What not to share with ChatGPT and other AI chatbots
AI chatbots are useful exactly because you can paste in real-life material: an email thread, a contract, a spreadsheet. That's also how sensitive details slip out. Google's own advice for Gemini is not to enter information you wouldn't want a human reviewer to see, and that's a good rule for every AI tool (why).
Here's what to keep out of your prompts, and what to do instead.
The never-paste list
1. Identity numbers
Passport, driver's license, social security and national ID numbers. These are the hardest details to change if they leak, and an AI never needs the real number to help you fill in a form or write a letter.
2. Financial details
Credit and debit card numbers, bank account numbers, IBANs, and full statements. If you want help understanding a statement, replace the numbers first.
3. Passwords, keys and access codes
Passwords, API keys, recovery codes, one-time codes and private keys. If a password or key ever ends up in a chat, change it.
4. Health information
Diagnoses, test results, medications and therapy notes, yours or anyone else's. If you want a medical term or a lab report explained, remove the name, date of birth and record numbers first.
5. Other people's contact details
Names together with emails, phone numbers or home addresses. Pasting a whole email thread usually shares several people's details without their knowledge.
6. Client, patient and employee data
Case files, payroll lines, performance reviews, patient notes and customer records. At work this isn't just a privacy risk, it can break your company's rules or your professional obligations (guide for work use).
7. Confidential business material
Unreleased plans, contracts under negotiation, financial results before they're public, and source code. In 2023, Samsung banned generative AI tools on company devices after employees reportedly uploaded internal source code to ChatGPT.
8. Your exact location and routines
Home address, your children's school, and when you're away from home. Small details add up.
9. Photos and scans of documents
A photo of an ID, a letter or a receipt sends everything visible in it, including details you didn't notice.
What's usually fine
- General questions and explanations
- Your own writing, with names and specifics removed
- Public information
- Work tasks where the real details are replaced with placeholders
How to still get help: use placeholders
The AI needs the structure of your request, not the real values. Compare:
Instead of: Write a reminder to David Levi (david@company.com) that his card ending 4821 will be charged on March 3.
Write: Write a reminder to [NAME] ([EMAIL]) that his card ending [CARD] will be charged on [DATE].
You get exactly the same message, and you fill the details back in yourself. Here's the full method.
A safety net for busy days
Checklists work when you remember them. SafePrompt is a free Chrome extension that checks for you when you press Send on ChatGPT, Claude, Gemini, Grok, Copilot or DeepSeek. If it spots names, emails, phone numbers, IDs, card numbers or bank details, it warns you and offers to swap them for placeholders in one click. When the answer comes back, it puts the real values back on your screen. It runs entirely in your browser, so your prompts never go to SafePrompt.
The 10-second check before you press Send
- Does this contain an ID, card, account or password?
- Does it contain someone else's contact details?
- Is it client, patient or employee data?
- Would I be OK with a stranger reading this?
If any answer is wrong, replace the details first.