AI Chat Privacy and Safety Guide

AI chat feels private: it is just you and a text box. But every message you send travels to a server, is processed by an AI model and may be stored for a while, depending on the service. This guide explains what generally happens to your data, what you should never paste into any AI chat, how ChatG handles your conversations, and how to stay safe from the scams that AI makes easier.

What happens to a message you send

Details differ between services, but the general path is the same:

1. Your message, and usually the earlier messages in the conversation, is sent over an encrypted connection to the service's servers.
2. The service passes it to an AI model, which may be run by the service itself or by a third-party model provider.
3. The answer is generated and sent back to you.
4. Depending on the service and your settings, the conversation may be saved in your account, kept in logs for a period for security and abuse monitoring, reviewed by people, or used to train future models.

Several major consumer AI services may use your conversations to improve their models unless you switch that off, and some have human reviewers look at samples. Most offer a setting to opt out of training, and many have a temporary or incognito chat mode. Opting out usually applies to future chats only: data that has already been used in training cannot be pulled back out of a trained model. Check the privacy settings and policy of every AI service you use, because they change from time to time.

Never paste these into any AI chat

• Passwords, PINs, recovery codes and one-time codes. No legitimate AI task needs them.
• API keys, access tokens and private keys. Replace them with placeholders.
• Full card numbers, online banking details and government ID numbers such as passport or social security numbers.
• Other people's personal data, such as customer details, patient records or a colleague's HR file, unless you are allowed to share it and the tool is approved for it.
• Confidential work material, such as unreleased figures, contracts or source code, unless your employer has approved the tool.
• Anything you would not want shown on a screen in public. Treat an AI chat like a helpful stranger, not a diary.

How to get help without sharing private details

You can almost always get an equally good answer with the identifying details removed. Replace them with clear placeholders:

Instead of:
My client John Miller at 14 Oak Street owes us 3,400 euros on invoice
2024-118, and his phone number is [real number]...

Write:
A client [CLIENT] owes us [AMOUNT] on an invoice that is 45 days
overdue. Write a firm but friendly reminder email. I will fill in the
details myself.

For health or legal questions, describe the situation in general terms ("a 50-year-old with...") rather than giving your name and records. For documents, delete names, addresses, account numbers and signatures before pasting, or paste only the paragraph you need help with.

How ChatG handles your conversations

Here is what happens on ChatG specifically, in plain terms:

• No account. You do not sign up or give us your name or email address to chat.
• Your history stays on your device. Chat history is saved only in your browser's storage on this device, not on our servers. Clearing your browser's site data deletes it, so export your chats if you want a backup.
• Messages are processed to answer you. To generate a reply, your messages are sent to the AI processing service we use, which relies on third-party model providers. We do not store your messages beyond short technical logs.
• Daily limits without accounts. To count free daily messages and prevent abuse, we use a hashed identifier derived from your IP address.
• Share links can be read by anyone who has the link. If you create a share link, a read-only copy of that conversation is stored on our server at a chatg.org/s/ address. Shared pages are not indexed by search engines, you cannot edit them afterwards, and to have one removed you contact us at hello@chatg.org. Check a conversation for personal details before you share it.
• Ads and analytics. ChatG is free because of ads shown around the content, never inside answers. Our advertising partner and Google Analytics use cookies and similar technologies, and you will see a consent banner where the law requires one.

The full details are in our privacy policy and terms. If you use a shared or public computer, delete your chats in the chat settings when you finish.

Scams that AI makes easier

Scammers use AI tools too. Knowing the patterns protects you and your family:

• Polished phishing. Scam emails and texts no longer give themselves away with spelling mistakes. Judge a message by what it asks for, such as urgent payments, login links, gift cards or codes, not by how well it is written.
• Voice cloning. A short recording can be enough to imitate someone's voice. If a "relative" calls in distress asking for money, hang up and call them back on a number you already know. Some families agree on a code word for emergencies.
• Fake videos and images. Be sceptical of shocking clips of public figures, especially ones asking you to invest or donate.
• Fake AI apps and websites. Some apps pose as well-known AI products to take payments or collect data. Install apps only from official stores and check the developer's name.
• "AI trading bots" that promise guaranteed returns are a common investment fraud.

You can use AI defensively, too. Paste a suspicious message with your personal details removed and ask the chat for the warning signs. Then verify through official channels, never through the links or numbers in the message itself.

Safety when acting on AI answers

• Health, legal and money questions: AI can explain general concepts and help you prepare questions, but it is not a doctor, lawyer or financial adviser. Confirm with a qualified professional before you act. In an emergency, contact your local emergency services.
• Instructions that affect safety: electrical work, chemicals, medication doses and car repairs should be checked against official manuals or with an expert.
• Hidden instructions in content: AI tools that read web pages, emails or documents can be manipulated by instructions hidden in that content, a technique known as prompt injection. Be careful about letting any AI agent act on your behalf, such as sending emails or making purchases, without reviewing each action.
• Children: ChatG is not intended for children under 13. Parents of older children should talk with them about what not to share and why answers need checking.

Our guide to fact-checking AI answers shows how to verify what you read, and how AI chatbots work explains why they make mistakes.

Privacy checklist

• I removed names, numbers and addresses that the answer does not need.
• I did not paste passwords, keys or codes.
• I checked my employer's or school's rules before pasting their material.
• I know where this service stores my history and whether it trains on my chats.
• I read through a conversation before creating a share link.
• I cleared my chats on a shared computer.

More answers about how ChatG works are in the FAQ and on the about page. If you are using AI for code or for work documents, our guides on AI for coding and AI for small business include specific advice on what to keep out of the chat.

FAQ

Can the people who run an AI service read my chats?

Depending on the service, staff or contractors may review some conversations, for example for safety or to improve models. Read each service's privacy policy. On ChatG, your history is stored in your browser rather than on our servers, but your messages are sent to an AI processing service to generate answers.

Does deleting a chat delete it everywhere?

Not necessarily. Deleting a chat usually removes it from your history, but copies may remain in logs or backups for a period set by the service's policy. On ChatG, deleting removes the chat from your browser; a share link you created stays online until you ask us to remove it.

Is private browsing enough to keep my chats private?

Private or incognito browsing stops your browser keeping the history after you close the window, and on ChatG it means your chats disappear when you close it. Your messages are still sent to the AI service to be answered, so it does not hide their content from the service.

What should I do if I pasted something sensitive?

Change any password or key you exposed straight away. Delete the chat and check the service's options for deleting your data. For a ChatG share link, email hello@chatg.org with the link and we will remove it.