What you share with an AI service

What it isCopy Anchor Link

When you use a chat app, your messages go to the provider's servers. What happens next depends on the service, your plan and your settings. Many consumer apps keep your conversations for some time and may use them to improve their models unless you switch that off. Business and developer plans usually have stricter terms, such as not training on your data. Policies change, so check the current settings and privacy page of the service you use. Some apps have a temporary or incognito chat, and most let you delete history.

What to keep outCopy Anchor Link

Passwords and security codes. Government ID, bank or card numbers. Medical records. Private details about other people, such as a friend's problem or a customer's address. Your employer's secrets, unreleased plans, source code or customer data, unless your workplace has approved the tool for it.

A worked example: redactingCopy Anchor Link

Before: "Dear Maria Lopez, your account 4421-8830 is overdue by $240. Please call 555-0142." You want help making it politer. After redacting: "Dear [customer], your account [number] is overdue by [amount]. Please call [phone]." The model improves the wording just as well. You paste the real details back in yourself.

What happens to your messageCopy Anchor Link

Your text travels over the internet to the provider's computers, where the model reads it and writes a reply. The provider may store the conversation, let staff review some of it for safety, or use it to improve future models. Which of these applies depends on the service and plan. That is why the settings page matters more than the logo on the app.

A second example: when redacting is not enoughCopy Anchor Link

Redacting works when the details are separate from the question. Imagine you paste a story about a classmate's family problem with the name removed. People who know your school could still guess who it is from the events. Details such as a rare job, a small town or a date can identify someone without a name. The safer move is to describe the situation in general terms or not share it at all.

Trade-offsCopy Anchor Link

Strict privacy has a cost. The more context you give a model, the better its help, so withholding everything makes answers vague. Pick the level by risk. A recipe or a poem needs no care. Health, money and other people's secrets need a lot. Some tools can run on your own computer, so the text never leaves it. Those models are usually smaller and less capable, which is the trade.

What goes wrongCopy Anchor Link

Data leaks mostly through habit, not hacking. You paste a whole document when two paragraphs would do. You upload a screenshot with notifications visible. You use a work account on a personal device. You will not see any symptom, which is the problem. Prevention is the only fix, because you cannot take text back once it is sent.

Try it yourselfCopy Anchor Link

Open the privacy or data settings of one AI tool you use.

1. Find whether your chats can be used to train models.
2. Find how to delete history or start a temporary chat.
3. Write down what you found in one line.

Expect the answers to be in different places than you guessed. Repeat the check every few months, because policies change.

Common mistakesCopy Anchor Link

Using a personal account for work documents. Assuming deleted chats vanish everywhere at once. Pasting a screenshot that contains tabs, names or messages you did not mean to share. When in doubt, ask yourself whether you would be comfortable if this text appeared in a stranger's inbox.