Straight answer
Is ChatGPT private?
Short answer: your chats are not private in the way a notebook is. They leave your device exactly as you typed them, arrive with your account and your internet connection attached, are stored, can be read by people, and by default can be used to improve the models. Here is what that means in practice — and how to get the same quality of answer without it.
What actually happens
Four things most people don't realise.
What arrives
The full text of your message
Everything you typed or pasted leaves your device exactly as written — the client's name, the real numbers, the private situation you were asking about.
What is attached
Your account and your connection
The message does not arrive anonymously. It carries the account you signed in with and the number your internet connection travels under, which barely changes from hour to hour.
What is kept
A history you can't fully see
Conversations are stored, kept for a period even after deletion, and can be read by people for safety and quality reviews.
What it can become
Training material, unless you opt out
By default, chats can help improve the models. Turning that off is a setting most people never find.
None of this means ChatGPT is careless. It means the whole design assumes you are willing to hand over the message first and trust what happens next. That is a reasonable trade for a recipe. It is a poor one for a client file.
Reduce the exposure
Three things worth doing right now.
Turn off model improvement. In the app's data controls, stop your chats being used for training. It is off-by-default for nobody, so you have to go and do it.
Stop pasting identifiers. Take out names, addresses, account numbers and dates of birth before you ask. The answer is usually just as good without them.
Never paste secrets. Passwords, keys and access codes should never go into a chat window, in any product.
The problem is that all three depend on you remembering, every single time, while you are busy. That is the part Kelven takes over.
The Kelven approach
Keep the answer. Drop the exposure.
Kelven sits between you and the AI companies. On your own device, it replaces the personal parts of your question with stand-ins, removes passwords and codes outright, then asks on your behalf — so the AI company sees Kelven's connection, never yours.
The answer comes back and your real names and figures are put back in on your screen. You read the complete version. Nobody else does.
And because Kelven isn't tied to one company, the same question can go to whichever model is genuinely best at it.
You type
“Draft a reply to Sarah Whitfield at her private home address about the $240,000 shortfall on her build.”
The AI company reads
“Draft a reply to PERSON_1 at PROPERTY_1 about the AMOUNT_1 shortfall on their build.”
You read
The finished reply, with Sarah's name, her address and the real figure back in place.
More answers
Common questions.
Is ChatGPT private?
Not in the way most people assume. Your conversations are sent to the company's servers as you typed them, stored against your account, and can be reviewed by people for safety and quality. Unless you turn it off, conversations can also be used to improve their models. Your internet connection arrives with every message too.
Does ChatGPT share or sell my data?
OpenAI says it does not sell personal information, but that is a different promise from not receiving it. Your messages still arrive in full and are kept for a period, and business partners and reviewers can be involved. The safest assumption is that anything you type is readable by someone other than you.
How do I stop ChatGPT using my conversations?
In the app's data controls you can switch off improving the model for everyone, and you can delete conversations. That reduces training use, but it does not change the fact that the full text and your connection still reach their servers first.
Is it safe to paste confidential work into ChatGPT?
Client contracts, patient details, salaries and unreleased plans are usually covered by a duty of confidentiality, and pasting them into a general AI chat sends them outside your control. Many employers and professional bodies advise against it.
How is Kelven different?
Kelven works on your device before anything is sent. Names, addresses, figures and other personal details are replaced with stand-ins, passwords and codes are removed, and Kelven asks on your behalf so the AI company never sees your connection. Your real details are put back into the answer on your screen.
Is anything guaranteed?
No. No automatic check catches everything, and the outside AI company still reads the stood-in version of your request in order to answer it. Kelven shows you exactly what was sent so you can judge it yourself.