CyberSandberg

Cyber AI

What to tell a chatbot, and what to keep to yourself

Checked

A chat box can feel like a private notebook. You type a question about a rash, a debt or a hard week, and an answer appears. But what you type goes to the company that runs the chatbot. Depending on the service and your settings, it can be saved to your account, read by trained reviewers and used to improve the company's AI. That's no reason to stop using these tools. It is a reason to decide what you share, and to spend five minutes on your settings.

Talk about it at home: jump to the Kitchen Table Talk

We use four widely used chatbots as examples: ChatGPT (from OpenAI), Gemini (Google), Claude (Anthropic) and Copilot (Microsoft). They are examples, not a ranking or an endorsement. The same questions apply to any chatbot you use.

What the companies say happens to your chats

Date checked: Oct 22, 2026. Setting names and how long chats are kept change often. Confirm on the company's own help page before you rely on a detail.

Gemini. Google says human reviewers, including trained reviewers from its service providers, read some Gemini chats to help improve its services. Its privacy page asks: "Please don't enter confidential information that you wouldn't want a reviewer to see or Google to use to improve our services." Google says your Gemini activity is deleted automatically after 18 months by default, and you can change that to 3 months or 36 months, or choose not to auto-delete it. Chats that reviewers have looked at are kept for up to three years, even if you delete your activity.

ChatGPT. OpenAI's setting is called "Improve the model for everyone." OpenAI says that when it's off, your new conversations won't be used to train its models, but they can still appear in your chat history.

Claude. Anthropic says it uses chats from its consumer plans to improve its models if you choose to allow that, or if a conversation is flagged for a safety review. It says Incognito chats are not used to improve Claude.

Copilot. Microsoft says that in the updated Copilot app, available since August 18, 2026, your prompts, responses and file contents aren't used to train foundation models. If you allow personalized ads, Microsoft says those ads might use your chat history and saved memories. If you haven't updated the app, Microsoft's privacy FAQ for the older version says your chats can be used to train its AI models unless you opt out.

The details differ, but the pattern is the same: each company keeps at least some of what you type, for at least some time.

What to keep out of any chat

Some things don't belong in a chat box, whatever your settings say:

You can still get help. Describe the situation with made-up names and round numbers. "A relative owes about $5,000 on a credit card" gets you the same explanation as the real statement would.

Our rule of thumb: if you wouldn't write it on a postcard to a stranger, don't type it into a chatbot.

Do this now: Think of the last thing you asked a chatbot. Would it pass the postcard test?

A five-minute settings check

Open the chatbot you use most and look for four things.

1. Training. In ChatGPT, open Settings, then Data controls, and find "Improve the model for everyone." In Gemini, the setting is called Keep Activity. Google says that if it's off and you don't send feedback, your future chats aren't used to improve its AI models. In Claude, Anthropic calls the choice Model Improvement, and it's in your Privacy Settings.

2. Memory. Some chatbots remember details about you from one chat to the next. In ChatGPT, go to Settings, then Personalization, then Memory to review or delete what it remembers. In Copilot, Microsoft says to open Settings, select Personalization, then select Manage next to Saved memories. It also says turning memory off doesn't delete memories already saved.

3. Temporary chats. OpenAI says ChatGPT's temporary chats don't appear in your history, don't create memories and aren't used to train its models, but may be kept for up to 30 days for safety. Google says Gemini's temporary chats aren't used to train its AI models, and it stores them with your account for 72 hours. Claude's version is called Incognito.

4. Feedback buttons. The thumbs up and thumbs down can share more than you'd expect. OpenAI says that if you give feedback, the entire conversation may be used to train its models, even if you've opted out of training. Anthropic says feedback stores the whole related conversation for up to five years and may be used to train its AI models.

Do this now: Change one setting tonight, and decide whether you want to keep using the feedback buttons.

Deleting isn't always erasing

Turning off training is not the same as deleting. OpenAI says the training setting "does not delete or hide saved chats." It also says deleting a chat doesn't necessarily delete a separate memory ChatGPT saved from that chat. And as noted above, Google keeps chats that reviewers have looked at for up to three years.

If you want something gone, delete the chat and any memory made from it, then check the company's help page for anything else it keeps.

Work and school accounts follow different rules

OpenAI says that by default it doesn't use content from ChatGPT Business, Enterprise or Edu workspaces to train its models, and that your organization's own privacy, retention and access policies still apply. Microsoft says its pages on consumer Copilot privacy don't apply when you sign in with a work or school account.

Before pasting anything from work or school into a chatbot, follow your employer's or school's AI policy. If there isn't one, ask before you paste.

Do this tonight

Open the chatbot you use most. Check the training and memory settings, then delete one old chat you'd rather not keep. Next, learn how to check the answers a chatbot gives you.

Teach it this week

Cyber AI

Would you tell a stranger? Deciding what's safe to type into a chatbot

Big idea: A chatbot isn't a diary. What you type goes to the company, can be kept for months or longer, and may be read by reviewers. Share like you're writing a postcard.

Background for the teacher or parent: Google asks Gemini users not to enter confidential information they wouldn't want a reviewer to see, and OpenAI explains ChatGPT's training and temporary chat settings. Read both: Google: Gemini Apps Privacy Hub and OpenAI: data controls in ChatGPT.

Classroom version (10 minutes)

  • Objective: Sort information by how safe it is to share with a chatbot, and name one privacy setting.
  • Hook (2 min): Hold up a postcard. Who might read it before it arrives? Who might read a chat?
  • Explain (3 min): Chats can be saved, and some are reviewed by people. Temporary chats and training settings help, but don't make a chat private. Share the postcard rule.
  • Activity (4 min): Pairs sort made-up cards into three piles: fine, change the details first, never. Examples: a math question, a friend's phone number, a password, a parent's bank account number.
  • Exit ticket (1 min): Rewrite one "change the details" card so it's safe to ask.

Kitchen Table Talk (10 minutes)

  1. Ask who uses a chatbot, and for what.
  2. Read the rule aloud. Our rule of thumb: if you wouldn't write it on a postcard to a stranger, don't type it into a chatbot.
  3. Together, find the training and memory settings in the chatbot you use most.
  4. Tonight's action: each person deletes one old chat they'd rather not keep.

Use made-up examples only; never real passwords, account numbers or personal details. Free to use; please keep the source credit.

This lesson also has its own page: Would you tell a stranger? Deciding what's safe to type into a chatbot.

Education only, not financial, legal or tax advice. Not a registered investment adviser.

Sources

Primary sources

  1. OpenAI: data controls in ChatGPT, help article, date checked: Oct 22, 2026
  2. Google: Gemini Apps Privacy Hub, help article, date checked: Oct 22, 2026
  3. Anthropic: is my data used for model training?, privacy center article dated March 16, 2026, date checked: Oct 22, 2026

Supporting links

  1. Microsoft: Copilot for individuals, your privacy controls and choices, date checked: Oct 22, 2026
  2. OpenAI: memory in ChatGPT, date checked: Oct 22, 2026
  3. Microsoft: Copilot for individuals, your activity history, date checked: Oct 22, 2026
  4. Microsoft: privacy FAQ for Microsoft Copilot (older app), date checked: Oct 22, 2026