BeginnerLesson 8 of 305 min read

Privacy and safety: what to share and what not to

Learn where your chats go, what never to paste, how to use data controls, temporary chats and memory, and how to stay safe with AI.

In this lesson you will

  • Explain where your chat data goes
  • Sort information into green, amber and red
  • Use data controls, memory and temporary chat
  • Follow a safe-use checklist

About 15 minutes

A chat is not a diary, and not a vault

Imran, in a hurry, pastes an entire customer spreadsheet into ChatGPT, names, phone numbers and addresses included, to ask for a summary. It works beautifully. It also means he has just sent other people’s personal data to an outside company, possibly against his employer’s rules. Nobody was hurt, but a good habit was missed. Privacy with AI tools comes down to a few clear ideas: know where your words go, decide what you will and will not type, and use the settings that give you control. This lesson gives you a simple, calm approach, with no need for fear.

Good to know

Privacy policies and settings differ by plan and change over time. This lesson explains the principles. Always read OpenAI’s current privacy information and the options in your own settings, and follow your school or employer’s rules.

Where does your chat go?

When you send a message, it travels to OpenAI’s servers to be processed and the reply comes back. Your chat is usually stored in your account so you can see it in your history. Depending on your plan and settings, conversations may also be used to help improve the models, and a small number of conversations can be reviewed by people for safety or quality purposes. Business and enterprise offerings usually have stricter default privacy terms than personal accounts. The point: treat a chat as something you are sending to a company, not as a private notebook.

The traffic-light rule

Colour Examples Guidance
Green: fine to type General questions, public information, your own made-up examples, ideas, recipes, drafts with no personal details Go ahead
Amber: think first Your own name and job, a work document, a school essay, a photo of your home, your city Remove names or details you do not need, or check the rules
Red: do not type Passwords, bank or card numbers, ID numbers, secret keys, private medical records, other people’s personal details, confidential business or client data Never paste into a chat

Replace details before you paste

You can often get the same help without revealing anything. Swap real names and numbers for placeholders before you paste.

Original: "Dear Mr. Ahmed Raza, your account number 1234-5678 is overdue by 30 days..."
Rewritten: "Dear [Customer], your account [number] is overdue by 30 days..."

Settings that give you control

  • Data controls. Look in settings for an option about whether your chats may be used to improve the models. Many people switch this off for peace of mind. Read what it says on your plan.
  • Temporary chat. Many versions have a temporary or private chat mode. Chats there are not added to your history and are not used to train the models, though they may be kept for a short period for safety reasons, according to OpenAI. Use it for one-off sensitive-ish questions.
  • Memory. If the memory feature is on, ChatGPT can save facts about you across chats. You can view, delete individual memories, clear them all or turn the feature off.
  • Delete chats. You can delete individual chats or clear your history from settings.
  • Export your data. There is usually an option to download everything stored about you.
Do a ten-minute privacy check-up

Open settings once. Read the data controls, decide about training and memory, switch on two-factor authentication, and look at what memory has saved. Then you can use the tool relaxed.

Other people’s information

It is not only your own privacy. Photos of friends, messages from a group chat, a student’s name and grades, a patient’s details: other people trusted you with that information, and they did not agree to it going to an AI company. As a rule, remove or change identifying details, and if in doubt, ask permission or leave it out.

Work and school rules

Many companies and schools set rules about AI tools. Some ban pasting confidential material; some provide an approved business version with stronger protections. Using a personal account for sensitive work data can break your contract. If you are not sure, ask your manager or teacher before you paste anything.

Safety beyond privacy

  • Scams. AI can write very convincing fake emails, voices and messages. Be wary of urgent requests for money or codes, even if they sound perfect. Check using another route, such as calling a known number.
  • Emotional reliance. It can feel like a friend. It can be a comfort in small ways, but it is not a person, a therapist or a doctor. For serious worries, talk to someone you trust or a qualified professional, or contact local emergency or crisis services if you are in danger.
  • Children. Younger users should use it with a parent or teacher, and age rules apply.
  • Harmful or illegal requests. The assistant is designed to refuse them. Do not rely on it for dangerous instructions, and do not use it to harm others.
  • Copyright and honesty. Be careful about presenting AI output as human work where honesty matters, and about copying other people’s protected work into your own products. Check the usage terms for commercial use.
Common mistake

Pasting a document first and wondering about the consequences afterwards. Check the colour of the information before you paste.

Common mistake

Believing a chat is automatically secret because it is on your screen. Anyone who uses your logged-in device or account can read your history. Lock your phone and log out of shared computers.

A safe-use checklist

  1. Use the official app or site, with a strong password and two-factor authentication.
  2. Decide your data controls and memory settings.
  3. Do not paste passwords, financial details, IDs or other people’s private data.
  4. Use placeholders for names and numbers.
  5. Use temporary chat for sensitive-ish questions.
  6. Check your employer or school rules.
  7. Verify important answers before acting on them.

Check what you learned

Five quick questions. Pick an answer and you will see straight away whether it is right and why.

Which of these should you never paste?

  1. A recipe request
  2. A poem idea
  3. A holiday question
  4. Passwords and card numbers

Secrets and financial details belong in the red zone.

What does a temporary chat do?

  1. It makes the reply longer
  2. It is not saved in your history
  3. It costs money
  4. It deletes your account

It gives you a more private way to ask one-off questions.

How can you ask about a customer letter safely?

  1. Paste everything
  2. Skip the question
  3. Send it by text
  4. Replace names and numbers with placeholders

Placeholders give the same help without private data.

What does the memory feature do?

  1. Saves facts about you across chats
  2. Boosts the battery
  3. Fixes spelling
  4. Blocks adverts

You can view and delete what it saves.

Who should you ask if you are unsure whether you can use AI for a work document?

  1. A stranger online
  2. The chatbot itself
  3. Your manager or the company policy
  4. Nobody

Workplace rules can be strict about confidential data.

Summary

  • Treat a chat as something sent to a company.
  • Never paste passwords, financial or other people’s private details.
  • Use placeholders and temporary chat.
  • Check your settings and your school or workplace rules.

Next lesson

Next: working with files, images and voice, plus tools that go beyond plain text.

Questions

Can OpenAI staff read my chats?

OpenAI says limited review can happen for safety and quality, and its policies describe the details. Read the current privacy policy for your plan.

Is the paid version more private?

Not automatically. Personal accounts, including paid ones, have their own data settings. Business and enterprise plans often have stricter defaults.

Should I switch off chat history?

That is your choice. Turning off training use, or using temporary chat for sensitive-ish questions, is a sensible middle path.

Progress is saved in this browser.