Every setting that controls what ChatGPT keeps about you, what OpenAI can do with it, and how to shut the door. Verified click by click on a real account in September 2026.
The plan · 3 steps
Before you start
1 switch
turns training off
30 sec
to do the whole lockdown
3 settings
that matter most
This guide is not an argument for using ChatGPT less. I use it every single day and I am not stopping. It is an argument for knowing exactly what you are handing over, and taking back the parts you never meant to give. There are three settings that do almost all of the work, and a handful more worth ten seconds each. I went through every one of them on my own account the week I wrote this, so the buttons are where I say they are.
What ChatGPT Actually Keeps
Unless you have changed a setting you have probably never looked at, everything you type into ChatGPT can be used to train the model. That is the default on a personal account. Your work drafts, your health questions, the message you rewrote four times at 2am, the client email you pasted in to fix the tone. All of it is eligible.
| What | Where it lives | Who can see it | What you can do |
|---|---|---|---|
| Your chats | Chat history in the sidebar | You, and OpenAI for training if the switch is on | Turn training off. Delete or archive chats |
| Memory | Personalization, then Memory | ChatGPT uses it in every new chat | Read it, edit it, or turn it off |
| Shared links | Data controls, then Shared links | Anyone with the link | Delete old ones |
| Plugin access | Settings, then Apps (or Plugins), then Connected accounts | Each plugin you connected | Remove what you don't use |
Rick's tip
Why the default is on: training on real conversations is how the model gets better at the things people actually ask. OpenAI says it takes steps to protect privacy before anything is used. The point is not that this is sinister. The point is that it is a choice, and right now it is being made for you.
Turn Off Training
Open ChatGPT on the web (this one lives in the web settings, and it applies to every device on your account). Click your name in the bottom left corner, then choose Settings, then Data controls in the left-hand list. The very first line says Improve the model for everyone. Click it.
A small panel opens called Model improvement. The first switch is the one you want. Turn it off. Underneath it are two more switches, one for your audio recordings and one for your video recordings from Voice Mode. Turn those off as well. Click Done.
Rick's tip
On the phone app the path is the same idea: tap your name or the two lines in the corner, then Settings, then Data controls, then Improve the model for everyone. The switch looks identical.
That is the big one. From this moment on your conversations stop being training data. It does not change what ChatGPT can do for you, it only changes what OpenAI can do with it. You will not notice any difference in the answers.
One exception worth knowing: if you hit the thumbs up or thumbs down on a reply, the whole conversation around that feedback can still be used, switch or no switch. So do not rate the sensitive ones.
Rick's tip
Heads up: this does not reach back in time. Anything already used before you flipped the switch stays used. And a chat you delete can take up to thirty days to fully disappear from OpenAI's systems. Everything from today forward is yours, and that is the part you control.
Read And Clean Out What It Remembers
Click your name again and choose Personalization. Scroll down until you reach the section called Memory. Next to Memory summary there is a Manage button. Click it.
What opens is a written overview of you. Not a list of bullet points, an actual page of prose: who you are, what you do, how you like things written, what you have been dealing with lately. Read the whole thing once. Most people find at least one line they did not remember telling it.
At the bottom of that page is a box that says Ask or update. That is how you edit it. You type an instruction and it changes the summary. Some things worth typing:
If you want the nuclear option, click the three dots next to the Memory summary title and choose Delete and turn off memory. ChatGPT will treat every chat as if it has never met you. The trade-off is real though: memory is what makes it good at your writing style and your recurring tasks. If you want to see exactly what it keeps and how to edit it, my ChatGPT memory guide covers the whole thing. My advice is to prune it, not kill it.
Use Temporary Chat When It Matters
Start a new chat and look in the top right corner for the pill-shaped Temporary button. Click it and the chat opens in a mode where nothing is saved to your history, nothing is added to your memory, and nothing is used for training while it stays temporary. When you close it, it is gone (OpenAI can keep a copy for up to 30 days for safety checks, but it never shows up anywhere you can see it).
When you start one, ChatGPT now asks whether the temporary chat should be personalized (it can use your memory and custom instructions) or non-personalized (a blank slate). Check which one is selected before you type, because non-personalized is the one you want for anything sensitive. You cannot switch a chat between the two after it starts. If a temporary chat turns out to be worth keeping, there is a Save option that turns it into a normal chat, and from then on it follows your regular settings.
Think of it as incognito mode for ChatGPT. Use it for anything you would rather not have sitting on a record: a medical question, a legal problem, a money situation, a conversation about someone else, the argument you are trying to win with your partner. It is the same ChatGPT with the same abilities. It just forgets you afterwards.
Rick's tip
Do this right now. Open ChatGPT, do Step 1, and come back. It takes thirty seconds and you will never think about it again. Then read your memory summary tonight when you have five minutes. That one is worth doing slowly.
Every Other Data Controls Setting
Once training is off, the rest of the Data controls page is worth a look. None of these take more than ten seconds and most people have never scrolled down far enough to see them. If you want the rest of the settings tour, my hidden ChatGPT settings guide covers the five most people never turn on.
| Setting | What it does | My call |
|---|---|---|
| Location | Lets ChatGPT use where you are for local answers | Leave off unless you use it for restaurants and weather |
| Information shared with apps | The partners ChatGPT shares your name and email with, like booking sites | Open it. Remove anything you don't recognize or don't use |
| Shared links | Every chat you've ever turned into a public link, still live | Delete the old ones. People forget these exist. Deleting the chat itself removes its link too |
| Archived chats | Chats you hid from the sidebar. Still stored | Fine to keep, but know they're there |
| Archive all / Delete all | One button for the whole history | Export first (below), then decide |
| Export data | Emails you a full copy of everything (can take up to 7 days, and the download link expires 24 hours after it arrives) | Do this once a year. It's your record |
| Marketing privacy | Two switches for ad measurement and personalized marketing | Both off |
| Ads controls (Free and Go only, its own Settings section) | Personalize ads, Past chats and memory, and Delete ads data | Both switches off, and hit Delete ads data |
Shared Links, The One People Forget
Every time you have hit Share on a chat and sent someone a link, that link is still live unless you deleted it. Go to Data controls, find Shared links, click Manage, and look at the list. If there is a chat in there from a year ago that had your address or a client's details in it, anyone who still has the link can still read it. Delete the ones you don't need, or use More actions, then Delete all shared links, and start clean. Deleting a link does not delete the chat itself, and it does not pull back copies people already made.
Export Before You Delete
If you are tempted by Delete all, hit Export first (Data controls, then Export data, then Export, then Confirm). OpenAI emails you a link to a file with every conversation you have ever had. It can take up to a week to arrive and the link only works for 24 hours, so download it as soon as you see it and keep it somewhere safe. Export is web only, and it cannot recover chats you already deleted. Then delete with a clear conscience, because the good prompts you wrote eight months ago are not gone, they are just not on OpenAI's servers anymore.
What Plugins And Connectors Can See
The moment you connect ChatGPT to Gmail, Google Drive, your calendar, Slack or any other app, you have given it a second door into your life. This is where most of the genuinely useful things ChatGPT does come from, so the answer is not to avoid connectors. The answer is to know what each one can reach.
Go to Settings, then Apps (it says Plugins on some accounts). Pick any app and open Connected accounts, and you will see exactly which of your accounts it can reach. The three dots next to an account is where you disconnect it. Read that list the way you would read the permissions screen on a new phone app. A separate line under Data controls called Information shared with apps covers the partners ChatGPT hands your name and email to for things like restaurant bookings, and it is worth a look for the same reason.
Disconnecting an app does not delete the chats that already used its data. Those stay until you delete the chats.
The least-access rule: Connect the smallest thing that does the job. If you want ChatGPT to summarize one folder of documents, connect that folder, not your whole Drive. If you want it to draft replies, give it read access to your inbox and send the emails yourself. Every connector you add is one more place your data flows, and one more thing to remember when you eventually want to clean up.
What You Should Never Paste In
Turning training off fixes the biggest problem, but it does not make ChatGPT a vault. Your chats are still stored on OpenAI's servers, still visible to you on every device, and still subject to the usual things that happen to accounts, like someone guessing your password. So there is a short list of things I do not put into a regular ChatGPT chat, ever, regardless of settings.
| Never paste this | Why | Do this instead |
|---|---|---|
| Passwords, API keys, bank logins | Stored in plain text in your history | There is no reason to. Ever |
| Card numbers, ID numbers, passport details | Same. And they live in your export forever | Describe the situation without the number |
| Other people's private details | It's their data, not yours to hand over | Swap names for Client A, my sister, my landlord |
| Contracts with names and amounts | Business confidentiality, and often your own NDA | Redact first (prompt below), then paste |
| Medical records with your name on them | The most sensitive category there is | Temporary chat, and strip the name and date of birth |
| Anything you'd be uncomfortable seeing in a leak | Because that is the honest test | Temporary chat, or don't |
The Redaction Trick
Most of the time the useful part of a document is the structure, not the names. A contract review works just as well with the parties called Company A and Company B. Here is how I strip a document before it goes anywhere near a chat that gets saved:
Five Prompts To Audit Your Own Account
Settings tell you what is allowed. These prompts tell you what is actually in there. Run them in a normal chat, because you want ChatGPT to use its memory to answer.
Lock The Door Too
Everything above is about what OpenAI keeps. The other half of privacy is who can get into your account. Three settings, all under Settings, then Security, all worth two minutes.
If you have health records or bank accounts connected, those get their own treatment: connected health data is never used for training regardless of your other settings, and Finances keeps its own separate memory you can view and delete from the Finances page.
Common Mistakes
These are the ones I see over and over, including a couple I made myself.
The Quick Reference
Bookmark this page. Everything above, in the order you should do it.
| # | Do this | Where | How often |
|---|---|---|---|
| 1 | Turn off Improve the model for everyone (plus the two voice switches) | Name, Settings, Data controls | Once |
| 2 | Read your Memory summary and prune it | Name, Personalization, Memory, Manage | Every couple of months |
| 3 | Use Temporary chat for anything sensitive | Pill-shaped Temporary button, top right of any chat | Every time it matters |
| 4 | Turn off both Marketing privacy switches | Data controls, Marketing privacy | Once |
| 5 | Review your connected apps and their accounts | Settings, Apps (or Plugins), Connected accounts | Whenever you add a connector |
| 6 | Delete old Shared links | Data controls, Shared links, Manage | Twice a year |
| 7 | Export your data | Data controls, Export | Once a year |
| 8 | Redact documents before pasting | Temporary chat, redaction prompt | Every time |
Rick's tip
The whole thing in one line: training off. Memory pruned. Temporary chat for the sensitive stuff. Nothing pasted that you wouldn't want to see in a leak. That is the entire lockdown, and steps one to three take under a minute.
Use it every day. Just decide what it gets to keep.