Brooke Wright
Brooke Wright · @wright_mode
FREE CHEAT SHEET

5 things I’d never type into AI

My cheat sheet for keeping money, health and other people’s details out of AI chats, plus how to switch off the share links you’ve already made.

The 5 things + swaps Copy-paste scrub prompt ChatGPT & Claude steps Built by Brooke
Something went wrong. Please try again.

No spam. Unsubscribe anytime.

The 5 things I’d never type into AI (and what to do instead) Jump to the 5 → Join the Membership
Wright Mode — Free Cheat Sheet

5 Things I’d Never Type Into AI

My cheat sheet for what stays out of ChatGPT, Claude and every other AI chat, what to do instead, and how to switch off the share links you’ve already made.

The 5 things + swaps Copy-paste scrub prompt Turn off old share links Aussie examples

What's inside

These five stay out of my AI chats, full stop. I can’t always predict which chat I’ll end up sharing, or what an app will do with sharing next week (hi, Meta). So I don’t risk it. Each one comes with a swap, so you still get the help.

1. Money stuff2. Health stuff3. Other people’s details4. Passwords & logins5. Confidential business stuff
1

Money stuff

Bank statements, BSB and account numbers, card numbers, your TFN, payslips, tax returns, loan applications, screenshots of your banking app.

Why it’s risky: Numbers plus your name is a scammer’s starter kit, and a share link puts it in front of whoever the link gets forwarded to.
Do this instead: Mask the numbers (BSB XXX-XXX, account ending XXXX), round the amounts (“about $40k revenue”) or describe the situation instead of pasting the statement.
2

Health stuff

Your Medicare number, diagnoses, test results, medications, mental health details, your kids’ health info, a staff member’s medical certificate.

Why it’s risky: Australian privacy law treats health info as “sensitive information” for a reason. Once it’s out, you can’t take it back.
Do this instead: Keep it general (“a chronic condition that means I work 20 hours a week”), leave out names and Medicare numbers, and keep anything medical out of chats you share.
3

Other people’s details

Client names with their contact details, a complaint email with the signature block still on, screenshots of DMs or group chats, staff info, your kids’ school details.

Why it’s risky: It isn’t yours to share, and they never agreed to end up on a public page.
Do this instead: Swap names for roles (“Client A”, “my VA”, “Supplier B”), delete signature blocks and phone numbers before you paste, and summarise the conversation instead of screenshotting it.
4

Passwords, logins and keys

Passwords, 2FA and verification codes, recovery codes, API keys, screenshots of a login screen, “here’s my login, can you check it?”

Why it’s risky: Whoever reads it can walk straight into your account. No hacking required.
Do this instead: Never paste them. Write [API KEY] or [PASSWORD] as a placeholder when you need setup help. Already pasted a key? Make a new one today and delete the old one.
5

Confidential business stuff

Client contracts, anything under an NDA, launches and pricing you haven’t announced, staff performance or HR issues, leave calendars, supplier quotes, a client’s numbers.

Why it’s risky: You could be breaking a contract or someone’s trust, and a share link can’t be un-sent.
Do this instead: Describe the situation, not the document (“a client contract with a vague scope and 30-day payment terms”), paste only the clause you need with names removed, and check the NDA before you paste anything it covers.

The scrub-it-first prompt

Honest disclaimer: whatever you paste, the AI sees. This prompt won’t make it OK to paste a bank statement. Use it as a second pass. Strip the big stuff by hand first (numbers, passwords, health details), then let it catch the names and small details you missed.

In ChatGPT, run it in a Temporary Chat. It stays out of your history and isn’t used for training (OpenAI may keep it for up to 30 days for safety).

I'm about to use the text below in an AI chat I might share. Anonymise it first. Replace: - People's names with roles (Client A, Staff member 1, Supplier B) - Business names, street addresses, phone numbers and email addresses with [BUSINESS], [ADDRESS], [PHONE], [EMAIL] - Any account, card, BSB, TFN, Medicare or reference numbers you spot (I should have removed these already) with XXXX - Health details, dates of birth and exact dollar amounts with general versions (e.g. "a health condition", "around $5k") Keep the meaning and structure the same so I can still work with it. Then list anything that still looks identifying or sensitive, so I can check it by hand. Text: [paste text you've already stripped of passwords, card numbers and health details]

Then paste the cleaned version into the chat you actually work in (or share), not the original.


A shared chat is a public page. Anyone with the link can read it, and links get forwarded to people you’ve never met. So my rule is simple: if you wouldn’t put it on a billboard, don’t put it in a chat you’ll share.

What a share link feels like

  • “I only sent it to my VA”
  • “She won’t send it on”
  • “It’s just a chat, who’d copy it?”
  • “I can always delete it later”

What it actually is

  • Anyone with the link can open it
  • It can be forwarded or posted anywhere
  • People can copy or screenshot it
  • Turning it off doesn’t delete their copies

It keeps happening

1
Meta Muse, October 2026. Shared Muse chats were public links. Security researchers found around 6,800 of them in public DNS records (part of the internet’s public address book), opened a few hundred, and found things like a small-business loan application, a bank fraud alert with partial account details and a staff holiday calendar. Meta paused sharing on 20/10/2026.
2
ChatGPT, 2025. Shared chats turned up in Google search results. OpenAI removed the option that made them discoverable.
3
Claude, 2026. WIRED reported shared Claude chats showing up in Google and Bing. Anthropic now tells search engines not to index shared chats.

Even the AI companies say it

OpenAI’s own help page: “Do not include health information, financial details, passwords, account numbers, or other sensitive personal information in shared content.” (source)

Anthropic’s: “The simple rule: treat a public link as public. If you wouldn’t post the contents on the open web, don’t put them behind a public link.” (source)



This one’s a different setting from sharing. It stops your new chats being used to train future models.

ChatGPT

Settings → Data controls → Improve the model for everyone → turn it off → Done. In the app: sidebar → your profile icon → Data controls.

Claude

Settings → Privacy → switch off the model-improvement toggle (Anthropic’s help page calls it “Help Improve our AI models”). Same path in the mobile app.

Both only cover new chats from here on. Anything already used in training stays there. And it doesn’t replace switching off your share links (OpenAI says changing this setting doesn’t delete them), so do both.


Ready to go deeper?

Where to go from here.