AI chatbots feel private. You type into a box, an answer appears, and nobody else seems to be involved. That feeling makes it easy to paste in a medical letter, a payslip, a tricky message about a family member, or the full text of a work contract without a second thought.
The short answer to "is it safe?" is: it depends on what you share, which service you use, and which settings you have on. Using an AI chatbot is not inherently dangerous, but it is not like writing in a private diary either. This guide explains what generally happens to your messages, what you should keep out of any chatbot, and the simple habits that protect you.
Every AI service works a little differently, and policies change, so always read the privacy page of the service you use. But the general pattern across consumer AI chatbots looks like this.
The AI model does not run on your phone or laptop. When you press send, your message, and usually the recent conversation history, is sent over the internet to servers where the model runs. If the chatbot is built on a model from another company, your message may pass through more than one organisation: the app you are using and the AI provider behind it.
Many services keep a history of your conversations so you can scroll back or continue them later. Separately, providers often keep messages for a period for purposes such as security, abuse detection and troubleshooting. Deleting a chat from your screen does not always mean every copy disappears immediately; backups and retention periods vary.
Some providers use conversations from free consumer accounts to improve their future models, often by default with an option to opt out. Some also have staff review a small sample of conversations for quality and safety. Business and paid plans often have different, stricter terms, but that is not guaranteed for every service. We cover this in more detail in Do AI Chatbots Train on Your Chats?
Privacy risks with chatbots usually come from ordinary things: a shared or stolen device with your chat history open, a data breach at a service, a weak password on your account, or a browser extension that can read pages you visit. Thinking about these is as important as thinking about the AI itself.
Some information is risky to share with any online service, and a chatbot rarely needs it to help you. Keep these out:
Avoiding all personal context would make AI much less useful. Saying "I'm 52, I have a bad knee and I want to start exercising" gets you far better advice than a generic question, and it does not identify you. The test is simple: could this detail, on its own or combined with others, identify me or harm someone if it leaked?
General facts about your situation, your age range, your country, your goals and your preferences are usually low risk. Your full name with your address, employer and date of birth together are not.
Before pasting a document, swap identifying details for placeholders. "Write a reply to [LANDLORD] about the leak at [ADDRESS] reported on [DATE]" gets you the same quality of help as the real details. You can fill them back in yourself afterwards.
A temporary chat is usually not saved to your visible history, and the conversation disappears when you close it. That protects you from someone opening your account later and reading it. It does not usually mean your messages skip the AI provider: the model still has to receive your message to answer it, and the provider may still keep it for a limited time under its own terms.
Deleting history works similarly. Deleting a chat removes it from your account, and most services delete underlying copies over time, but backups may persist for a while. Read the service's privacy policy for the details, and if you need to, most providers let you request deletion of your account and data.
Since we are asking you to read the fine print elsewhere, here is ours in plain terms:
Automatic detection of personal details is helpful but not perfect, so the habits above still apply on every plan. The plain-English summary is in our Privacy and safety guide, and the full legal version is in the Privacy Policy.
AI chatbots are safe enough for most everyday use if you treat them like any other online service: share what helps the answer, leave out what identifies you or others, and know where your history is kept. A few seconds spent removing a name or an account number is the most effective privacy setting there is.
If you'd like to try an AI chat with these options built in, Tobava AI is free to try, no account needed.