Field note
ChatGPT Temporary Chat: What It Does at Work
Temporary Chat keeps a conversation out of your history and memory, but it is not a data-loss control for files, connected apps or anything you paste that you should not have shared.
Temporary Chat keeps a conversation out of your history and memory, but it is not a data-loss control for files, connected apps or anything you paste that you should not have shared.
What Temporary Chat does
Temporary Chat is a switch at the top of a new ChatGPT conversation. At the time of writing, OpenAI describes it like this: the chat does not appear in your history, it does not use or add to your saved memories, and it is not used to train models. OpenAI may keep a copy for a limited period, currently stated as up to 30 days, for safety and abuse review. The exact wording is in the OpenAI Help Centre article on Temporary Chat, and it changes, so check it before you quote it to staff.
That makes Temporary Chat useful for one-off work where the risk is contamination: you do not want a client's name, a candidate's details or a pricing negotiation to turn up as a remembered fact in next week's unrelated chat. It is also handy for a colleague using your machine for five minutes.
What it does not do
It does not stop the data leaving your network. The text still goes to OpenAI and is processed there. If a data class is not allowed in ChatGPT, it is not allowed in a Temporary Chat either.
It does not make files vanish. A document uploaded to a Temporary Chat is still uploaded, still processed, and still subject to the retention period above.
It does not undo actions. If the chat used an app or connector to create a calendar entry, draft an email, or write to a spreadsheet, that record now lives in the other system and stays there.
It does not override the workspace. On Business and Enterprise plans the admin may have logging, retention or compliance settings that apply regardless of the switch. Ask what they are.
It does not change your legal position. UK GDPR obligations apply to what you entered, not to whether the conversation shows in a sidebar.
A realistic example
The practice manager at a small dental group needs to draft a difficult letter to a patient about an outstanding balance and a complaint. She uses a Temporary Chat so the patient's name and history do not end up in memory, which is sensible. She then pastes the full patient record, including treatment notes, "for context". The letter is good. The record has now been sent to a third-party processor without a documented basis, the data-protection lead has no idea, and the fact that the chat disappeared from the sidebar afterwards is irrelevant to the ICO.
The right version is the same task with a two-line summary instead of the record, on the company's Business workspace, with the letter reviewed by a person before it goes out.
When to use it, and when not to
Use Temporary Chat for:
- sensitive one-off drafting where memory carry-over is the real risk;
- quick personal questions on a work account;
- testing a prompt that contains dummy data before you build a Project around it; and
- letting someone else use your account briefly.
Do not use it as:
- a way to handle data that is otherwise banned from ChatGPT;
- a substitute for a Business or Enterprise workspace;
- a control for connectors or apps; or
- a reason to skip the review step before something is sent to a customer.
Controls that actually reduce the risk
Decide the data classes first. A short list of what never goes into any AI assistant, whatever the mode, does more than any switch. Turn memory off entirely for roles that handle patient, client or HR data, then Temporary Chat becomes a convenience rather than a control. Keep business use on a company workspace so an admin can see and set retention. Restrict apps and connectors to an approved list so a "temporary" chat cannot quietly write to a live system.
The ChatGPT memory privacy review covers the memory side, and the AI chatbot HR data boundaries note is the one to hand to whoever runs people matters.
What to write into policy
- Temporary Chat reduces persistence inside ChatGPT; it does not change what data may be entered.
- The banned data list applies in all modes, on all plans, on all devices.
- Files and connector actions in a Temporary Chat are treated the same as in any other chat.
- Roles handling client, patient or HR data have memory switched off, and the setting is checked at onboarding.
- Anything drafted with AI that goes to a customer is reviewed by a named person first.
- The OpenAI wording relied on, and the date it was checked, is recorded with the policy.