Knowledge · Ira Berg, 12 November 2026
What you should never tell an AI
In short: Anything you would not write on a postcard does not belong in an AI chat either. So no passwords, no ID photos, no addresses, no secrets and nothing about other people without their permission. The good news: the AI almost always helps you just as well if you simply leave such details out.
Why this matters
A conversation with ChatGPT, Gemini or Claude feels private. It is one window, just you and the AI. In fact, everything you type ends up on a company's servers. Depending on the provider and your settings, it is stored, used to improve the AI and in some cases read by people who check quality. Google even says this explicitly for Gemini.
That does not mean someone is deliberately searching your chats. But it does mean: once something is typed in, it is no longer fully in your hands. A well known case from 2023 shows how quickly this happens. Samsung employees had copied secret program code into ChatGPT to find bugs. Afterwards the company temporarily banned its use completely.
The 7 rules
1. No passwords, PINs or login details
Not even “just quickly to check if the password is secure”. A password that has been typed in somewhere is no longer a secret. If you want to know what a secure password looks like, ask for the rules, not about your password.
2. No details that make you findable
Full name together with address, phone number, school and class gives a pretty exact picture of you. On their own these are often harmless, together they are not. For most questions “I am 15 and in 9th grade” is enough.
3. No photos of ID cards, school reports or bank cards
Such documents contain a lot of data at once, often including numbers that can be used for fraud. If you need help understanding a form, type out the question and leave out the personal fields.
4. Nothing about others without their permission
Screenshots from the class chat, photos of friends, messages from your teacher: all of this contains data from people who were not asked. Everyone has a right to their own image and their own data. It gets especially sensitive when photos of others are to be changed with AI. That can quickly become hurtful and in the worst case is a criminal offence.
5. Health and very personal things only without names
Many teenagers ask an AI about things they do not talk about with anyone else. That is understandable, and an AI can give good first information. But leave out names and identifying details. And importantly: an AI is not a doctor and not a therapist. If you are really not okay, talk to a person you trust: parents, friends, a teacher or a counselling service.
6. No secrets that are not yours
Family matters, things someone told you in confidence, documents from your parents' jobs: if it is not your secret, you may not pass it on. Not even to a machine.
7. Nothing you might regret later
The rule of thumb for everything else is the postcard test. Imagine your input is written on a postcard that passes through many hands. Does that feel uncomfortable? Then phrase the question more generally.
How to still use AI well and safely
- Placeholders instead of names: write “Person A said in the group …” instead of the real name. The answer will be just as helpful.
- Turn off training: in the settings of ChatGPT, Gemini and Claude you can decide whether your conversations may be used to improve the AI. Have a look and decide consciously.
- Use temporary chats: the big providers have a mode in which conversations do not end up in your history. Good for questions that feel a bit uncomfortable.
- Think first, then type: most questions can be asked in a way that nobody can tell who it is about. This short pause is also good practice for thinking along.
Families can include these points in their AI rules for the family.
And Monkey Score?
Monkey Score notices how you work with the AI, but not what you write. Only numbers are stored, for example how long an answer was visible or whether you asked a follow up question. Texts are never stored and nothing leaves your device. More about privacy
FAQ
Does anyone read my chats with ChatGPT, Gemini or Claude?
That depends on the provider and your settings. Conversations can be stored, used to improve the AI and in some cases checked by people. Google points this out explicitly for Gemini. So always assume someone could read along.
Is a chat gone when I delete it?
Not necessarily right away. Providers sometimes keep data for a while, for example for security reasons or legal obligations. Once something is typed in, it is no longer fully under your control.
Can I upload photos of friends to an AI?
Only with their permission. Photos show faces and often places too. Everyone has a right to their own image. So ask first, especially if the picture is to be edited or changed.
How can I still use AI for personal questions?
Leave out names, addresses and other identifying details and replace them with placeholders like “Person A” or “my school”. The AI can help you just as well without knowing who you are.
Prompt to try
Copy this straight into ChatGPT, Gemini or Claude:
“I want to ask you something personal without revealing who I am. Help me rephrase my question so that it contains no names, places or other details that could identify me: [your question]”