Staff should use temporary chat for one-off questions they won't need again and that contain no customer or confidential details: checking wording, explaining a term, testing a prompt, a personal question on a work seat. It stays out of history and memory and isn't used for training, but OpenAI may keep a copy for up to 30 days.
The common misunderstanding is treating it as a privacy mode. Temporary chat changes what ChatGPT remembers and what appears in your history; it doesn't change what you've sent to OpenAI. And because an unsaved temporary chat can't be reopened after you leave it, it's the wrong choice for anything the business might need to find again, which covers more staff work than people expect.
What a temporary chat changes, and what it leaves alone
You start one by opening a new chat and selecting Temporary. Before the first message, you choose a personalised or unpersonalised version. Here is how the three options compare, based on OpenAI's temporary chat help page as of September 2026:
| Behaviour | Normal chat | Temporary, personalised | Temporary, unpersonalised |
|---|---|---|---|
| Appears in chat history | Yes | No, unless saved | No, unless saved |
| Uses existing memories, custom instructions and plugins | Yes | Yes | No |
| Creates or updates memories | Yes, if Memory is on | No | No |
| Used to improve OpenAI's models | Depends on your plan and training setting | No, while it stays temporary | No, while it stays temporary |
| Copy kept by OpenAI | Under your normal retention | Up to 30 days, for safety | Up to 30 days, for safety |
One more behaviour matters for staff: a temporary chat can be saved to history. Saving turns it into a regular chat, even one that started unpersonalised, and from that point it follows the account's normal personalisation, training and retention settings. So "temporary" is a starting state, not a guarantee, and a quick save undoes the protection.
On a business workspace, the organisation's own retention, memory and compliance settings still apply around all of this, and admins decide which features members can use. On ChatGPT Business, training on your content is already off by default, so there the main benefits of temporary chat are a tidy history and a memory that isn't cluttered with stray facts.
What a temporary chat adds therefore depends on how the seat is already set up. For each common set-up:
| How the seat is set up | What a temporary chat adds |
|---|---|
| Individual plan, model training on, Memory on | Keeps that chat out of training, out of memory and out of history |
| Individual plan, training switched off, Memory on | Keeps it out of memory and history |
| Business seat, Memory on | Keeps it out of memory and history |
| Business seat, Memory off | Keeps it out of history, and nothing else |
The pattern is worth telling staff plainly: the better your account settings, the less temporary chat does. On a business seat with Memory off, it is mainly a way to keep history tidy, and the "never" rules later on matter far more than which mode someone picks.
Personalised or unpersonalised: pick before the first message
The personalised version uses your existing memories, custom instructions and plugins, without adding anything new to memory. The unpersonalised version ignores all three. That second option is useful and easy to misuse, and one example shows both sides.
An illustrative picture framer asked, in an unpersonalised temporary chat:
In two sentences a customer would understand, what does
conservation framing mean?The illustrative reply:
Conservation framing uses acid-free materials and UV-filtering glass
to protect artwork from fading, yellowing and damage over time. It
costs more up front but helps preserve the value and color of your piece.The content was fine, but the spelling color gave it away: unpersonalised mode had ignored the framer's custom instructions, which set the shop's house spelling and its plain, friendly tone. For a one-off explanation that's easily fixed by hand. For anything customer-facing that should sound like the business, use the personalised version, which keeps the house style without adding to memory. Unpersonalised earns its place when you want an answer free of your own context: a second opinion on pricing that isn't coloured by months of your earlier chats, or a check of how a prompt behaves for someone with no history, before you put it in a shared project.
That prompt check is worth seeing in action. An illustrative garden centre manager wanted a prompt the whole team could use for plant care cards, and tested it unpersonalised first:
Write a plant care card for [plant name] in our usual format.In her own normal chats the prompt worked, because months of earlier chats had taught ChatGPT what "our usual format" meant. Unpersonalised, it produced a 300-word essay with headings, nothing like the 40-word cards on the shop shelves. The test exposed that the prompt leaned on her history, which colleagues' seats wouldn't have. The fixed version spells the format out: "four lines: light, water, soil, one tip; under 45 words; no Latin names unless asked". Tested unpersonalised again, it produced a usable card first time, and only then went into the team's shared project.
Five situations where staff should use it
- One-off lookups. A dry cleaner's counter assistant checking what a care-label symbol means, or how to phrase "we can't guarantee removal of this stain" politely. Nobody needs that chat next week.
- Personal questions on a work seat. Someone planning a holiday or asking about a medical appointment on their work account. A temporary chat keeps those details out of the seat's memory, where they'd otherwise shape later work answers.
- Testing a prompt before sharing it. Try a new prompt unpersonalised to see how it behaves without your context, then add the finished version to the team's project.
- A second opinion free of memory. Unpersonalised mode gives an answer that hasn't absorbed your earlier assumptions, which helps when you suspect ChatGPT is simply agreeing with what you told it last month.
- Commercially sensitive but non-personal drafts on an individual plan. A sole trader on Plus who hasn't switched model training off can use a temporary chat so that one chat isn't used for training. The better fix is turning the training switch off, which covers every chat.
A sixth case deserves its own paragraph, because it's where temporary chat protects people rather than the business: the shared counter computer. An illustrative dry cleaner has one PC at the counter signed in to one ChatGPT seat, used by three staff across the week. Anything typed in a normal chat there is readable by the next person who opens the history, and it feeds the seat's memory, so one person's questions shape the answers the others get. Temporary chats close both doors: an unsaved one isn't in the history for the next person, and it adds nothing to memory. It's still a stopgap. Three people on one seat means nobody knows who asked what, and separate seats, even at the Business minimum of two, fix that properly.
One limit to explain honestly: temporary chat isn't a way to hide work from the business. OpenAI's help pages say a managed workspace's own retention, compliance and access settings still apply, and the organisation's policies still govern the account. Staff should treat a temporary chat on a work seat as private from their own history and memory, not from their employer.
Four situations where it's the wrong tool
- Customer or patient details. A copy may be kept for up to 30 days, so temporary chat doesn't make it acceptable to paste something you wouldn't otherwise paste. If a receptionist pastes a customer's address and complaint on 1 September, a copy may exist until 1 October. The rules for what belongs in ChatGPT at all are in whether it's safe to put customer data into ChatGPT.
- Work someone will need again. Quotes, supplier disputes, complaint replies, anything with reasoning a manager may ask about. Once the tab closes, an unsaved temporary chat is gone from the person's view.
- Client work. Ongoing work for a client belongs in that client's project with project-only memory, as set out in using ChatGPT Projects to keep client work separate.
- Anything that should be a business record. On a business plan the organisation may want chats kept, for handovers and for when someone leaves. Temporary chats bypass that, which is a problem covered in who owns AI chat history when an employee leaves.
An optician's week of chats, sorted
To see how often each option fits in real work, take an illustrative optician's practice with four ChatGPT Business seats: the practice manager, two dispensing opticians and a receptionist. The manager reviewed one week's chats across the four seats, 60 in total, and sorted them by what should have happened.
| What the chat was | Count | Right tool |
|---|---|---|
| One-off lookups: terms, wording, a lens coating explained | 26 | Temporary, personalised |
| Personal questions on a work seat | 6 | Temporary |
| Drafts reused later: recall letters, a supplier complaint, a job advert | 18 | Normal chat or a project |
| Anything including a patient's name, prescription or health detail | 8 | Not ChatGPT, or anonymised first under the practice's rules |
| Prompt tests before adding to the shared project | 2 | Temporary, unpersonalised |
So 34 of 60 chats, 57%, suited a temporary chat, but only 3 had actually used one. The memory on the receptionist's seat had picked up six details from personal chats, including a child's school and a dentist appointment. Meanwhile the 8 patient-detail chats were the real problem, and temporary chat wouldn't have fixed them. The practice changed two things: a one-page guide, below, and a rule that patient details are anonymised before any chat, temporary or not, following the practice's anonymisation steps. A month later the manager's recount showed 29 temporary chats out of 58, no patient names in any chat, and the receptionist's memory reduced to two work-related entries.
The mistake that shows up the next morning
A pet shop assistant drafted a firm email to a supplier about a short delivery in a temporary chat, sent a version, and closed the tab. The next morning the supplier disputed the quantities, and the owner asked how the figures in the email had been worked out. The chat, with the assistant's pasted delivery note and ChatGPT's working, had gone from her view. Reconstructing it took 40 minutes and a call to the driver.
Two habits would have prevented it. Anything involving money, a dispute or a promise goes in a normal chat from the start. And if a temporary chat turns into real work halfway through, save it before leaving, which converts it into a regular chat. Staff who know the save option exists use temporary chat more freely, because they know they can keep something if it turns out to matter.
A one-page staff guide you can adapt
Most AI policies say something like "use temporary chat for anything sensitive". That wording is what caused the optician's patient-detail problem, because it suggests temporary chat makes sensitive data safe. A before-and-after of the rule itself:
| Before | After |
|---|---|
| Use temporary chat for anything sensitive. | Use temporary chat for one-off questions with no customer, patient or staff details. Sensitive details don't go into any chat unless anonymised first. |
The full guide, as the optician's practice pinned it by the reception computer:
WHICH CHATGPT CHAT TO USE
Temporary chat (New chat, then Temporary):
- One-off questions: wording, terms, how something works
- Personal questions on your work account
- Choose the personalised option unless you're testing a prompt
Normal chat:
- Anything you or the manager may need again
- Anything involving money, a complaint, a supplier or a promise
Our shared project:
- Recall letters, standard replies, anything using practice templates
Never, in any chat:
- Patient names, prescriptions, dates of birth or health details
unless anonymised using the practice steps
- Passwords, card numbers, bank details
If a temporary chat turns into real work: save it before you close it.
Questions: ask [practice manager's first name].Adapt the examples to your business, and keep it to one page. A dry cleaner's version might list garment-care questions under temporary and customer compensation claims under normal chat; a music teacher's might put term-fee reminders in a project and a question about a pupil's needs under "never" unless anonymised. Where this sits within a wider policy is covered in writing an AI usage policy for a small business.
A two-minute team briefing, with three practice questions
Guides get read once. A short briefing at a team meeting sticks better, and an image helps: a temporary chat is a whiteboard wiped when you walk away, except that the cleaner keeps a photo of it for 30 days. So you can scribble on it freely, but you still don't write anything on it you wouldn't want photographed.
Then run three quick scenarios and let people answer before you do. For an illustrative garden centre team:
- "I want help wording a notice about our price rise." Temporary, personalised: it's a one-off question and the house style should apply.
- "I'm drafting the actual price rise letter going to 400 trade customers." Normal chat, or the team project: it will be reused, questioned and possibly revised, so it must be findable next week.
- "I need to text a customer whose order is late, and I've got her number and order details here." No chat at all for the personal details: write the message with a placeholder such as [first name] in any chat, then add the real details in the till system or phone.
The third one is the one people get wrong, and it's the reason the briefing is worth two minutes of a meeting.
Checking the guide is being followed
Three quick checks, once a month, take about 15 minutes for a small team:
- Look at the memory on each seat. Personal or stray entries mean personal questions are still going through normal chats. Delete what shouldn't be there and remind that person of the rule.
- Skim a sample of normal chats for customer or patient details. The temporary chats you can't see, which is precisely why the "never" list has to hold whatever mode someone picks.
- Ask one question at the team meeting: did anyone lose something they needed because it was in a temporary chat? If yes, the "normal chat" list needs another example from your own work.
How ChatGPT's memory works on each plan, and how to review and clear it, is explained in how ChatGPT's memory settings work for business. The short version for this guide: temporary chats keep memory clean, and a clean memory makes every normal chat more accurate.
Further reads
- AI Acceptable Use Policy for a Small Professional Firm — Fold temporary-chat rules into a full acceptable use policy.
- Shadow AI: Is Your Team Using AI Without Telling You? — Staff using AI you don't know about, and how to bring it into view.
- ChatGPT Plus vs ChatGPT Business: Which Plan Does a Team Need? — Why training defaults differ between individual and business seats.
- Patient Data and AI: A Confidentiality Checklist for Small Practices — For practices handling health information, a stricter checklist.
- How to Train Staff to Use AI in a Small Business — How to teach habits like this so they stick.
- How to Stop AI Tools Training on Your Business Data — The exact training switches in ChatGPT, Claude, Gemini, Copilot and Perplexity, plus the hidden AI features most owners forget to check.
- Can Therapists Use ChatGPT for Session Notes? — Personal plan, de-identified shorthand, business plan or a therapy note tool: where each lands, and why removing a name rarely removes the client.
- How to Keep Customer Data Private When Your Team Uses AI — Eight steps to keep customer details out of the wrong AI tools, with a filled-in data table, redaction examples, policy wording and a quarterly audit.
- Shadow AI: How to Stop Staff Pasting Client Data Into Free Tools — Stop client data leaking into free chatbots with an approved tool, a traffic-light data rule, browser blocks and a no-blame reporting route.
- ChatGPT Connectors: What They Can See in Your Drive and Inbox — What ChatGPT's Drive, Gmail, Outlook and SharePoint apps can read, where training and memory come in, and the admin switches that narrow access.
- Are Custom GPTs Private? What Happens to the Files You Upload — Who can see a custom GPT's chats and knowledge files, how consumer and business plans differ, and what the December 2026 retirement means for your uploads.
- Is ChatGPT Safe for Business Use? Risks, Settings and Plan Choice — The five real risks of using ChatGPT at work, the settings that fix most of them, and which plan fits a sole trader, a small team or a clinic.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI help pages (Temporary chat in ChatGPT; Chat and file retention in ChatGPT; Data controls in ChatGPT; Memory in ChatGPT; ChatGPT privacy centre), checked September 2026.