ChatGPT Settings: Check Memory, Instructions, and Data Use

ChatGPT settings for work need three separate checks: what personalizes responses, what instructions you provide, and how your content may be used. Open Settings in the account and workspace you actually use, then review Personalization and Data controls before adding work material.

Keep approved project rules in an explicit record as well. Tools like fenbs hold those rules outside personal account preferences, so a colleague or another connected assistant can read the same instruction. A preference for short answers and a requirement for a manager to approve a client promise have different jobs.

What should you check before changing anything?

1. Confirm your account and active workspace. Ask your organization which account is approved for the material you plan to use.
2. Open Personalization and review the memory and custom-instruction controls available to you.
3. Open Data controls and check your model-improvement preferences separately from personalization.
4. Try a harmless example in a new conversation, then record any work rules the test revealed you need.

Write down the setting names you actually see. OpenAI explains that available controls depend on your account and workspace in its consumer data guidance↗ (opens in a new tab). If a work account has a different screen from your personal one, ask the workspace administrator before treating the difference as an error.

Use a fictional project for the first test. For example, ask for a short agenda about choosing office furniture. You can evaluate tone, length, and remembered preferences without exposing a real employee discussion. Our guide to using AI at work covers choosing an appropriate first task.

What belongs in custom instructions?

Custom instructions are guidance you deliberately supply about how ChatGPT should respond. On the web, OpenAI places the controls under Settings and Personalization; its custom instructions guide↗ (opens in a new tab) gives the platform-specific steps. Use that page when the labels on your device differ.

Keep broadly useful preferences here: American spelling, plain language, or a request to identify assumptions. Avoid placing a temporary client deadline in general instructions. Otherwise an unrelated conversation may receive guidance you intended for one assignment, and you may forget where the instruction came from.

• Useful general preference: Explain unfamiliar terms in a short sentence.
• Useful response preference: Put the proposed next step before supporting detail.
• Project-specific rule to keep with the project: A person must approve changes to the delivery date.

Test instructions with both a normal request and an exception. If you usually want concise answers, ask for a detailed checklist and inspect the result. Good instructions help express your preference; you still need to review the actual response. Our work prompt examples show how to state the task itself.

How do you review memory without assuming it is complete?

Under Personalization, inspect the Memory controls your account offers. OpenAI’s memory documentation↗ (opens in a new tab) explains that available experiences differ, and a memory summary does not show everything that might influence personalization. Review the displayed information, correct stale details, and check relevant sources when available.

Treat a remembered detail as something to inspect. Say you once discussed a draft launch date and later changed it. Ask which date the answer is using, then supply the approved date directly. A fluent response is not evidence that it found the latest approval.

Deleting a conversation does not necessarily remove a separate saved memory. Follow OpenAI’s removal guidance for the sources involved rather than assuming a single deletion clears every reference. When the fact is a work decision, store the approved answer and who chose it with the project; fenbs explains that shared record.

How are data controls different from personalization?

Data controls govern choices such as model-improvement preferences. Personalization concerns information and guidance used to shape responses. Review both: changing the way an answer sounds does not establish how the content is handled, and changing a data-use preference does not approve a document for upload.

OpenAI’s consumer data guidance distinguishes personal accounts from managed workspaces. In a work setting, check your organization’s policy and the controls your administrator provides. Do not paste confidential material into a personal account simply because you changed one toggle. Ask the document owner what use is permitted when you are uncertain.

Keep the practical check small: account, workspace, content classification, approved task, and reviewer. For a public meeting agenda this may take a moment. For employee, customer, or other restricted material, wait for the organization’s answer before proceeding. Settings are one part of that decision.

When should you inspect Temporary Chat options?

Check the options before starting a conversation you intend to keep temporary. OpenAI’s Temporary Chat guide↗ (opens in a new tab) distinguishes personalized and unpersonalized conversations: the personalized option can use existing memory, instructions, and plugins. Neither creates or updates memories while temporary.

A temporary chat stays out of history and model improvement while it remains temporary; saving it converts it into a regular chat subject to the applicable settings. Temporary does not mean that no copy can be retained. Check the official guidance and your workspace policy before choosing it for work.

For a clean writing-style test, inspect the unpersonalized option and use invented text. Compare that result with your usual conversation to see whether your preferences help. Avoid changing several settings at once, because you will have trouble identifying which change affected the answer.

How to do this with fenbs

1. Create a free account, then add one project whose instructions need to survive a new conversation.
2. Record the approved project rules in Decisions and rules, naming the person who decided them. Leave unanswered choices open for a person.
3. Follow the integration guide for a supported assistant and approve the access you intend to grant.
4. Ask the connected assistant to read the project context before working and save progress when it stops, including outstanding review questions.

fenbs is free for one person, no credit card. It is not the right fit if your only goal is to change an account preference; do that in ChatGPT. Its purpose here is to preserve the work rules and progress that another supported assistant also needs.

After setup, ask a fresh session to explain the required review before it drafts anything. Check that explanation against the rule a person approved. For the next step after configuration, use our beginner workflow; the AI at work hub covers other practical tasks.