Mo Explains
I'll be honest up front. I haven't turned off the training setting in my own ChatGPT, and I love memory. It's nice starting a new chat and not having to catch it up on everything again.
But there's a short list of things worth keeping out of the chat box no matter what your settings are, because once you hit send, that text lives on somebody else's computer.
Is it safe to put personal information in ChatGPT?
For everyday stuff, mostly yes, as long as you know where it goes. Your chats get stored on the company's servers and can be used to train future models. And a small slice of them gets read by actual people.
Google says that part out loud on its own help page. The Gemini Apps Privacy Hub reads: "Please don't enter confidential information that you wouldn't want a reviewer to see or Google to use to improve our services, including machine-learning technologies." Chats that a reviewer looks at are kept for up to three years, and deleting your activity doesn't remove them. OpenAI does a version of this too, which I covered in the post on contractors reading ChatGPT chats.
Then there are bugs and mistakes. In March 2023 a bug let some ChatGPT users see other people's chat titles, and OpenAI said it may have exposed payment details for 1.2% of Plus subscribers who were active during a nine-hour window. And in 2025, shared chats people had marked "discoverable" started showing up in Google search until OpenAI pulled the feature.
What not to share with ChatGPT: the five things
These are the ones that cause real trouble if a stranger ever reads them.
- Passwords, PINs and the six-digit codes sites text you, plus recovery codes, security-question answers and crypto wallet phrases. No chatbot needs them for anything.
- Your Social Security number, full card numbers and bank account numbers. Paste a statement if you want help with a budget, but black out the numbers first.
- Confidential work stuff, like client files, contracts, internal reports, code and anything under an NDA. In 2023 Samsung engineers uploaded internal source code to ChatGPT, and the company banned the tools for staff and warned that breaking the rule could get people fired.
- Other people's private details. Your friend's diagnosis or your coworker's address isn't yours to hand to a company.
- Medical records, legal papers, tax forms, pay stubs or a photo of your passport, anything with your full name on it. You can still ask about your own symptoms or your taxes. Just strip out the name, date of birth and ID numbers before you paste a document.
What each AI company says it keeps
All three big chatbots store your chats by default, and each one has a setting that changes how they're used. The details are different enough that it's worth knowing which one you use.
OpenAI's data controls page says new chats aren't used for training once you turn off Improve the model for everyone, and that hitting thumbs up or down can send that conversation for training anyway. Anthropic changed its rules in 2025, and now if you allow training, Claude can keep your chats for up to five years. If you don't, it's 30 days. For Gemini, the setting that matters is Keep Activity.
The settings that shrink the risk
Turning off training and using temporary chats for anything sensitive covers most of it, and none of them take long.
- ChatGPT: go to Settings > Data controls > Improve the model for everyone and turn it off. For a one-off sensitive question, start a Temporary Chat. OpenAI says those don't train its models or create memories, but it may keep a copy for 30 days for safety.
- Gemini: turn off Keep Activity, or use a temporary chat. My guide to turning off Gemini has the exact taps.
- Claude: go to Settings > Privacy > Help improve our AI models and turn it off, or start an incognito chat with the ghost icon.
Think twice before you hit Share on a chat, too. Anyone with the link can read it.
Already shared something you shouldn't have?
Delete the chat, then fix the thing itself, because deleting only stops future use. OpenAI says deleted chats are removed from its systems within 30 days, with some exceptions.
If it was a password or code, change it now. If it was a card or account number, call your bank. If it was work material, tell your IT or security team, because they'd rather hear it from you than find out later. And check what memory saved under Settings > Personalization > Memory, then delete anything from that chat. OpenAI's memory FAQ says deleting a chat doesn't necessarily delete a memory it created.
Honest limitations
None of these settings take back what's already been used. Anthropic says chats already in a finished training run stay in that model, and turning off training today only covers what comes next. ChatGPT's temporary chats still sit on OpenAI's servers for up to 30 days, and a chat flagged for safety can be kept longer.
Connecting your bank or email to a chatbot is a separate set of risks. I covered the bank side in should you connect your bank account to ChatGPT.
Before you paste anything, skim it for those five. If one's in there, black it out first.
Related reading: OpenAI pays contractors to read real ChatGPT chats | What ChatGPT's memory knows about you | AI scams in 2026: voice cloning and romance fraud | New to AI? Start here
