ChatGPT Memory at Work: What It Remembers and How to Turn It Off

What ChatGPT Memory remembers

OpenAI's help article Memory in ChatGPT says that when Memory is on, ChatGPT can remember relevant preferences and details "from your chats and other available sources". Depending on plan and region, those sources can include:

• Past chats (through Reference chat history).
• Saved memories: details you asked it to remember, or that it saved as useful context.
• Custom instructions.
• Files in Library.
• Content from connected apps, such as Gmail.

OpenAI also says Memory does not retain every detail from every conversation, and that features and controls vary by plan, region, platform and workspace settings. When personal context shapes an answer, a Sources section may appear under the response showing which chat, memory, file or email was used.

How a staff member turns Memory off or deletes it

Each user controls Memory in their own settings. The steps from OpenAI's help article:

1. 1

   Open the Memory settings

   Go to Settings, then Personalization, then Memory. Available controls may include Memory, Reference saved memories, Reference chat history, Memory summary and Manage.

2. 2

   Review what it holds

   Open the memory summary or saved memories list, or ask ChatGPT what it remembers about you.

3. 3

   Delete specific items

   Delete the saved memory and the chat where the information was first shared. Also remove it from Library files and disconnect apps that contain it.

4. 4

   Turn Memory off if needed

   Use Delete and turn off memory to remove the remembered information shown in the summary and switch Memory off. Past chats stay unless you delete them.

Two details catch people out. Deleting a chat does not automatically delete a saved memory created from it, and turning off Memory does not delete past chats. OpenAI says it may keep logs of deleted saved memories for up to 30 days for safety and debugging, and that information remembered from chat history is scheduled for deletion within 30 days when Reference chat history is turned off.

Workspace controls in ChatGPT Business

OpenAI's Memory FAQ for ChatGPT Business says memory and personalisation controls exist at both the user and the workspace level.

For ChatGPT for Healthcare and ChatGPT Enterprise with Regulated Workspace, OpenAI says improved memory is disabled by default, and owners and admins choose whether to make it available. OpenAI's enterprise privacy page covers the wider data commitments for business plans.

When remembered client data becomes a PDPA issue

The PDPA applies to how your organisation collects, uses, discloses and protects personal data, whichever tool is involved. The PDPC's Advisory Guidelines on Use of Personal Data in Generative AI treat organisations that use AI tools as deployers with primary responsibility. Our summary of PDPC's generative AI guidelines covers them in detail.

Memory raises three practical risks for a small business:

• Purpose: a client's details pasted to draft one email can resurface in an unrelated answer later, possibly in front of a colleague.
• Retention: data you would normally delete after a job can persist in saved memories and chat history until someone deletes both.
• Personal accounts: if staff use their own free or Plus accounts for work, your company has no workspace controls, and the data sits outside your policies.

A simple company rule for ChatGPT Memory

You do not need a long policy. Most small teams can agree on a few rules and check them once a quarter.

Our ChatGPT at work policy guide has a fuller one-page policy, and AI training opt-out settings for every tool covers the related setting that controls whether chats are used to train models.

Do it yourself or get help?

The settings themselves take minutes. What protects client data is staff knowing what not to paste and remembering to clear memory, which is a training and habit problem more than a technical one.

IT Sorted is a Singapore-based team of developers with over 10 years of experience. Our Security Awareness Training (S$390) sets up a 12-month plan of 5-minute lessons in your own account, with Singapore-specific topics and completion tracking, so you can include safe AI use alongside scams and phishing. Platform licence fees are paid to the vendor.