ChatGPT Temporary Chat: what it keeps out, and what it can now read

You're about to ask ChatGPT about something you'd rather it didn't carry around: a client's restructuring, a partner dispute, a number nobody outside the room has seen. Temporary Chat is the tool for that, and on 27 August 2026 OpenAI changed what it does. Here is what it keeps out, what it can now read, how long OpenAI keeps a copy, and how it differs from deleting a chat or turning memory off. Every OpenAI statement below was read on 20 September 2026.

Short answer: a Temporary Chat is a ChatGPT conversation that stays out of your chat history, doesn't add anything to ChatGPT's memory, and isn't used to improve OpenAI's models, for as long as it stays temporary. Temporary used to mean "ChatGPT forgets this." Since 27 August 2026 it can also mean "ChatGPT reads what it knows about you, and forgets only this": you now choose, before the first message, whether the chat may draw on your existing memories, custom instructions and plugins (OpenAI's word for connected apps such as Gmail). You can also save a temporary chat, which turns it into an ordinary one. OpenAI says it "may keep a copy of a temporary chat for up to 30 days for safety purposes." If what you want is a clean room for one sensitive question, choose Unpersonalized, and don't save.

What a Temporary Chat is, in OpenAI's current words

OpenAI's help article "Temporary chat in ChatGPT" opens with the whole definition: "A temporary chat stays out of your chat history and is not used to improve OpenAI models while it remains temporary. It can use existing memories, custom instructions, and plugins to personalize responses. Before the conversation begins, you can turn personalization off. You can also save the chat later as a regular chat."

Compare that with how OpenAI described the same feature a few months earlier, in release notes that are still on its site. On 5 May 2026: "temporary chats that do not use or update memory or appear in history." On 12 June 2026: "Temporary Chats don't use existing memories or create new memories." Those two sentences are what many older guides still repeat. The second half of each is still true. The first half, "don't use existing memories," stopped being the whole story on 27 August.

What changed on 27 August 2026

The release note that day is titled "More controls in temporary chat," and it made two changes: "You can now choose to personalize responses in a temporary chat with memory, plugins, and custom instructions from your regular chat settings. You can also choose to save a temporary chat." OpenAI's own explanation of each, in its words, because the details matter:

• Personalized. "Personalized temporary chats can use your memories, custom instructions, and plugins to tailor responses. They do not create new memories and remain out of your chat history unless you save them. Personalization can be set only when you start a temporary chat and cannot be changed after the conversation begins."
• Non-personalized. "By default, temporary chats are not personalized. Non-personalized temporary chats do not use memory, custom instructions, or plugins, and they do not create new memories."
• Saving. "When a conversation becomes useful for the future, you can add it to your chat history. Saving converts it into a regular chat, so it then follows your account-level personalization settings and model-improvement preferences."

A note on the word "plugins." OpenAI's release notes use it for connected apps. The entry dated 28 August 2026 begins "You can now connect multiple accounts for Gmail, Google Calendar, and Google Contacts plugins in ChatGPT," and the entry dated 11 September 2026 describes plugins as something that "can combine reusable instructions with connected apps." So our reading, which OpenAI doesn't state in these words, is that a Personalized temporary chat may reach what you've connected, not only what you once typed into ChatGPT. OpenAI's pages don't say which plugins work inside a temporary chat. We checked every later entry in the release notes, up to the newest one on 17 September 2026, and none mentions Temporary Chat.

One thing OpenAI's pages don't agree on: the default

The release note says plainly that "By default, temporary chats are not personalized." The help article was revised after that (it was marked as updated within the previous day when we read it on 20 September 2026), and the word "default" does not appear anywhere in it. Its overview instead says the chat "can use existing memories, custom instructions, and plugins" and that "you can turn personalization off," which reads as though personalization may be the starting position. The labels differ too: the release note says "Non-personalized," the help article says "Unpersonalized."

We can't tell you from OpenAI's pages which setting your app will show first, and we're not going to guess. The safe habit costs two seconds: look at the control before you type anything, because, as the help article puts it, "You cannot change personalization after the conversation starts."

How to start a Temporary Chat

OpenAI documents four steps, and these are its words:

• "Open a new chat."
• "Select Temporary."
• "Choose Personalized or Unpersonalized before you send your first message."
• "Enter your message."

That is the whole procedure. OpenAI's article gives one set of steps and doesn't break them out for the website, the iPhone app and the Android app, so we won't invent three. Where the Temporary control sits may differ between them, and any walkthrough made before 27 August 2026 describes a version without the third step. The third step is the one that matters.

OpenAI's other pages do name a location, though both still describe the older behaviour elsewhere in the same text, so treat them as pointers: its "What is ChatGPT: FAQ" says to "Start a temporary chat from the model menu," and its macOS app article says to "click on the model name at the top of the application window and select the Temporary Chat toggle."

Two conditions sit above your choice. "Your account and workspace restrictions take priority over the personalization choice for an individual temporary chat," so on a firm-managed account the options you see may not be the ones described here. And the choice governs personalization only: "ChatGPT may still use information from prior conversations for limited safety and security purposes," which OpenAI describes as applying "in rare, high-risk situations."

What a Temporary Chat does not do

• It doesn't appear in your history, and you can't get back to it. "If you do not save a temporary chat, it will not appear in your history and cannot be reopened from history after you leave it." Copy out anything you need before you close the window.
• It doesn't add to memory. "Neither option creates or updates memories while the chat remains temporary." OpenAI's Memory article says the same from the other side: "Temporary chats do not create or update memories, even with personalization on."
• It isn't used for model training while temporary. "A temporary chat is not used to improve OpenAI models while it remains temporary, whether or not you choose personalization. If you save it, your model-improvement setting applies."
• It doesn't mean nothing is kept. "OpenAI may keep a copy of a temporary chat for up to 30 days for safety purposes." The same figure appears in OpenAI's retention article and in its Data Controls FAQ, which words it as "Temporary Chats are deleted from our systems after 30 days" and says they "May be reviewed only to monitor for abuse." Temporary describes your history and ChatGPT's memory. It does not mean the text was never sent or never stored.
• It doesn't cover third parties. "If a GPT uses actions, data sent to a third party through an action is governed by the recipient's privacy policy. The recipient may retain that data for longer than 30 days and may use it for other purposes." For a sensitive conversation, use plain ChatGPT, not a custom GPT that calls outside services.
• It doesn't hide the chat from a company's compliance tooling. "For eligible Enterprise customers, temporary chats are available through the Compliance API for 30 days, including when the workspace has a different custom retention period." If your ChatGPT seat belongs to your firm, temporary is temporary for you, not necessarily for the firm.
• It doesn't stay temporary once saved. Saving "converts it into a regular chat, including when it started without personalization," and "If Memory is on, the saved conversation may be used to personalize future responses." Uploaded files can come along: "eligible files may also be saved to Library when you save the chat, subject to availability and storage limits."

A routine for one sensitive client conversation

Putting OpenAI's statements together, this is the order we'd follow. The steps are OpenAI's; the judgement calls are ours.

• Start it as temporary, and read the control. If the goal is that nothing ChatGPT already knows about you, your firm or your other clients colours this answer, choose Unpersonalized. If you want your usual tone and standing instructions applied to a one-off draft and don't mind your memory shaping it, Personalized is the right tool: the conversation still won't feed memory.
• Decide what you paste with the 30 days in mind. A temporary chat keeps a matter out of your history and out of memory. It doesn't keep it off OpenAI's systems for that window. If a name or a figure would be a problem sitting in a copy for a month, leave it out or replace it.
• Skip custom GPTs for this one. Anything a GPT sends to an outside service is governed by the recipient's privacy policy, not OpenAI's 30 days.
• On a firm account, ask before assuming. Workspace restrictions override your per-chat choice, and Enterprise compliance access applies to temporary chats too.
• Take what you need, then leave without saving. Saving is the one action that undoes the point of the exercise.

Temporary Chat vs deleting a chat vs turning memory off

These three get used interchangeably, and they do different jobs. Wording in quotation marks is OpenAI's, from the articles named in the footnote, read on 20 September 2026; everything outside quotation marks is our summary.

Read across the memory row and the division of labour is clear. Temporary Chat is prevention. Deleting is clean-up of the transcript. Turning memory off is a switch for the whole account. Only the first keeps a conversation from reaching memory at all, and it only works if you chose it before you started typing.

Where Temporary Chat runs out: you have to know in advance

Temporary Chat is a good tool, and for a one-off sensitive question it's the right one. Its limit is that it's all-or-nothing, per chat, decided up front. Real client work doesn't sort itself that neatly. The sensitive detail usually turns up in the middle of an ordinary conversation: you're drafting a proposal in a regular chat and, twenty minutes in, you mention what the client's CFO told you in confidence about the margin problem. Nothing about that chat was temporary, and the fact may now be part of what ChatGPT remembers.

For that case OpenAI's guidance is a list, not a button. Its Memory article: "To remove information saved as a memory, delete the saved memory and the chat where you first shared it. Also remove the information from any other sources where it appears," naming the memory summary, saved memories, regular and archived chats, files in Library and connected apps. And the place you'd go to check whether it worked is described by OpenAI this way: the memory summary "is a high-level view" that "does not necessarily include every detail or source that Memory can reference." The article's own FAQ asks "Does the memory summary include everything ChatGPT remembers?" and answers "No."

That's not a criticism of OpenAI's candour; it publishes all of this openly. It's a description of the design. In what OpenAI calls improved Memory, ChatGPT "continually updates a broader summary of relevant context," so the question "is that one fact still in there?" often doesn't have a line you can point at. (Where the older saved-memories experience is still offered, OpenAI describes it as "a more explicit list of individual memory items.") The step-by-step version of the clean-up is in how to delete ChatGPT memory, what survives a deleted chat is in does ChatGPT remember deleted chats, and how the summary itself behaves is in what the ChatGPT memory summary is.

The other shape: a memory you can open and prune one line at a time

Equerry is a personal assistant that works the other way round, and it's worth being clear that it is not a temporary chat: it's for the things you do want remembered. Nothing reaches it unless you send it. You CC or forward the one client thread, share a PDF or a screenshot from the share sheet, dictate a voice memo after the call, or type a note. There's no account to connect; it reads the thread you sent, not your inbox.

For the notes you save, what it keeps is a list you can open. In Settings → Your memory, each fact Equerry has remembered from a note sits on its own line with the note it came from beside it. If one of the lines is the thing you wish you hadn't shared, you delete that one line and leave the rest alone; delete a note you saved and what came from it goes too. When an answer draws on a note you saved, it carries a source chip: tap it to see the passage Equerry used and open the original. What Equerry sends to which providers, and for how long they keep it, is set out in our disclosure page.

Keep ChatGPT, and keep using Temporary Chat for the one-off question: choose Unpersonalized, don't save. For the client work you want held onto, the useful property isn't a memory that forgets everything or nothing. It's one where, for what you've saved, you can see each item, see where it came from, and remove it on its own. Equerry is on the App Store for iPhone and iPad.

Sources, all OpenAI's own pages, read on 20 September 2026 in a browser session because the help center refuses automated fetches: "Temporary chat in ChatGPT" (help.openai.com, article 8914046, formerly the Temporary Chat FAQ) for the definition, the four steps, both personalization modes, saving, training, the 30-day copy, GPT actions and Compliance API access; the ChatGPT release notes (article 6825453; read in full, newest entry 17 September 2026, oldest 15 December 2022) for the 5 May, 12 June, 27 August, 28 August and 11 September 2026 entries; "Memory in ChatGPT" (article 8590148) for turning memory off, Reference chat history, deleting remembered information and the memory summary; "Deleting and archiving chats in ChatGPT" (article 8809935) and "Chat and file retention in ChatGPT" (article 8983778) for deletion and the 30-day schedules; the Data Controls FAQ (article 7730893) for its wording on Temporary Chats; and "What is ChatGPT: FAQ" (article 12677804) and "Temporary Chats on the ChatGPT macOS app" (article 9295248) for where the control sits. The Temporary chat and Memory articles were both marked as updated within the previous day when we read them, so wording you see quoted elsewhere may predate that revision. Quoted wording is OpenAI's; where its pages are silent or disagree with each other, this guide says so instead of filling the gap. OpenAI changes these pages often; treat your own ChatGPT screen as the authority for your account.

Last updated 2026-09-21.

All product names, logos, and brands are property of their respective owners. Equerry is not affiliated with, endorsed by, or sponsored by any other product mentioned; comparisons are provided for informational purposes.