Your ChatGPT Privacy Audit

Delete your ChatGPT data: the full checklist

Look, a tidy chat list doesn't tell you what's been removed. Use these checks to clean up the information you want gone, then set what ChatGPT can use going forward.

Checked against OpenAI's help pages on 18/09/2026. Menu names can vary by account. Start on the web for the full Library controls.

1

Clear memories and remembered chat history

Open Settings → Personalization → Memory. Your account may show Manage memories, Saved memories or a Memory summary.
Delete the saved details you want removed. If your summary offers Delete and turn off memory, use it when you want to clear that remembered information and switch Memory off.
If you have Reference chat history, switch it off to remove information remembered from earlier chats. OpenAI schedules that remembered information for deletion within 30 days. The original chats and saved memories need separate deletion.
Remove the same information from the chats, files and other sources below. Check custom instructions for details you've written there too.

Turning Memory off alone doesn't clear past conversations. Switching it back on can let ChatGPT learn from chats you've kept. “Don't mention this again” also leaves the original source in place. Deleted-memory logs may remain for up to 30 days for safety and debugging. OpenAI's memory controls.

2

Delete chats, including archived ones

Open the chat's ⋯ menu → Delete, then confirm.
Check Settings → Data controls → Archived Chats → Manage and delete the conversations you want removed there too.
To clear the whole chat history, use Delete all chats in Data controls. This includes chats inside projects.

Archive keeps the conversation in your account. Delete removes it from your view and starts the deletion process. Deleted chats cannot be recovered, so save anything you need first. OpenAI's chat deletion instructions.

3

Check saved files and connected apps

Library files

Open Library in the sidebar. Select each uploaded or generated file you want removed, then Delete.
If Recently deleted appears, review it and use Delete forever for files you want permanently removed.

A file saved separately in Library survives deletion of the chat that used it. Connected Google Drive files stay in Google Drive; manage or delete those originals in Google Drive itself. OpenAI's Library guide.

Connected apps

Open Settings → Apps or Plugins, then select the app.
Review Connected accounts or the connection menu. Choose Disconnect for each account you no longer want ChatGPT to access.
Delete chats, saved files and memories containing information already brought in from those apps.

Disconnecting stops future access through that connection. It doesn't clear information already saved in ChatGPT. Other connected accounts can remain active. OpenAI's connection controls.

Disconnecting Google Drive also leaves the original Drive files intact. A separately managed workplace index needs your administrator's attention. OpenAI's Google app data controls.

If you use projects or custom GPTs

Check files stored in projects and knowledge uploaded to custom GPTs. They have their own retention rules; clearing ordinary chats isn't a complete check of those files. OpenAI's file-retention guide.

4

Turn off future model training

Open Settings → Data controls.
Turn Improve the model for everyone off.

This account-wide setting stops new conversations being used for training. Your chats can still appear in history. It doesn't do the cleanup steps above. OpenAI's Data Controls FAQ.

If you submit thumbs-up or thumbs-down feedback, OpenAI says the conversation attached to that feedback may still be used for training, even after opting out. The opt-out applies to new conversations; don't treat it as a promise to undo previous training. Business workspace content isn't used for training by default. OpenAI's model-training explanation.

Want to delete the account too?

This action is separate and permanent. It removes access to OpenAI services through that account, including ChatGPT and the API.

On the web, open Settings → Account → Delete account.
Follow the identity confirmation and permanent-deletion prompts.
If you can't sign in, submit a verified deletion request through the OpenAI Privacy Portal.

Check any Apple or Google Play subscription first and cancel billing you no longer want. Managed business, education and healthcare accounts need help from their organisation. OpenAI says account data is deleted within 30 days, with limited information retained longer where law requires or permits it. OpenAI's account-deletion guide.

What “deleted” means here

A chat disappearing from your screen is the first visible step. OpenAI schedules deleted chats for permanent removal within 30 days, with exceptions for data already de-identified and separated from your account, or information retained for security or legal obligations. Workspace retention policies can also apply. OpenAI's retention policy.

Library files have a similar deletion window and exceptions. OpenAI's Library deletion details.

How ChatGPT builds a profile of you

Every time you chat with ChatGPT, it's taking notes in the background. Two kinds, actually — and understanding the difference is the whole point of this audit.

🗂️ What it stores (memory)

Your name, your business, your city
How you like answers written
Projects and people you mention often
Preferences you've told it directly
Facts it decides are worth keeping

🔍 What it infers (guesses)

Your age, income and situation
Your mood and stress levels
What you avoid asking about
How you actually come across
Things you've never once said out loud

The stored stuff is the "memory" feature — you can see it, and you can delete it (section three shows you how). The inferred stuff is different. It's the picture the model builds up on its own from your writing style, your questions, and the topics you circle around. You never typed it, but it's there, shaping every reply you get.

Why this matters for you

None of this is sinister on its own — memory is what makes ChatGPT feel like it "gets" you. But most people have never seen the file, and never chosen what's in it. This audit puts you back in the driver's seat: see it all, keep what's useful, delete what isn't.

The 4 prompts

Open ChatGPT (works best on an account you've used for a while) and run these one at a time. Read each answer properly before moving to the next — the goal is to actually see what's been collected, not rush through. Hit copy, paste, send.

1 · The full file

Lay out everything you've stored and remember about me as one complete list. Include names, places, work habits, preferences — anything at all. Don't summarise and don't leave anything out. I want to see the full file you've built on me across all of our conversations.

2 · The inferred profile

Tell me everything you've figured out about me that I've never actually stated — the things you've inferred from how I write, what I ask, and what I avoid. My likely age, income and life situation. Be specific, not vague.

3 · The blind spots

Based on everything you know about me, name the personality traits I most likely don't see in myself — my blind spots, my insecurities, and the way I actually come across versus how I think I do. Be blunt and be specific.

4 · The predictions

Using your profile of me, predict how I'd handle things I've never told you about — stress, money, conflict, risk, and the decision I'm most likely to make next. For each prediction, tell me how confident you are and what behaviour of mine you based it on.

A word before you read the answers

Prompts three and four can hit a nerve — the model is reflecting patterns back at you, and it's often startlingly on the money. Treat it as a mirror, not a verdict. It's reading text, not your soul. If something feels off, it probably is.

View & delete your memories

The prompts show you what ChatGPT knows. This is where you start taking control of it. Your saved memories live in one settings screen — here's the exact path to find them and clear out anything you didn't mean to hand over. The full checklist above covers the other controls too.

1

Open your profile menu

Click your name or profile picture (bottom-left on the app, top-right on the web) to open the menu.

2

Go to Personalisation

Open Settings, then find the Personalisation tab. This is where ChatGPT keeps everything it's learned about you.

3

Find Memory

Inside Personalisation you'll see a Memory section, with a toggle and a "Manage memory" option.

4

Manage memory

Tap "Manage memory" to see the full list of saved memories, written out one by one in plain language.

5

Delete the ones you don't want

Hover or tap the bin icon next to any memory to remove it individually — or clear the whole lot in one go. The memory is removed from use; limited logs may be retained for up to 30 days for safety and debugging.

🧭 The path in one line

Profile menu → Settings → Personalisation → Memory → Manage memory → delete the ones you don't want. Bookmark this — you'll want to come back and tidy it every so often.

One thing to know

Deleting a memory stops ChatGPT using that fact going forward — but it won't scrub it from the raw text of old chats. If a conversation was genuinely sensitive, delete the whole chat too, not just the memory — the full checklist above shows you how, and why archiving it won't do.

Lock it down

Cleaning out old memories is the tidy-up. This is the ongoing habit — a short checklist so you decide what ChatGPT keeps from here on, instead of it deciding for you.

Turn memory off if you'd rather it saved nothing

Assuming "it forgets" — it doesn't, unless you tell it to

Use a temporary chat for anything sensitive

Pasting client data, passwords or bank details

Keep private personal details out of prompts

Sharing other people's information without their say-so

Check your memory list every month

Leaving old, wrong or awkward memories sitting there

Temporary chats are your friend

Need to ask something you don't want remembered? Start a temporary chat (the toggle sits near the top of a new conversation). It runs like a private window — it does not create memories or appear in history. A copy may be retained for up to 30 days for safety. Keep confidential client information out unless you have permission and suitable controls.

The golden rule

Never paste anything into ChatGPT you wouldn't be comfortable seeing again later — client records, passwords, financial details, other people's private information. If in doubt, leave it out or strip the identifying bits first.

Get more out of it, safely

Here's the part people miss: locking down your privacy and getting better answers aren't opposites. A few small habits give you both — a model that works harder for you, on details you actually chose to share.

1

Set custom instructions instead of hoping it remembers

In Personalisation, use custom instructions to tell ChatGPT how you like answers written — tone, length, your role. It's deliberate context you control, not something it guessed.

2

Use Projects for ongoing context

For something you work on regularly, create a Project and keep the background there. It gives the model what it needs for that job without saving it to your permanent memory.

3

Review your memory monthly

Make it a five-minute habit. Open Manage memory, delete anything out of date or off, and keep the useful bits. A clean memory means sharper, more relevant answers.

4

Give context on purpose, not by accident

When a task needs detail, paste it in for that conversation rather than letting it drip into memory over months. You get a great answer and stay in control of what sticks.

The mindset that changes everything

ChatGPT is brilliant when you treat it like a tool you direct, not a diary you confide in. Choose what you share, tidy what it keeps, and it becomes genuinely useful — without knowing more about you than you'd want any app to.

One honest caveat

Settings and menu labels shift as ChatGPT updates. If a name has moved since I wrote this, the path is still the same idea — look under your profile settings for Personalisation and Memory. And the inferred-profile answers are educated guesses, not facts. Read them as a mirror, not a diagnosis.