AI privacy and safety: send less, get the same help

Each message leaves your device to reach the model, and most tasks do not need your real name, numbers or addresses. Here is the route your words take on gpt.me and simple habits that keep private details private.

5 min readUpdated

Typing into a chat window can feel like writing in a diary. But every message you send leaves your device and is processed by computers elsewhere, as it is with any online AI service. That is fine for the great majority of tasks, as long as you choose what goes in. Most requests work just as well without your real name, address or account numbers. This guide explains the route a message takes on gpt.me, what to keep out of any AI chat, how to redact text and photos in seconds, and how to use AI sensibly for health, money and legal questions.

Where your message goes on gpt.me

gpt.me is an independent site, not run by OpenAI. When you press send in our chat, this is what happens:

  1. Your browser sends the message, plus the recent part of the conversation, to our server.
  2. From there it goes to our AI gateway, then to OpenRouter (a routing service for AI providers), and on to the OpenAI model you picked: GPT-6 Luna or GPT-6.1 Terra.
  3. The answer streams back the same way and appears in your browser.

Along the way, we do not store the text of your messages. Our gateway keeps technical records only: the model used, how large the request was, when it happened, and your IP address in hashed form, a one-way scrambled code. The hashed address is also what we use to count the daily free allowance and to block abuse. Images you attach are sent for that one answer only.

Your chat history is kept in your browser on your device, not on our server. That is why it does not follow you to another phone or computer, and why clearing your browser's site data erases it. The full details are in our privacy policy.

What OpenAI says about API data

Our chat reaches OpenAI's models through its API, the developer service, not through ChatGPT. OpenAI's published data controls for the API say that data sent to it is not used to train its models unless the customer opts in, and that abuse-monitoring logs, which may contain prompts and answers, are kept for up to 30 days unless the law requires longer. OpenRouter, which sits between our gateway and OpenAI, has its own privacy policy. Rules like these can change, so treat any online AI service as a place where text might be seen by systems or people other than you.

The practical takeaway is the same as for e-mail or cloud documents: share what the task needs, and keep the rest out.

Never paste these

Whatever chat you use, some things should not go into it:

  • Passwords, PINs and one-time codes, including the ones in screenshots of login pages.
  • Full card numbers, bank details, tax and ID numbers.
  • API keys, tokens and private keys in code you paste for review.
  • Other people's private information, such as a friend's medical history or a colleague's salary, unless they agreed.
  • Confidential work material your employer or a contract forbids sharing with outside services.
  • Anything you would be badly hurt by if it leaked. If in doubt, leave it out.

Redact without losing the help

Replacing details with placeholders keeps the meaning and removes the risk. The model does not need real names to fix a letter's tone.

Weak (too much personal data):
I'm Maria Kowalski, 14 Elm Road, account 55301278. My bank
charged me twice on 3 March. Write a complaint.

Better:
I'm a customer of [BANK]. I was charged twice for the same
payment of [AMOUNT] on [DATE]. Write a short, polite complaint
asking for a refund within 14 days. Leave placeholders for my
name and account number.

A few quick habits:

  • Use consistent placeholders like [NAME], [CLIENT], [CITY], so you can fill them back in with find and replace.
  • Round or shift numbers when only the shape of the problem matters.
  • For long documents, paste only the section you need help with.
  • For a summary of a long private document, the summarizer works the same way: redact first, then paste.

Try it with a ready-made request: ask GPT to redact a text for you. Check the result before you use it, as it may miss something.

Photos and screenshots

A picture often shows far more than the question needs. Check it first:

  • cut away everything except the area your question is about;
  • check the edges of screenshots for names, e-mail addresses, open tabs and notifications;
  • cover faces, number plates, addresses on envelopes and documents you did not mean to include;
  • do not photograph ID cards, bank cards or prescriptions with your details showing.

Our chat sends a picture only with the message you attach it to. Shared conversation links do not include images. More advice is in the guide to asking about images.

Because your chats live in your browser, anyone who uses the same browser profile can open them. On a shared or public computer, delete the chats when you are done, or use a private window, which forgets everything when closed. You can rename, pin and delete chats in the sidebar, delete all of them in settings, and export the ones you want to keep as a file.

A share link creates a read-only copy of a conversation on our server, at an address like gpt.me/s/ followed by a code. Anyone with the link can read it. Shared pages are not indexed by search engines, and they stay online until removed; to have one taken down, use the contact form. Before sharing, scroll through the chat for anything personal.

AI can help you understand a diagnosis letter, prepare questions for a doctor, compare loan offers or decode a contract clause. It can also be wrong while sounding certain, and it does not know your full situation. Use it for general understanding, then confirm with a professional before you act. Do not change medication, sign a contract or make a large financial decision based on a chat answer alone. In an emergency, contact local emergency services, not an AI.

The postcard test: before you send something, ask yourself whether you would be comfortable writing it on a postcard. If not, replace the sensitive parts with placeholders. You will almost always get the same quality of help.

FAQ

Do you read my conversations?

No. We do not store message text on our server. Your history stays in your browser unless you create a share link, which stores a copy of that one conversation.

Does the chat remember me between conversations?

No. Each chat stands on its own; the model has no memory across chats. Your old conversations are saved only in your browser, where you can reopen them.

Do I need to sign in?

No. There is no account on gpt.me, so we do not hold your name or e-mail address. The daily allowance uses a hashed IP address instead.

Is it safe for children?

The site is not meant for children under 13. Teenagers should follow their school's rules, and parents may want to read our terms and the FAQ together with them.

How do I delete everything?

Open the chat settings and choose to delete all chats, or clear this site's data in your browser. Share links you created stay online until you ask us to remove them through the contact form.

Got a question? Ask it now

No account and no download. Open the chat, type what is on your mind, and the answer starts appearing in a second or two.

Start chatting