Turn Off ChatGPT Memory and Choose What Work Carries Forward

To turn off ChatGPT memory, open Settings, choose Personalization, then Memory, and disable the memory controls available to your account. Turning memory off and deleting information are separate actions, so choose whether you want a pause in personalization or removal of something already stored.

For work, you can still carry forward a few approved instructions deliberately. fenbs keeps project rules and notes that you choose to record outside personal ChatGPT memory. That lets you stop relying on remembered preferences without starting every project explanation from scratch.

Which settings should you change first?

1. Open the account and workspace you actually use for the conversation. Check this before changing settings if you switch between personal and work accounts.
2. Open Settings, then Personalization, then Memory. Read the labels shown in your account rather than following a screenshot from a different interface.
3. Turn off the available Memory control. If your account shows separate saved-memory and chat-history reference controls, review both so the remaining personalization matches your intention.
4. Close settings, start a new ordinary conversation, and give it only the context required for that task. Keep the settings change separate from any later decision to delete information.

OpenAI documents several memory interfaces, so the labels can differ by account. Use its current memory settings guide↗ (opens in a new tab) to match the controls you see. If a work account does not expose a setting, ask the workspace administrator what is available.

Write down the outcome you wanted before making more changes: no personal references in new answers, no new remembered details, or removal of an old fact. Those goals can require different actions. A short goal keeps you from deleting useful project material when all you wanted was a less personalized drafting session.

What is the difference between off, delete, and training controls?

OpenAI treats model improvement as a separate preference. Review Data controls in ChatGPT↗ (opens in a new tab) if training is your concern. Make each choice on its own merits: whether future answers use personal context, whether a record remains stored, and whether eligible content can help improve models.

For example, a freelance designer may want neutral ideas for a new client while keeping earlier design discussions available to read. Another person may need to remove an outdated personal detail completely. The same sequence of clicks is not necessarily appropriate for both people.

Do you need to remove anything already saved?

Decide that separately after you have disabled the controls you intended to disable. Review the settings description before selecting a combined action with a name such as Delete and turn off memory. A control that changes use and a control that deletes a record have different consequences.

If removal is the goal, make a small source checklist: the remembered item, the conversation where you supplied it, and any file or connected source that also contains it. Keep this review narrow. You do not need to erase unrelated work just because a personal preference has become outdated.

For the adjacent questions, see how memory works, removing remembered information, a full-memory warning, long-term recall, or missing recall.

If you later turn memory back on, review your intentions again rather than treating the earlier switch as permanent cleanup. Keep the decision to resume personalization deliberate, especially when you move between unrelated clients or between personal questions and work.

Can you make a single conversation less personalized?

For a separate conversation, start Temporary Chat and choose Unpersonalized before the first message if that option is available. OpenAI distinguishes personalized and unpersonalized temporary conversations in its Temporary Chat guide↗ (opens in a new tab). Read the choice at the start instead of assuming the word temporary means every personalization source is excluded.

This is useful for an independent review. Say you are comparing two website concepts and want an answer based only on the supplied criteria. Give the criteria explicitly, ask the assistant to identify missing information, and avoid mentioning earlier preferences unless they are relevant to the comparison.

Also check any direct guidance you have supplied. OpenAI provides separate controls for custom instructions↗ (opens in a new tab). A remembered preference and an instruction you deliberately entered can produce similar wording, so an answer that sounds familiar is not a reliable settings test by itself.

What work context should you provide after switching memory off?

Supply a compact task brief: the goal, the audience, the approved constraints, and the specific output you need. Add the source material required to check the answer. Keep a current approval separate from a suggestion that is still being discussed.

• Goal: draft the introduction for a client proposal.
• Approved rule: use the client’s product names exactly as supplied.
• Current source: the approved outline and the latest product sheet.
• Open question: the client has not approved a delivery date.
• Next step: produce a draft that leaves the delivery date unresolved.

This small record makes review easier because you can compare the output with visible instructions. If work belongs in a ChatGPT project, our project memory guide covers that separate scope; if you change apps, the ChatGPT alternatives guide considers a different choice.

Put the lasting rule and the unresolved decision in fenbs Decisions and rules when they need to survive beyond this conversation. You can then provide project context intentionally while leaving personal memory disabled.

How to do this with fenbs

1. Create a free account and add the project whose context you need to keep.
2. Write the instructions you approved in Decisions and rules. Record who decided them and their scope so one client’s rule is not applied to unrelated work.
3. Add a short AI context note containing the current goal and relevant source references. Leave out personal details the project does not require.
4. Connect a supported assistant using the guide for that app. Ask it to read the project context before working and save progress before stopping.

fenbs is free for one person, no credit card. It does not import ChatGPT memories or erase information from ChatGPT. It is not the right fit if your only goal is to change a personal preference setting; use the ChatGPT controls for that.

How can you keep the choice deliberate over time?

At the next session, supply or read the current project record before drafting. Ask the assistant to state the constraints it is using and flag missing approvals. That check tests whether the work has enough context, without inviting it to invent what it thinks you previously wanted.

When a rule changes, replace it through a recorded human decision and update the next step. Keep this routine small enough to repeat: read, work, check, save. The broader AI memory topic hub covers other ways context can persist when you need a different arrangement.