9 Things You Should Never Type Into ChatGPT (Or Any AI)

Most people treat ChatGPT like a private diary that happens to talk back. You paste in whatever you're working on, ask your question, and move on — without ever wondering where that text goes.

Here's the uncomfortable truth: on most personal plans, it doesn't just disappear. By default, your conversations on ChatGPT's Free, Plus, and Pro tiers are used to help train future models. Real humans can review them. And — unlike a conversation with your doctor, lawyer, or therapist — AI chats carry no legal confidentiality. If it ends up in a data breach or a legal request, "but I told it in private" means nothing.

This isn't a reason to stop using AI. These tools are genuinely useful and most of what you type into them is harmless. But there's a short list of things that should never go into a chat box, and a few simple habits that let you keep using AI freely without exposing yourself. Let's go through both.

Before you type something sensitive, ask: if this exact text showed up in a leak, a court filing, or in front of a stranger reviewing chats — would I be okay?

First, where does your text actually go?

Three things can happen to anything you type into a consumer AI tool:

It can become training data. Many tools, by default, feed your conversations back into the system to improve future versions. In theory, fragments of what you typed could influence how the model responds to someone else down the line.

A human can read it. All the major companies use human reviewers to check conversations for quality and safety. The "private" chat you had at 2am may pass across a stranger's screen.

It can be exposed later. Data breaches happen. Legal requests happen. And courts have generally found that AI conversations don't get the confidentiality protection that privileged conversations do. Once it's typed, you've lost control of it.

With that in mind, here's the list.

1. Passwords, PINs, and login credentials

This sounds obvious, but people do it constantly — "here's my password, help me think of a stronger version," or pasting a whole login to troubleshoot something. Don't. A chat box is not a password manager. Anything you type could be stored, reviewed, or exposed, and credentials are the single most damaging thing to leak.

2. Financial account details

Credit card numbers, bank account and routing numbers, your full account balances tied to your name. People paste bank statements in to "analyze my spending" or card details to "check if this charge looks right." If you want AI to analyze your finances, strip the identifying numbers first — the AI can find spending patterns perfectly well from amounts and categories alone.

3. Government ID numbers

Social Security numbers, passport numbers, driver's license numbers, national ID numbers. These are the master keys to identity theft. There is almost no legitimate reason to type one into a chat box, and a lot of downside if it surfaces later.

4. Other people's personal information

This is the one professionals get wrong most often, and it's the most legally dangerous. Pasting in a client's details, a patient's records, an employee's review, or a customer list isn't just risking your privacy — you're handing over data that isn't yours to share. If you work in healthcare, law, finance, HR, or education, this can violate confidentiality rules you're legally bound by (HIPAA, attorney-client privilege, and so on). Anonymize first, every time.

5. Confidential or proprietary work material

Unreleased products, internal strategy documents, source code under NDA, trade secrets, anything marked confidential. There's a well-known case of employees at a major company pasting proprietary code into ChatGPT to debug it — and effectively leaking it. If your employment contract or an NDA covers it, assume the chat box is the same as posting it publicly.

6. Identifiable medical information

Your specific health conditions, diagnoses, test results, or medications tied to your name. It's completely fine to ask AI general health questions ("what are common causes of lower back pain?"). It's a different thing to build a detailed, identifiable medical profile of yourself in a system with no medical-privacy protection. Keep the question general; keep your identity out of it.

7. Private details about your kids

Children's full names, schools, schedules, photos, or anything that could identify or locate them. This deserves its own line because the instinct to ask AI for parenting help is strong and the information often slips in without thinking. Strip identifying details — the advice doesn't get worse, and your family's safety stays intact.

8. Anything you've promised to keep confidential

Information covered by a contract, a confidentiality agreement, a settlement, or a professional duty. Even if it's not technically "personal data," typing it into a tool with no confidentiality protection can put you in breach of an agreement you signed. When in doubt, keep it out.

9. Anything you'd be harmed by if it became public

This is the catch-all test, and it's the most useful one to internalize. Before you type something sensitive, ask: If this exact text showed up in a leak, a court filing, or in front of a stranger reviewing chats — would I be okay? If the answer is no, don't type it. That single question covers the eight specific cases above and everything else this list doesn't name.

How to keep using AI freely (and safely)

The point isn't fear. It's a few small habits that let you use these tools for almost everything while keeping the genuinely sensitive stuff out:

Turn off training. On ChatGPT, go to Settings → Data Controls and switch off "Improve the model for everyone." This stops your future chats from being used to train models. Claude and Gemini have equivalent toggles in their settings. It takes thirty seconds and protects every conversation going forward.

Use Temporary Chat for sensitive sessions. ChatGPT's Temporary Chat (the option at the top of a new conversation) keeps the chat out of your history and out of training. It's the right mode for anything you'd rather not have stored.

Anonymize before you paste. This is the real unlock. Replace names with "the client," account numbers with "XXXX," your company with "a mid-size firm." You get the same quality of help on a contract, a financial analysis, or a tricky email — the AI doesn't need the real names to be useful. (If you're not sure how to ask in a way that still gets a great answer, see Why ChatGPT Gives You Generic Answers — the same specificity that makes prompts good lets you describe a situation fully without naming names.)

Use a business or enterprise plan for sensitive work. If your job regularly involves confidential material, the team and enterprise tiers are built differently — your data is typically excluded from training by default and covered by stronger agreements. We broke down whether the paid tiers are worth it in Is ChatGPT Plus, Claude Pro, or Gemini Advanced Worth It?.

The takeaway

Three things to remember:

One. Your AI chats are far less private than they feel. On most personal plans they're used for training by default, humans can review them, and they have no legal confidentiality.

Two. Keep a short list out of the chat box entirely: passwords, financial details, ID numbers, other people's data, confidential work material, identifiable medical info, your kids' details, anything under an agreement — and anything you'd be harmed by if it leaked.

Three. You don't have to stop using AI. Turn off training, use temporary chats for sensitive sessions, and anonymize before you paste. Those three habits cover almost everything.

AI is one of the most useful tools you'll ever have. Use it for nearly everything — just not as a vault for the things that would actually hurt you to lose.