Developing with ChatGPT

Model training, deletion, and the thirty-day window

ChatGPT

One toggle decides whether your chats train future models. A separate set of rules decides how long anything survives after you delete it, and files do not follow the same rules as chats.

Applies to
ChatGPT
Last verified
Reviewed by
Timothy Fehr

Two separate questions get confused with each other constantly. Whether your conversations train future models is one setting. How long anything survives after you delete it is a different set of rules, and the answer changes depending on whether the thing you deleted was a chat or a file.

The toggle

It lives in Settings → Data Controls, worded "Improve the model for everyone".

For personal accounts the direction of the default is visible in how OpenAI words its own policy: individual services "may use your content to train our models" and you opt out. So the choice most people are living with is the one they never made.

Turning it off applies to the whole account. The documentation is explicit that it does not matter which device you are on, and that you can change the setting at any time with no restrictions. Your existing conversations stay in your history; they simply stop being training material.

Deleting, archiving, and the thirty days

Chats stay in your account until you delete them by hand. Delete one and it leaves your account immediately, then is scheduled for permanent deletion from OpenAI's systems within 30 days. Two exceptions are stated: content already de-identified and disassociated from you, and content OpenAI must keep longer for security or legal reasons.

Archiving is not deleting. It hides a chat from the sidebar and keeps it in your account under exactly the same retention rules. People reach for it thinking it is a soft delete. It is a soft hide.

Temporary Chat is the closest thing to a clean surface: no history, no memories, not used for training, and deleted within 30 days without you doing anything. OpenAI notes such chats may still be reviewed to monitor for abuse.

Files do not follow the chat

This is the part that surprises people, and it is stated plainly in the retention documentation: deleting a chat does not delete files you saved to Library. Chats and Library files are managed separately.

Files uploaded into a custom GPT or a project live until that GPT or project is deleted, and are removed within 30 days after that.

So "I deleted the conversation" and "the document I pasted in is gone" are two different claims, and only one of them follows from deleting a chat.

Business accounts invert the default

The default flips on the business side. OpenAI states it does not train on inputs or outputs from ChatGPT Team, ChatGPT Enterprise, or the API by default, and that organisations are opted out of data sharing unless they explicitly opt in.

Retention becomes an administrator's decision rather than yours. On Enterprise, Edu and Healthcare workspaces the admin controls how long data is kept, with deleted conversations removed within 30 days unless legally required otherwise. Business works the same way and extends it to unsaved conversations. API inputs and outputs may be retained up to 30 days to detect abuse, and zero data retention is available for eligible endpoints.

You will encounter the claim that OpenAI keeps deleted consumer chats forever because of litigation. The retention documentation's own wording already contains the hinge: deletion happens within 30 days unless OpenAI must retain content longer for security or legal obligations.

On the litigation specifically, OpenAI states that content covered by the court order is stored separately in a secure system under legal hold, where it cannot be accessed or used for anything except meeting legal obligations.

What the pages consulted here do not state is the current scope of that hold: which accounts and which date range remain covered today. Treat any specific figure you read elsewhere as unverified unless it comes from OpenAI's own litigation updates.

Try this

Open Settings → Data Controls and read what the toggle currently says. Then ask yourself the only question that matters: did I choose this, or is it a default? Most people find it is a default. Then decide deliberately, either way.

What goes wrong

Assuming archive means delete. It hides the chat and keeps everything.

Assuming deleting a chat removes the file you uploaded. Library files are separate and survive.

Treating a personal account and a work account as the same product. They run on different contracts with opposite training defaults.

Turning the toggle off and believing it reaches backwards. It governs new conversations only.

Quoting a retention figure from a comparison article. These change, and aggregators do not tell you when they went stale.

How to check it worked

Look at the setting for the account you are actually signed into rather than the one you remember configuring, and check it on the surface you use most. If you administer a workspace, check what a newly created account defaults to, because that default is what almost all of your users will have. Then delete a test chat and confirm that any file you uploaded in it is still sitting in Library, which is the behaviour most people are surprised by.

Sources

  1. Data Controls FAQ — OpenAI Help Center Tier 1 2026-09-10
  2. Chat and File Retention Policies in ChatGPT — OpenAI Help Center Tier 1 2026-09-10
  3. How your data is used to improve model performance — OpenAI Tier 1 2026-09-10
  4. Enterprise privacy — OpenAI Tier 1 2026-09-10
  5. Fighting the New York Times' invasion of user privacy — OpenAI Tier 1 2026-09-10