ChatGPT Privacy: 7 Things You Should Never Tell a Chatbot

By Chris Clark, founder of Senior Tech Cafe

ChatGPT privacy comes down to one plain fact: what you type is stored on OpenAI’s servers, can be read by its staff if a conversation gets flagged, and on a personal account can be used to train future versions unless you switch that off. So anything you wouldn’t write on a postcard stays out of the chat.

Remember when privacy meant simply closing your curtains and lowering your voice? Those were simpler times.

Now we’re chatting with AI like it’s our neighbor over the fence. Only this neighbor remembers everything, never blinks, and isn’t really a person.

Yes, ChatGPT is pretty amazing. It’s friendly. It’s helpful. It can tell you how to poach an egg or unpack a sonnet. But this clever little assistant isn’t bound by a pinky swear or doctor-patient confidentiality.

Here’s what actually happens to what you type, the settings that keep more of it to yourself, and seven things you should never tell ChatGPT or any other AI chatbot. Not because it’s evil, but because protecting your privacy is always the smart move.

Is ChatGPT Private?

ChatGPT isn’t private in the way a doctor’s office or a sealed letter is. It’s safe for everyday questions, but personal details you type in are stored, can be seen by OpenAI staff in some cases, and can end up in training data unless you change a setting. Treat it like a helpful stranger, not a confidant.

Here’s what happens to a conversation on a personal account:

• It’s saved in your chat history until you delete it. OpenAI says deleted chats are removed from its systems within 30 days, unless it has to keep them for legal or security reasons.
• It can train future models. On personal accounts this is switched on until you turn it off.
• It can be reviewed by people. OpenAI uses automated tools plus trained staff to check conversations flagged for possible misuse.
• It can be remembered. With memory on, ChatGPT saves details you mention and brings them up in later chats.

You can’t make ChatGPT private, but three settings make it a lot less chatty about you:

1. Turn off training. Open Settings, tap Data controls, and switch off Improve the model for everyone. New chats stay in your history but won’t be used to train models.
2. Use Temporary Chat for anything personal. Start a new chat and turn on Temporary Chat. It stays out of your history, isn’t used for training, doesn’t create memories, and is deleted after 30 days.
3. Check what it remembers. In Settings, find Memory (it sits under Personalization on most versions). You can delete individual memories or turn memory off.

None of this changes the golden rule below. Settings limit what happens after you type something. The safest information is the information you never typed.

What Not to Tell ChatGPT: 7 Things to Keep Out of the Chat

What not to tell ChatGPT boils down to seven kinds of information. For each one, you’ll find why it’s risky, what to do instead, and a short list of questions that are fine to ask.

1. Personally Identifiable Information (PII)

“Hi, My Name Is…Wait, Let’s Not Go There!”

Let’s kick things off with the basics, the kind of info you rattle off without thinking at the doctor’s office or when setting up your cable:

• Full name
• Address
• Phone number
• Email
• Birthdate
• Social Security number
• Passport number
• Driver’s license details

That’s what the pros call personally identifiable information, or PII. It might feel safe to drop it into ChatGPT, especially when you’re sprucing up a résumé or writing a letter. Resist the urge. Big time.

The first reason is simple: if someone ever gets into your ChatGPT account, they can read everything you’ve typed there. And it only takes a handful of those juicy details to swipe your identity and leave you cleaning up a mess that makes a root canal look like a spa day.

The second reason is that once it’s typed, it tends to stick around: in your chat history until you delete it, in ChatGPT’s memory if it decides the detail is worth saving, and possibly in training data if you haven’t switched that off.

Here’s the smarter move: if you’re working on a document with ChatGPT, leave out the real stuff. Use a made-up name like “Gladys McSnazzlepants” or “Hank O’Keyboard.” Just be sure to plug in your real info before sending it off, or you might end up with a job offer addressed to Ms. McSnazzlepants. Which, depending on your mood, might not be the worst thing.

3. Passwords and Login Info

“1234 Is a Bad Password, and ChatGPT Shouldn’t Know It Anyway”

Passwords are already a pain to remember. You know what’s worse? Handing them over to a chatbot, even by accident.

ChatGPT never needs your passwords. Ever. Not for your email, not for your Amazon account, not even for your online mahjong game. The same goes for the six-digit codes that companies text you when you sign in. Anyone asking for one of those, chatbot or person, is a red flag.

Watch out for answers to common security questions too, like your mother’s maiden name, your first pet, or the make and model of your first car. They sound innocent, but scammers use them to reset your accounts and lock you out.

If remembering strong passwords feels like juggling flaming bowling pins, there are better tools for the job. Here’s how to remember your passwords without writing them on the fridge, and how passkeys let you skip passwords on many sites altogether.

🟢 Safe to Ask:

• “What makes a strong password?”
• “Should I use the same password for more than one website?”
• “Are password managers safe?”
• “What’s two-factor authentication?”
• “How do I tell if a login page is fake?”

🔴 Not Safe to Share:

• Email, bank, or account passwords
• Security question answers like your mother’s maiden name or pet’s name
• Account usernames or logins
• “Here’s my password. Does it look strong enough?”
• “The code I just got is 438172. What should I do with it?”

5. Medical Information

“Dr. GPT Is Not a Licensed Physician”

Sure, ChatGPT can explain sciatica like a champ. It might even have you nodding along as it breaks down why your knees now snap, crackle, and pop like a bowl of cereal.

But don’t be fooled. This chatbot is not your primary care physician.

Here’s the real rub: ChatGPT isn’t bound by HIPAA, the law that makes sure your real doctor can’t go blabbing about your blood pressure at a dinner party. When you type in personal health details, the protections you get are whatever OpenAI’s own policies say, not the ones your doctor’s office has to follow.

OpenAI has added a separate Health space inside ChatGPT and says conversations there aren’t used to train its models. That’s an improvement, but it still isn’t covered by HIPAA, and it still isn’t a doctor.

So just like you wouldn’t hand your Social Security number to a stranger in line at Walgreens, think twice before sharing your diagnoses, test results, prescriptions, or health history with an AI, even one that’s really good at explaining gout.

Yes, ChatGPT can offer helpful general info. But it can also be dead wrong. And in the worst cases? Flat-out dangerous.

6. Work Secrets and Business Info

“Don’t Upload Your Boss’s Secret Plan to Rule the Industry”

ChatGPT is great for organizing ideas, fixing awkward sentences, and even making boring meeting notes sound a little less boring. But before you copy and paste your company’s next big idea into the chat, take a beat.

Because unless your job is “accidentally leak sensitive information to the internet,” this is one area where sharing can really backfire.

Just ask Samsung.

In 2023, a few of its engineers pasted confidential source code and internal meeting notes into ChatGPT to save time. Back then, ChatGPT used conversations to train its models by default, and there was no way to pull that data back out. Samsung banned employees from using it on company devices a few weeks later. Oops doesn’t even cover it.

Why Does This Matter?

Because while ChatGPT isn’t a cop, it does have guardrails. It flags sketchy stuff, and flagged chats can be reviewed by actual humans. If they decide a conversation involves a real threat to someone else’s safety, OpenAI says it may refer it to law enforcement.

So if your “joke” crosses the line, you could find yourself locked out, suspended, or permanently banned. Not exactly a tech drama worth streaming.

Easy rule of thumb: if it’s not something you’d blurt out in a crowded elevator, don’t type it into a chatbot. AI doesn’t always pick up on sarcasm, and it definitely doesn’t know when you’re “just kidding.”

So keep it clean, keep it legal, and save the true crime plots for your bookshelf.

If You Wouldn’t Say It to a Stranger on a Bus

Think of ChatGPT as a really helpful stranger: great with words, never interrupts, and somehow knows everything from Shakespeare to scrambled eggs. A solid companion for solving problems, but not the one you should be spilling your life story to.

The golden rule of ChatGPT privacy: treat AI tools as smart assistants, not trusted confidants. Turn off training, use Temporary Chat for anything personal, and keep the seven things above out of the chat box entirely. The same rules go for AI companion apps, which are built to get you talking.

For the rest of the checklist, start with our online safety guide for seniors.

And some things are better left unsaid entirely. Especially the story about how you “liberated” the flamingo from your neighbor’s lawn. Let’s keep that between us.

Questions About ChatGPT Privacy

Is ChatGPT Private?

Not in the way a doctor’s office is. Your chats are stored on OpenAI’s servers, staff can review conversations flagged for safety problems, and on personal accounts your chats can be used to train future models unless you turn that off. Temporary Chat and the Data controls settings cut down what’s kept, but nothing you type is truly private.

Does ChatGPT Remember What I Tell It?

It can. With memory turned on, ChatGPT saves details you mention and can draw on past conversations in new ones. You can see and delete what it has saved, or turn memory off, in Settings. A Temporary Chat doesn’t create memories at all, which makes it the better choice for anything personal.

If I Delete a Chat, Is It Gone?

Mostly. OpenAI says deleted chats are removed from its systems within 30 days, unless it has to keep them for legal or security reasons. Deleting a chat doesn’t erase anything ChatGPT already saved to memory from it, so check your saved memories too. Anything already used for training can’t be pulled back out.

Can I Stop ChatGPT From Training on My Conversations?

Yes. In ChatGPT, open Settings, then Data controls, and turn off Improve the model for everyone. New conversations won’t be used for training after that, though they still appear in your history. Temporary Chats are never used for training. The setting doesn’t reach back to chats you had before you switched it off.