What Never to Tell a Chatbot (and When Not to Trust It)

Rosa just got a letter from her health insurance company. It is full of codes and terms she does not understand. She has heard AI can explain things, so she opens a chatbot and starts typing everything: her full name, her date of birth, her member ID, her Social Security number, and the whole letter. Halfway through, she stops. "Wait," she thinks. "Where does all of this go?"

Good instinct. AI can be a big help with letters like that, and you will learn how in the next lesson. But first, two important skills: keeping your private information private, and knowing when to double-check what the AI tells you.

Where your words go

When you type into a chatbot, your words travel over the internet to the company that runs it. Depending on the company and your settings, what you type may be saved, and it may be read by people who work there or used to improve the program.

You do not need to be afraid of this. You just need to treat a chatbot like a helpful stranger. You would happily ask a stranger to explain a confusing word. You would not hand them your bank card.

What never to type into a chatbot

Keep these out of any chatbot, including the boxes in this course:

• Social Security number
• Bank, credit card, or account numbers
• Passwords and PINs, including the codes a bank texts you
• Your full medical records, or detailed health history tied to your name
• Your full name together with your address and date of birth. Together, these are what someone needs to pretend to be you.
• Other people's private information, like a family member's medical details or account numbers

A good rule: if you would not write it on a postcard, do not type it into a chatbot.

How to get help without sharing private details

The good news is that the AI almost never needs your private details to help you. It needs to understand the situation, not who you are.

The trick is to blank things out. Before you paste in a letter or describe a problem, replace private details with a short label in square brackets. For example:

• Instead of your account number, write [my account number]
• Instead of your name, write [my name]
• Instead of your address, write [my address]
• Instead of your doctor's name, write [my doctor]

So instead of this:

"My name is Rosa Delgado, account 4417-2290-1185, and Lakeview Health says I owe $340."

Write this:

"My name is [my name], account [my account number], and my health insurance company says I owe $340."

The AI can explain the situation just as well. It does not need the real numbers. Amounts and dates are usually fine to keep, because they help the AI understand what is going on.

When not to trust the AI

Now the second skill. In Lesson 1 you learned that a chatbot writes by choosing likely words. That is why it can be confidently wrong. It can give you a phone number that does not exist, a deadline that is off by a week, or a rule that was true years ago but has changed.

It will not warn you. It sounds just as sure when it is wrong as when it is right.

So here is a simple way to decide how much to trust an answer.

Low stakes: go ahead

Dinner ideas, a birthday card message, explaining a word, a first draft of a friendly note. If it is a little off, no harm done. You can fix it yourself.

Higher stakes: check with the real source

Anything involving money, health, the law, deadlines, or your benefits. Use the AI to understand and prepare. Then confirm the important facts with the real source:

• Phone numbers and websites: get them from your card, your bill, or the official website. Never from the AI.
• Deadlines and amounts: check the original letter or call the sender.
• Medicine and health: ask your doctor or pharmacist.
• Legal questions: talk to a lawyer or a free legal aid office.
• Money decisions: talk to your bank or a trusted advisor.

The AI is not a doctor, lawyer, or financial advisor. And in an emergency, call 911.

A helpful habit: ask it how sure it is

You can ask the AI, "What parts of your answer should I double-check, and with whom?" It will usually point out the parts most likely to be wrong. That is a great starting point for your checking.

Practice blanking things out

Press Run to have the AI help you blank out a message before you send it anywhere. Every name and number in this example is made up. You can change the words first.

Practice checking an answer

Here is a second practice. Press Run to ask the AI to point out which parts of an answer need checking. You can change the question to one you care about.

What you just learned

• Your words go to the company that runs the chatbot, and they may be saved.
• Never type your Social Security number, account numbers, passwords, full medical records, or full identity details.
• Blank out private details with labels like [my account number]. The AI can still help.
• The AI can be confidently wrong. For money, health, law, and deadlines, use it to understand, then confirm with the real source.
• Phone numbers and websites should come from your card, bill, or the official site, never from the AI.

Try it yourself

Think of a real situation you might want help with, like a bill, a form, or a message. Do not paste it here. Instead, use the box below to practice describing it the safe way. Press Run, and change the words to fit your situation, keeping private details blanked out.

In the next lesson, you will put both skills together to understand a confusing letter or bill, step by step.