investigate recall and conversation boundaries instead.

OpenAI’s Memory guide↗ (opens in a new tab) explains that controls vary by account availability and workspace settings. Use the controls actually presented to you, rather than assuming a screenshot from another account must match yours.

Are you viewing a different account or workspace?

Check the account menu instead of identifying the account by a familiar conversation title. Two accounts can contain similar work, and a copied prompt can make either one feel familiar. Write the active account privately in your test notes so you can compare like with like.

OpenAI says account switching keeps memory and settings separate↗ (opens in a new tab). A different setting after switching accounts is therefore not evidence that the original account changed itself.

For a managed workspace, ask the administrator whether the setting you need is available to your role. Describe the missing or disabled control precisely. Do not try to work around an organization’s restriction by moving its project information into an unrelated personal account.

Is the conversation using a different kind of context?

Compare the same harmless question in a regular conversation and the conversation where the problem appears. Keep your prompt identical. If you change both the chat type and the wording, the answers cannot tell you which change mattered.

Temporary chat has its own personalization choices. OpenAI’s Temporary chat documentation↗ (opens in a new tab) distinguishes using existing personalization from creating new memories: temporary conversations do not create or update memories while they remain temporary.

Projects add another boundary. OpenAI explains project-only memory↗ (opens in a new tab) separately from general personalization. Our guide to ChatGPT project memory covers that boundary; keep the deeper project question separate from whether an account switch persists.

Did a setting change, or did an answer miss the fact?

These are different failures with different next steps. If the visible control remains enabled, write down what the answer missed and where you expected the fact to come from. Was it an instruction in the current conversation, a remembered preference, or a decision buried in another discussion?

Do not repeatedly delete and recreate remembered information to test a switch. That changes the material being tested and makes a clean comparison harder. For the separate question of deliberate disabling, see ChatGPT memory off.

What if memory seems to turn back on?

Apply the same test in reverse. Confirm the visible control, account, and workspace before concluding that ChatGPT re-enabled it. Also distinguish familiar wording from personalization: information in your current prompt can shape an answer without proving that a remembered record was used.

If you can reproduce a visible setting change, follow OpenAI’s Support instructions↗ (opens in a new tab). Include the sequence, time and time zone, browser or app details, and a screenshot with private content removed. A report that names the control is more useful than “memory is broken.”

For ongoing work, give the assistant the current approved facts directly while you investigate. The AI memory hub explains other continuity options, and fenbs keeps explicit project context available without relying on personal recall.

How to do this with fenbs

1. Create a free account and add the project you need to keep moving.
2. Record the decisions and continuing rules a person has approved. Add only the background required for the current work.
3. Put the next actions into tasks and write Where we left off, including what is untested or waiting for a person.
4. Connect a supported assistant using its setup guide. Ask it to read the project context and state the next step before acting.
5. Review the result and ask it to save progress before the session ends. Keep your ChatGPT troubleshooting notes separate from the project’s actual progress.

fenbs is not the right fit if all you need is a working personal preference toggle. It is useful when other people or later AI sessions need a clear record of the work, regardless of what happens to that toggle.