Skip to content
← All guides

The AI Privacy Checklist

Your AI remembers more than you think, and deleting the chat doesn't make it forget. This is the 5-step checklist to see everything it knows about you, wipe what it shouldn't have, and stop it from saving more. Built for Claude, with the same steps for ChatGPT and Gemini.

Why this matters

Zapier surveyed 1,005 workers in May 2026. 79% have typed sensitive information into AI: performance reviews, salaries, customer IDs, client contact lists. 31% have given it login credentials or API keys. (Zapier's write-up.)

That's not a reason to stop using AI. Memory is what makes it good: it knows your business, how you write and what you're working on, so you don't explain yourself every time. The goal is to keep the useful stuff and get rid of what it should never have had.

Here's what happens to the stuff you type:

  • Your chats
    They sit in your history until you delete them.
  • Your memory
    Claude pulls details out of your chats and saves them as memory. Deleting the chat doesn't delete those memories.
  • Their servers
    With "Help improve Claude" on, your chats can be kept for up to 5 years and used to train future models. With it off, deleted chats are gone from Anthropic's servers within 30 days.

What never goes into AI

The easiest privacy fix is the stuff you never type. Keep these out of every chat, on every AI:

  • Passwords, API keys and login details. Ever. If a tool needs access, use its proper connection screen, not the chat box.
  • Full card or bank numbers, and ID numbers (SSN, passport, driver's license).
  • Health and therapy details you wouldn't want saved with your name on them.
  • Other people's private info: a client's contact list, a coworker's review, a friend's situation. It's not yours to share.

Need help with a document that has this stuff in it? Swap the details out before you paste: "Client A" instead of the name, "$X" instead of the salary, "[address]" instead of the address. The AI gives you the same quality answer either way.

1. See everything it knows about you

  1. Open your memory. In Claude, go to Settings > Your Claude and click Manage next to Your memory. On the older settings layout it's Settings > Memory (or Settings > Capabilities on the oldest one).
  2. Read every topic. Everything Claude remembers is listed there, grouped into topics. Click into each one.
  3. Then ask it directly. Open a new chat and paste this. It's the question from the video, and the answer is usually eye-opening.
Prompt 1: The question
What do you know about me that I never actually told you?

Want the full picture? This one makes it list everything, sorted, with the sensitive stuff flagged.

Prompt 2: The full privacy audit
I'm doing a privacy check on my account. List everything you remember about me, from memory and from anything you can pull from my past chats.

Don't summarize, group or leave anything out. Put each item on its own line, and sort them into:
1. Things I told you directly
2. Things you figured out on your own (guesses about me, my habits, my job, my health, my money, my relationships)
3. Anything about other people (names, details about my family, friends, coworkers or clients)

Then flag every item that's sensitive: passwords, API keys or login details, card or bank numbers, ID numbers, my home address, health or therapy details, my income or debts, and private details about other people.

2. Delete anything you don't want it knowing

  1. Go back to your memory (Settings > Your Claude > Manage).
  2. Open the topic with the thing you want gone.
  3. Click Delete to remove the whole topic, or use the edit icon to remove just one part and keep the rest.

Faster way: tell Claude in any chat. Memory updates apply to your next conversation.

Prompt 3: Make it forget
Forget these things about me and remove them from your memory completely:
- [the first thing, e.g. "my home address"]
- [the next thing, e.g. "anything about my therapy sessions"]
- [the next thing, e.g. "my salary"]

When you're done, list what you removed, then list everything you still remember about me so I can check nothing sensitive is left.

Want a totally clean slate? Turn the memory setting off and reset your memory. Claude starts from zero. Turn it back on whenever you want, and it rebuilds from your chats from then on.

3. Delete it from memory too (the one nobody knows)

This is the big one. Deleting a chat does NOT delete what Claude remembered from it. Anthropic says it in their own help docs: when you delete a conversation, the memory entries made from it aren't removed (Claude Help Center).

So if you told it something private and then deleted the chat, it can still remember it.

  1. Delete the memory first. Find it in Settings > Your Claude > Manage and delete it (or use Prompt 3).
  2. Then delete the chat. Sidebar, open the chat menu, Delete.
  3. Check it's gone. Start a new chat and ask what it knows about that topic.

4. Turn off "Help improve Claude" (optional)

  1. Go to Settings > Privacy.
  2. Turn off "Help improve Claude".
  • On Your chats can be kept for up to 5 years and used to train future models
  • Off Anything you delete is wiped from Anthropic's servers within 30 days

It's optional because some people are fine helping train the model. But a lot of people clicked through the pop-up when this setting came out and have no idea it's on. Check yours.

5. Go incognito for anything private

About to ask something private, like a health question, a money situation or a hard conversation? Use incognito.

  1. Start a new chat (outside a project).
  2. Click the little ghost icon in the top right.
  3. Ask away. Nothing you say there gets saved to your history or added to memory.

Want your normal chat history but no memory for one task? Before you send the first message, open the + menu and turn memory off for that chat.

Already pasted a password or API key?

Deleting the chat doesn't un-share it. Treat it as leaked and fix it at the source. This takes 5 minutes:

  1. Change the password on that account, and anywhere else you used the same one.
  2. Turn on two-factor login for that account if it's not on already.
  3. API key? Go to the dashboard that made it, revoke (delete) the old key and create a new one. The old one stops working the second you revoke it.
  4. Then clean up the AI side: delete it from memory (step 3), then delete the chat.

Stop it from saving the wrong things

You can tell Claude what never goes into memory. Paste this once and it applies to every chat after.

Prompt 4: The memory guardrail
Remember this rule about me for every future chat:

Never save these to memory, even if I mention them: passwords, API keys, login details, card or bank numbers, ID numbers (like my SSN or passport), my home address, health or therapy details, my exact income, and private details about other people.

If I paste something like that by accident, tell me right away so I can delete the chat and change it.

The same checklist for ChatGPT and Gemini

Use more than one AI? Run the checklist on every one. The settings live in different places, but it's the same 5 moves. These apps rename settings often, so if a label doesn't match exactly, search the settings for the key word (Memory, Training, Activity, Temporary).

StepChatGPTGemini
See what it knowsSettings > Personalization > Memory > Manage, then ask it Prompt 1Settings > Saved info, then ask it Prompt 1
Delete itDelete items in Manage memories, or tell it "forget that"Delete items in Saved info, or tell it "forget that"
Delete it there tooDeleting a chat doesn't delete saved memories. Clear both.Delete the chat and its entry in Activity
Stop the trainingSettings > Data controls > turn off "Improve the model for everyone"Settings > Activity > turn off Keep Activity
Go incognitoTemporary Chat (top of a new chat)Temporary chat (in the chat menu)

OpenAI's own breakdown of its settings is here.

Your 2-minute monthly check

Memory keeps growing. Put this on your calendar once a month:

  1. Ask what's new. Paste the prompt below.
  2. Delete anything sensitive it flags, in memory first, then the chat.
  3. Glance at Settings > Privacy to make sure "Help improve Claude" is still how you want it.
Prompt 5: The monthly check
Monthly privacy check. Compared with what you knew about me before, what new things have you saved about me recently? List them, and flag anything sensitive so I can delete it.

The checklist

  1. See what it knows: Settings > Your Claude > Manage + Prompt 2.
  2. Delete it: delete the topic or use Prompt 3.
  3. Delete it from memory too: deleting the chat isn't enough. Memory first, then the chat.
  4. Stop the training: Settings > Privacy > "Help improve Claude" off.
  5. Go incognito: the ghost icon for anything private.