What happens to what you type

It is late and the room is quiet. You type something into the AI app that you have not said to anyone. A money problem, a fight at home, a question about your body. It feels private, like talking to yourself in an empty room.

It is not private in that way. This lesson tells you what it is instead, and what you can change.

Where your message goes after you press send

Your phone does not answer you. It cannot. The model is far too big to sit inside a budget Android phone.

So your words leave your phone. They travel over your mobile data to a company’s computers, thousands of kilometres away. The answer comes back the same way, in about a second. Where your question actually goes follows that whole journey.

Those computers do not read your message and forget it. They store it. That is exactly why you can open the app tomorrow and see yesterday’s chat sitting there.

So here is the honest picture. A chat is not a diary in a locked box. It is more like a letter you handed in at a company office. The office keeps a copy of the letter.

Saved and used for training are two different things

People mix up two questions here. Keep them apart, because the app has separate settings for them.

The first question is: is my chat saved? Almost always yes. It is saved in your account so you can open it again. Companies also keep messages for a while to check for abuse of the tool.

The second question is: are my words used to improve the model? That means your chat becomes practice material for the next version. You saw how that practice works in how a model is made.

These two are not one switch. Turning off training does not delete your history. Deleting your history does not turn off training.

The apps differ, and they change their rules. In August 2026, ChatGPT used chats from ordinary personal accounts to improve its models by default. Free and paid personal accounts both. You had to go and turn that off yourself.

Claude changed its rules for ordinary users in 2025, after many people had already signed up. Users were asked to choose. Allowing training meant chats were kept for 5 years. Saying no meant the older, shorter period of about 30 days.

Google keeps your Gemini activity by default and deletes it automatically after 18 months. You can turn that keeping off. Even then, Google holds chats up to 72 hours simply to run the service.

Finding the switch in your own app

App screens change every few months. So learn the shape of it, not the exact taps.

Start where your name or profile picture is. From there, open settings. Then look for a section named for data, privacy or activity. The words differ by app. The section is always there.

Inside it, look for a switch worded something like improve the model for everyone, or keep activity, or help improve our services. Read the full sentence printed next to the switch. That sentence tells you what the company actually does.

While you are in there, answer these four questions for yourself.

1. Are my chats saved in this app?
2. Is there a switch about improving the model?
3. Is it on or off right now?
4. Where do I delete one chat, and where do I delete all chats?

One more habit is worth building. When the app shows a pop-up about new terms or new privacy rules, do not tap Accept straight away. Look for the switch inside the pop-up first. The setting it comes with usually suits the company, not you.

A real person may read some chats

Google says plainly that some Gemini chats are read by human reviewers. Not all of them. A small share, picked for safety checking and for improving quality.

Google puts its own warning in writing on its privacy page.

Please don't enter confidential information that you wouldn't
want a reviewer to see or Google to use to improve our services.

There is more. Chats picked for human review are kept for up to 3 years. Deleting your activity does not delete those. They are only cut loose from your account.

Do not read this as somebody watching you type. Nobody is sitting there. Most chats are never opened by a person at all. Treat it as a thing that could happen, and plan for it.

The plan is one sentence. Type only what you would be willing to read out loud in the village square.

That rule covers photos too. If you upload a picture of an Aadhaar card, a passbook or a marksheet, the whole picture goes across, numbers and all. Cover the numbers first, or type the words without them. Giving it your own book, notes or photo is still the best way to get good answers, and it works fine with the numbers hidden.

Here is a bank form. I have put XXXX where my real details go.
Name: XXXX
Account number: XXXX
Village and district: XXXX
Explain what each line of this form is asking me for, in simple words.

What delete really does

Deleting a chat takes it out of your view. That is real, and it is worth doing. On a shared phone it is the single most useful thing you can do.

But be clear about the limits. There are three.

First, copies may stay on the company’s computers for a while, for abuse checks. One company keeps messages about 30 days for that reason, even after you turn training off.

Second, if a chat was picked for human review, deleting your activity may not remove it. That is the 3 year point from the section above.

Third, anything already used as practice material cannot be pulled back out of a model. Turning training off today is not backwards. It changes what happens from now on.

None of this is a reason to panic. It is a reason to decide before you type, not after.

Two ways other people end up reading your chat

The first way is the share button. Pressing share does not send a copy. It makes a web page with an address, and anyone holding that address can open it. There is no password on it.

In August 2025 this went wrong in public. Thousands of shared ChatGPT chats turned up in Google search results. People had shared chats about their health, their work and their family. The company turned that option off and worked with search engines to take them down. Copies can still sit in caches for a while.

So the rule is short. Treat share as publish. Never press share on a chat that holds your phone number, your family details or your money troubles.

The second way is memory. Many apps now keep short notes about you that carry across chats. Your name, your exam, your village, the work you do. It makes answers better, and most people never look at the list.

On a shared phone, the next person can read that list. If you use AI to help your family, it may hold their details as well as yours. Find the memory section in settings, read every line, and delete the ones you do not want kept. On a shared or cybercafe computer, log out before you stand up.

What the law here gives you

India has its own data protection law now. The rules under it were notified in November 2025. They arrive in stages over about 18 months, and companies have to follow the main rules from 2027.

In plain words, it gives you rights over your own information. You can ask a company what personal data it holds about you. You can ask it to correct that data, update it or erase it. You can also name someone to act for you if you cannot act yourself. The company has to reply to you within 90 days.

There is one more part that matters here. Under this law, anyone under 18 is a child. A parent or guardian has to agree before a company handles a child’s data.

Note the gap. The AI companies’ own terms usually say 13 and above, with a parent’s permission. Indian law counts you as a child until 18. So if you are 15, 16 or 17, use these tools in the open. Tell a parent or an elder brother or sister that you use them, and what for. That is honest, and it saves an argument later.

Do not wait for the law to protect you. Most of it is not fully in force yet, and a rule on paper cannot take back a photo you already sent. Prevention is the part you control today.

So carry one sentence out of this lesson. Assume anything you type could be read by someone, and let that decide what you type.

Next comes AI scams and how they work. This lesson was about your own words going out. That one is about words coming in, from a voice or a face that is not real.