ChatGPT memory for work: what to keep in it, and what not to

What to keep where

A worked example: a memory review

Illustration: a made-up business. An account manager opens her saved memories after three months and finds eleven. She sorts them:

• Keep (4): her role, short replies, British spelling, no emoji.
• Move to a project (4): two clients' price lists and two clients' contacts, each into that client's project.
• Move to custom instructions (2): "never confirm a delivery date" and "always add a 'check' line".
• Delete (1): a colleague's phone number it had picked up from a chat.

The two rules now read as proper instructions:

Can't: confirm a delivery date. Instead, say the operations team will confirm it within one working day.
Format: the draft, then one line starting "Check:" on anything to confirm.

After the move, a new chat about one client no longer brings in another client's prices.

Set it up

1. Open Settings > Personalization and look at both memory settings.
2. Review saved memories: keep preferences, move rules and client facts, delete anything about other people.
3. Put rules in custom instructions, client facts in projects, methods in skills.
4. Use Temporary Chat for anything you don't want remembered, and choose Unpersonalized before the first message if it shouldn't use your memories either.
5. On a personal account, check data controls too; see what not to paste.

Next step: the team working the same way

Memory makes one person's ChatGPT fit them; it does nothing for the team. Getting your team to use AI the same way covers sharing one method. For the three places rules can live, see custom instructions vs projects vs skills.

Questions

Should I use ChatGPT memory for work?

For stable preferences, yes: your role, the format you like, words you avoid. Not for rules or facts other people rely on, such as a client's pricing or the team's method: memory is personal, changes over time and isn't something a colleague can see. OpenAI's docs say memories are separate from required guidance, which belongs in instructions.

How do I turn off ChatGPT memory?

In Settings > Personalization. OpenAI's Memory FAQ, read through search results, describes two settings: Reference saved memories and Reference chat history. Turning off saved memories also turns off chat history, and doesn't delete what was already saved; delete those separately. For a single chat, use Temporary Chat: OpenAI's Temporary Chat FAQ says it doesn't create or update memories, and if you choose Unpersonalized before the first message, it doesn't use your existing memories or custom instructions either.

Where should work rules go instead of memory?

Rules for every chat go in custom instructions; the background for one client or job goes in a project; a method the team repeats goes in a skill. All three can be read and changed on purpose, which memory can't be in the same way.