ChatGPT Privacy Checklist

ChatGPT can remember details from your conversations and use them to personalize answers. Here's how to review what it knows, question what it assumes, and choose what you want it to keep.

See What It Remembers

If you came from the PROFILE video, start here. Open a regular chat in ChatGPT. Use the two prompts on these first pages, then follow the settings checklist.

Prompt 1: Your memory review

Lay out the personal information about me that is available to you in this chat as a detailed list. Don't compress separate details into a vague summary.

Group it by work, preferences, routines, location, relationships, and any other categories supported by the information you can actually access. Identify the source when available, such as this conversation, a saved memory, a past chat, custom instructions, a file, or a connected app.

Separate things I explicitly told you from your inferences. Don't fill gaps or invent details. Say what you cannot access or verify. Don't describe this as a complete account-data export or a list of everything OpenAI stores.

Read through the answer and mark anything that is wrong, outdated, too personal, or about someone else. You don't have to remove useful preferences just because you want to remove an old address or client detail.

Stay in the same chat. This follow-up gets at the second part of the video: the conclusions ChatGPT might draw from the context it has.

Prompt 2: Show me the evidence

Now review the inferences you made about me that I did not explicitly tell you.

For each one, show the inference, the specific clue you can actually point to, and at least one other possible explanation. Label it as uncertain. If you cannot identify supporting context, say unsupported and withdraw the claim.

Distinguish an inference you are making now from information you can verify was already remembered. Don't invent an income, diagnosis, personal weakness, or prediction about my future to make the answer more interesting. If you found no supported inferences, say so.

How to read the answer

An answer that sounds personal can still be a guess. Asking about a city doesn't establish that you live there. Asking about a salary doesn't establish your income. A conversation about procrastination doesn't prove a permanent personality trait.

The useful question is: what supports this? If Sources appears beneath a response, open it and review the available evidence. Missing evidence means you should treat the claim as unverified.

What these prompts can and cannot show

They can help you review the context ChatGPT can access and challenge assumptions in its reply. They cannot produce a guaranteed complete hidden profile, prove that every inference was previously stored, or reveal a reliable prediction of your future.

For an account-data copy, use the export option later in this guide. A confident answer in a chat is not a substitute for that process.

On the web, click your profile icon at the bottom left > Settings > Personalization. Find ChatGPT memory, then Memory summary > Manage. The summary is an overview, so also review saved memories if that link appears.

On other versions, the route may read Personalization > Memory > Manage memories. Use the controls available in your account. The important part is reviewing both the summary and any separate saved-memory items. If you want to keep memory and make it work better for you, see my ChatGPT memory guide.

Prompt 3: Remove a specific topic

Choose a topic from the review, replace the bracket, and paste this into the same chat. Use a short topic label rather than copying a sensitive number or address again.

Forget the saved memories you can remove about [PERSON OR TOPIC]. Do not create a replacement memory containing those details.

Tell me which changes you actually completed, which you could not complete, and which original chats, instructions, files, or connected sources I need to review myself. Don't claim that this deletes my chat history or all copies of the information. Don't repeat the sensitive details in your confirmation.

Return to the memory controls and check the result. For a wrong detail you want to keep, use the available correction field instead. If you want memory off, use its off switch. Where offered, Delete and turn off memory also removes the remembered information shown in the summary.

Keep the distinction clear

Turning memory off alone does not delete your existing chats or necessarily erase saved memories. Deleting a saved memory does not remove the conversations that mention it. A request to stop mentioning something also does not delete its source.

If your account has Reference chat history, turning it off schedules the information remembered from past chats for deletion within 30 days. The chats themselves remain until you delete them. Deleted saved-memory logs may be retained for up to 30 days for safety and debugging.

To clean up a topic, check everywhere you put it. A sensitive detail can appear in more than one conversation or document.

For a chat you want removed, use its three-dot chat menu > Delete, then read and confirm the deletion. Archive only moves it out of the main list. It keeps the conversation in your account.

Deletion is permanent, but not instant everywhere

Deleted chats leave your account view immediately and are scheduled for permanent deletion within 30 days. Exceptions can apply for security or legal obligations, or if content was already de-identified and separated from your account. Deleted chats cannot be recovered through ChatGPT or Support.

Deleting a chat does not remove a file saved separately in Library or Space. Disconnecting an app stops future access, but does not erase conversations that already used its data or delete the original material in that app.

Click your profile icon > Settings > Data controls. Find Improve the model for everyone and turn it off. The web interface may show the switch directly. If your version opens a separate panel, turn it off there and select Done if shown.

This stops new conversations from being used to train OpenAI models. It applies across devices when you are signed in to the same account. It does not erase your history, turn off memory, or undo any training that has already happened.

One exception to remember

If you submit feedback, such as a thumbs-up or thumbs-down on a response, the associated conversation may still be used for training. Avoid submitting that feedback on a conversation you want excluded.

Three different controls

You can choose to keep useful memory while opting out of training. You can also disable memory if you prefer less personalization. Choose each setting for its own purpose.

Business, Enterprise, Edu and Healthcare workspaces are not used for model training by default. Workplace access and retention policies still apply, so use your organization's approved tools for work information.

Start a New chat, choose Chat if a Chat/Work switch appears, then select Temporary chat at the top right. Before sending your first message, open Personalized and choose Unpersonalized if you don't want existing personal context used.

You cannot change that personalization choice after the conversation starts. Limited safety-relevant context can still apply in rare, high-risk situations.

Temporary chats stay out of history unless you save them. OpenAI may retain a copy for up to 30 days for safety. Saving one turns it into a regular chat that follows your normal memory and model-training settings.

Remove identifiers before you upload

Temporary Chat still sends your message to OpenAI. Replace names, account numbers, addresses and unnecessary identifiers on your own device first. Keep passwords and API keys out of your prompts. My guide to using ChatGPT and Claude safely covers more habits like this.

Prompt 4: A second redaction check

I have already replaced private identifiers in the text below with placeholders. Check for remaining identifying details, including combinations of clues that could identify a person or organization. Don't try to reconstruct the original identities.

Flag anything I should remove before we work on the actual question. If a detail is unnecessary, suggest a generic replacement.

[PASTE ONLY THE ALREADY-REDACTED TEXT]

Open Settings > Plugins. Select an installed plugin and review its connected accounts and permissions. Some versions call this section Apps. Keep connections you use and disconnect ones you no longer want ChatGPT to access.

Use narrower permissions when the app offers them. Folder-level access or read-only access is not available for every connection. Review the actual permission screen instead of assuming it is restricted.

Prompt 5: Review available connections

Help me review the apps and plugins available to this chat. For each one, report only the access you can verify: what it can read, what actions it can take, and whether an account connection is confirmed.

Separate verified access from unknown permissions or unavailable information. Don't treat a tool listing as proof that you can access an account. Tell me which settings I need to inspect myself and suggest narrower access where it is actually supported.

Do not open private records, send messages, change permissions, or disconnect anything during this review.

Information shared with apps is a separate item under Data controls. Review it as well as your plugin connections. Third-party services have their own data practices, and ChatGPT's training switch does not change those policies.

Shared links and account security

Open Settings > Data controls > Shared links to review links you created. Remove ones you no longer want accessible. Removing a link cannot recall screenshots or copies someone already saved.

Open Settings > Security and login to review the sign-in protections available to you. Use a passkey or multi-factor authentication where offered, and review signed-in sessions for unfamiliar access.

Get an account-data copy

Open profile > Settings > Data controls > Export data. Select the export control and review its confirmation. Request the export, then wait for it to arrive before deleting anything you want to keep.

Exports can take up to seven days. The download link expires after 24 hours. Download it while signed in to the same account, open the ZIP, and check that the conversations you need are included. Store it privately because it brings a lot of personal information together. For a full walkthrough, see my export guide.

The export contains available chat history and relevant account data. It cannot recover already-deleted chats. Managed workspaces may restrict self-service exports. For data access, deletion requests, or an unavailable export option, use the OpenAI Privacy Portal or your workspace administrator.

Location, marketing and ads

Review location permissions if you use local recommendations. Turning off precise location access does not erase places you mentioned in chats. Approximate location and contextual clues can still influence local answers. A confident location guess is not proof of live tracking.

In Data controls, review Marketing privacy if offered. Ad personalization may have separate controls on eligible accounts. These choices are separate from memory and model training, so review the options shown in your own account.

Prompt 6: Finish the checklist

Use this after you have checked the settings. You can answer with on/off status and general topic labels. You don't need to paste the private details you removed.

Help me finish my ChatGPT privacy review. Here is what I personally checked and changed: [LIST SETTINGS AND GENERAL TOPICS ONLY].

Compare my notes with memory, original and archived chats, custom instructions, files, model training, Temporary Chat, connected apps, shared links, location permissions, and account security.

Make two lists: checked based on my report, and still needs my review. Don't claim you verified account settings you cannot inspect. Give me the next three useful actions without repeating private information.

Common Mistakes and Easy Fixes

Keep the review useful

You are deciding what information is worth keeping. Writing preferences can be useful. An old home address, a private client conversation or a detail about someone else's health may not be.

If you accidentally pasted a password or secret key, deleting the conversation is only part of the response. Replace or revoke that credential through the service that issued it.

Check again after changes

Revisit your settings when you connect a new app, change devices or notice an answer using context you didn't expect. Ask what source it used and review that source directly.

Quick Reference: Your Privacy Checklist

Before sharing a document

Remove unnecessary identifiers locally. Use placeholders like Client A or Company B. Check that you are allowed to share the remaining material, then choose the appropriate chat mode.

Use it every day. Decide what it gets to keep.

Published October 14, 2026 · Checked October 2026