More than half of New Zealanders now use AI at least weekly, according to InternetNZ’s 2026 research. It’s useful for drafting a tricky email, making sense of a letter, or planning a week of meals.
The problem is that a chatbot feels like a private conversation, and it isn’t one. Whatever you paste leaves your device and ends up on the company’s servers, usually overseas.
If you wouldn’t tell a crowd of strangers, don’t tell a chatbot.
What happens once you hit send
01 · Stored
Kept on the company’s servers, often overseas and outside NZ privacy law.
02 · Reviewed
Some conversations are read by staff or contractors to check quality and safety.
03 · Trained on
Most free accounts use your chats to train future models unless you opt out.
04 · Leaked
Databases get exposed. One AI app left 300 million private chats open online in January 2026.
5 things you shouldn’t tell an AI chatbot
01 / 05
Personal ID numbers
Your IRD number, NHI number, bank account, passport, driver licence, RealMe login or KiwiSaver details.
These are all a scammer needs to pretend to be you, open accounts in your name, or get into the ones you already have. A chatbot doesn’t need the real number to help you fill in a form.
Do this instead
Swap the real number for a placeholder like [IRD NUMBER] before you paste.
02 / 05
Passwords and codes
Passwords, PINs, two-factor codes and recovery phrases.
It can be tempting to hand them over when you’re stuck. No AI tool needs your password, and once it’s stored, you can’t control whether it ends up in front of someone else, or whether an AI tool uses it to log in without your permission.
Do this instead
Describe the problem (“I’m locked out of my bank app”), not the login.
03 / 05
Work and customer info
Customer lists, client files, internal documents and anything your employer would call confidential.
In 2025 a contractor in New South Wales uploaded a spreadsheet of flood victims’ details to ChatGPT, exposing up to 3,000 people. Under New Zealand’s Privacy Act, it’s the organisation that’s accountable, not the AI company.
Do this instead
Check your workplace AI policy and use the approved tool, not your personal free account.
04 / 05
Other people’s messages
Texts, emails and group chats you’ve received from someone else.
You might be comfortable with your own words being stored. The person on the other end didn’t get a say. When the Washington Post looked at 47,000 shared ChatGPT chats, people had pasted in more than 550 email addresses and 76 phone numbers.
Do this instead
Describe the situation and what you want to say back. Leave their words out.
05 / 05
Health and medical details
Test results, diagnoses, prescriptions and letters from your doctor.
There’s no doctor–patient confidentiality with a chatbot. This is some of the most sensitive information you have, and once it’s out you can’t take it back.
Do this instead
Ask general questions without your details, and take the specifics to your GP or Healthline on 0800 611 116.
Turn off training
Most free AI tools train on your chats unless you disable the setting. Switching it off only takes about a minute per app.
These menus change often. If yours looks different, use the official instructions link under each one.
OpenAI
ChatGPT
- Open your profile menu and go to Settings.
- Choose Data controls.
- Turn off Improve the model for everyone.
Note: Temporary Chat isn’t used for training, but is kept for up to 30 days.
Gemini
- Go to Settings & help, then Activity.
- Click On at the top.
- Choose Turn off, or Turn off and delete activity.
Note: With activity off, chats are still kept for up to 72 hours to run the service.
Anthropic
Claude
- Click your name and choose Settings.
- Choose Privacy.
- Turn off the model improvement toggle (Help improve our AI models).
Note: Incognito chats aren’t used for training.
Microsoft
Copilot
- Select your profile icon, then your name (on mobile: Account).
- Choose Privacy.
- Turn off Training on conversation activity and Training on voice conversations.
Facebook · Instagram · WhatsApp
Meta AI
- In the Meta AI app, open Settings, then Data controls.
- Turn off Help improve our AI.
Heads up: Meta AI is also built into Facebook, Instagram and WhatsApp, where the controls vary.
What to do instead
If you still want or need to use AI for something sensitive, these habits lower the risk.
- Strip out names and numbers before you paste.
- Check screenshots for open tabs, notifications and names in the background.
- Don’t press “Share” on a chat unless you’d be fine with anyone reading it.
- Use the official app. Third-party “chat with AI” apps often store your chats themselves.
- Be wary of AI browser extensions that ask to read every page you visit.
Already pasted something?
Already shared something sensitive with an AI chatbot? It’s not too late to limit the damage:
- A password: change it now, and anywhere else you’ve used it.
- Bank or card details: ring your bank on the number on the back of your card.
- An ID number: contact the agency that issued it.
- Anything else: delete the chat, and ask the provider to delete your data.
Your rights and where to get help
AI tools are covered by the Privacy Act 2020, the same as any other organisation that holds your information.
- How your data was handled: the Office of the Privacy Commissioner.
- Online harm and scams: Netsafe, 0508 638 723.
- Accounts or details compromised: Own Your Online.
- An AI tool did something it shouldn’t: add it to our AI Incident Tracker.
Before you type
Would you be okay if a stranger read it?
If not, maybe don’t tell a chatbot. Save the one-page checklist, or send this guide to someone who uses AI every day.