What You Should Never Type Into an AI: Privacy in Practice

You have probably done this before: opened ChatGPT or another AI tool, pasted in a full email, a contract excerpt or a list of customers, and asked for help. It is fast, it is useful and it feels harmless. But have you ever stopped to think about where everything you type actually goes?

The short answer: it does not go nowhere. What you write can be saved in your history, used to improve the model and, in some cases, accessed by other people. The good news is that you can get real value from AI every day without exposing what should stay private. In this article, you will learn what you should never type into an AI, why it is risky and what to do instead.

Where your words actually go

When you send a message to an AI, it is usually stored on the servers of the company that runs the service. Depending on your plan and account settings, those conversations may be used to train future versions of the model, may be reviewed by human quality teams and may sit in your history for a long time.

This does not mean someone is reading your chats right now. It means something more subtle and more important: you lose control over that information. If there is a leak, a shared team account or unauthorized access, what you typed can show up where it never should. And remember that the tool has other limits beyond privacy, as we explain in why AI hallucinates.

Personal data: the most common slip

The most frequent mistake is treating the AI chat like a notepad. In a hurry to get something done, people paste data that directly identifies someone. A few examples that deserve extra care:

• ID documents: national ID numbers, passports, driver's licenses and any number that identifies you or someone else.
• Contact details and addresses: home addresses, personal phone numbers and other people's private emails.
• Financial data: card numbers, bank details, credit limits and statements.
• Passwords and codes: passwords, tokens, API keys and verification codes. Never, under any circumstances.
• Health information: diagnoses, test results and medical reports with a patient's name.

The rule is simple: if the information identifies a real person and you would not want it circulating, it does not belong in an AI chat.

Customer data, company data and confidential documents

At work, the risk changes scale. Pasting a customer spreadsheet, a contract under negotiation or financial results that have not been published yet can violate privacy laws (such as GDPR or Brazil's LGPD), confidentiality agreements and your company's internal policies. Even if nothing leaks, you may be breaking a compliance rule without noticing.

The same goes for proprietary source code, business strategies, supplier data and any document marked internal or confidential. The dangerous part is that almost nobody does this in bad faith: the person just wants a summary of the contract or a quick analysis of the spreadsheet. The problem is that the information left the company's controlled environment and entered an external service, often through a free personal account, with no contract and no guarantees.

What to do instead

The solution is not to abandon AI, it is to use it with judgment. A few habits solve most cases:

• Anonymize before pasting: replace real names with "Customer A", remove ID numbers, emails and phone numbers. The AI does not need to know who the person is in order to help.
• Generalize the context: instead of pasting the whole contract, describe the situation. "I have a service agreement with a termination penalty, how does this kind of clause usually work?" works very well.
• Use fictional data: to test formulas, analyses and automations, create made-up examples with the same structure as your real data.
• Prefer enterprise versions: business plans usually offer data protection agreements and a guarantee that your conversations will not train the model. If your company provides one, use it.
• Follow internal policy: if your company has rules about AI use, they matter more than any shortcut. If it does not, suggest creating them.

Safe to type x better not

To make everyday decisions easier, use this quick reference:

Safe to type	Better not
Conceptual questions and requests for explanations	Passwords, tokens and access keys
Public texts or your own writing, with no third-party data	ID numbers and identification documents
Examples with fictional or anonymized data	Spreadsheets with real customer names and contacts
A general description of a work problem	Full contracts, proposals and confidential documents
Email drafts with no sensitive information	Unpublished company financial data
Generic code and snippets with no trade secrets	Proprietary code and embedded credentials

Checklist: what you should never type into an AI

Before you hit Enter, scan this list. If your message contains any item below, edit it before sending:

• Passwords and keys: for any system, personal or corporate.
• Identification documents: yours or anyone else's.
• Bank and card details: including bills and statements.
• Identifiable health information: reports and test results with a name.
• Customer data: names, contacts, purchase history, contracts.
• Confidential documents: anything your company classifies as internal.
• Information about third parties without consent: if the person did not authorize it, do not paste it.

Conclusion

Using AI safely does not require advanced technical knowledge, it requires habits: anonymize, generalize and think twice before pasting. People who master these practices use the tool with far more freedom, because they know exactly where the line is. If you want to learn how to use AI productively and safely, with method and without surprises, get to know the courses and programs at Data Lover and take the next step with confidence.