Is it safe to tell ChatGPT personal things? What happens to your chats

It’s late. You’ve typed half a message to ChatGPT about something you haven’t told anyone. Then you stop. Who else reads this?

You’re not the only one asking

Most ChatGPT use isn’t about work anymore. More than 70% of messages have nothing to do with work, up from 53% (NBER, September 2025). And much of it is personal:

• About 3 in 10 people who use generative AI type in personal or confidential information, including about their health and money. That’s from a survey of 2,600 people in 12 countries (Cisco, October 2024).
• In the Netherlands, 32% of adults asked an AI chatbot about a health complaint in the past year (Univé, October 2025).
• In the US, 13% of all adults have put their own medical information into an AI tool, like test results or a doctor’s note (KFF, March 2026).

Most people feel uneasy about it at the same time. In the same Dutch survey, 62% worried about what happens to their data. In the US, 77% worried about the privacy of medical information they give AI (KFF). And 71% of Americans think AI will make their personal information less secure (Pew, June 2026).

So people share anyway, with a knot in their stomach. That’s understandable. It helps to know what actually happens.

What happens to a normal chat

The details differ per service. The basics are the same.

• It’s saved under your account. The chat stays in your history until you delete it.
• It can be used for training. ChatGPT can use your chats to train its models until you turn off Improve the model for everyone (OpenAI). In Gemini, the setting that allows this is on by default for adults (Google).
• It can be kept for years. Claude keeps chats for five years if you allow training, and 30 days if you don’t (Anthropic, August 2025).
• A person may read it. Google has human reviewers read some Gemini chats. Those are kept for up to 3 years, even after you delete your activity (Google).
• It isn’t confidential. There’s no doctor–patient secrecy for a chat with AI. OpenAI’s CEO Sam Altman said so himself in July 2025, and called it “very screwed up” (TechCrunch).

What has really gone wrong

None of this is a reason to panic. But these things really happened:

• A court kept deleted chats. From May to late September 2025, a US court made OpenAI keep ChatGPT chats it would normally have deleted, including deleted and temporary chats (Engadget, October 2025). More on that here.
• 20 million chats went to court. In the same copyright case, a judge ordered OpenAI to hand 20 million anonymised chat logs to the lawyers of The New York Times and other news outlets (eWeek, December 2025).
• Shared chats ended up in Google. In July 2025, more than 4,500 shared ChatGPT chats could be found through Google, some with names and CVs in them. OpenAI removed the option that made this possible (Search Engine Journal).
• Grok’s too. In August 2025, a Google search showed more than 370,000 shared Grok chats. People weren’t warned that the link was public (Forbes).

One thing we found no evidence for: that AI companies sell your chats. They use them themselves, to train their models and to check for abuse. That’s reason enough to think before you type.

Asking something personal? Melo, Kamelo’s free private chat, answers on your own computer, so your question isn’t sent anywhere. How it works

Why leaving things out is harder than it sounds

Many people protect themselves by hand. They leave out names, make up a name, or stay vague. That helps a little, but less than you’d think.

In a study with 240 people, participants rewrote their own messages so an AI couldn’t guess something private about them. Only about 28% of the rewrites worked. Rephrasing, the most common trick, worked worst (arXiv, September 2025). The AI still read the clues between the lines.

And a story about your life is rarely only about you. Your mother’s diagnosis. Your partner’s debts. A colleague you clash with. Those people never chose to be in a chat.

What you can do

You don’t have to stop using AI. A few habits go a long way.

• Leave out what identifies people. Names, addresses, dates of birth, patient and account numbers. The question usually works just as well without them. How to do that.
• Turn training off, in every AI you use:
  • ChatGPT: Settings → Data controls → Improve the model for everyone.
  • Gemini: Settings & help → Activity, and turn off Keep Activity. Step by step.
  • Claude: Settings → Privacy → Help improve Claude.
• Use a temporary chat for personal questions. It stays out of your history and isn’t used for training. It does still go to the AI company: ChatGPT may keep it for up to 30 days for safety (OpenAI). What it does and doesn’t do.
• Check what it remembers. Memory can carry a detail from one chat into the next. How to see and delete it.
• Don’t share links to personal chats. Copy the part you need instead.
• Ask on your own computer. For what you really want to keep to yourself, use an AI that runs on your device. Then the question isn’t sent anywhere.

Where Melo fits, and where it doesn’t

Melo, by Kamelo, is free privacy software for your browser. It works in Chrome, Edge and Firefox. When you type something personal on ChatGPT, Claude, Gemini or another AI chat, it notices and offers a safer place: the site’s temporary chat, or Melo, Kamelo’s private chat, which answers on your own computer. It doesn’t change or hide anything in your message. The extension doesn’t talk to any server, not even ours, and the code is open source.

There’s always a catch, so here it is:

• Melo is small, and Chrome only. It uses the AI built into Chrome. That’s private, but less capable than ChatGPT. In Edge and Firefox, you get the temporary chat.
• A temporary chat still goes to the AI company. It’s just kept out of your history and training.
• Your AI account still knows it’s you. Kamelo can’t change that.
• Kamelo can miss a personal question. It helps you choose. It doesn’t decide for you.

And be picky about extensions, ours included. In December 2025, researchers found that a popular free VPN extension had been collecting people’s AI chats (Malwarebytes). What happened. Check what an extension can see before you add it.

AI can really help with personal things. Knowing where your words go lets you choose where to ask.