What you share when you use AI

📋 30-second summary

• What you type goes to the company’s servers (OpenAI, Anthropic, Google) and stays there for a while.
• Three destinations: stored in your account, possibly used to train future models, sampled for safety review.
• Rules depend on your plan: free and consumer can be used for training (you can turn it off in settings), business plans are opted out by default.
• Never give it: other people’s data, third parties’ sensitive documents, professional secrets, passwords or card numbers.
• Mental test: “if this showed up in a Google search a year from now, would it be a problem for me?” If yes, think twice.

When you type a message to ChatGPT, Claude, or Gemini, that message doesn’t vanish into thin air. It travels to a company’s server, gets read by a program that produces a reply, and leaves a trace that someone holds on to for a while.

Knowing what actually happens lets you decide calmly what to share and what to keep out. This isn’t a lesson for getting paranoid: it’s here so you can use AI casually for everyday things and pause for a second when you’re about to give it something sensitive.

What the AI sees when you talk to it

Everything you type in the input box gets sent to the company that runs the service: OpenAI for ChatGPT, Anthropic for Claude, Google for Gemini. Not just the text: if you upload a PDF, a photo, or audio, it sends those too.

There isn’t a person peeking at you from the other side. Your message turns into numbers, the model produces a reply, and it comes back. But a copy of what you wrote does stay on the company’s servers, at least for a while.

Serious companies don’t read conversations for fun. They do run sample-based checks to catch abuse (threats, child sexual abuse material, fraud) or to improve the product: if your conversation ends up in that sample, it may pass through the eyes of someone on their team, under confidentiality rules.

What if I use a less famous AI app?

“Serious companies” here means ones with a recognizable name, legal documents published openly on the site, and an app downloaded from the official Play Store or App Store. If a free AI app you stumbled on doesn’t have a readable privacy policy, or asks for odd permissions (contacts, address book, always-on microphone), don’t give it personal data. The rules in this lesson apply to OpenAI, Anthropic, Google, and a handful of other big companies. Unknown apps deserve more caution, not less.

What happens after you send the message

Three things, with variations from product to product:

1. Storage. The conversation gets saved in your account so you can find it later. It usually sits there until you delete it yourself.
2. Training The phase in which an AI learns by reading huge amounts of text: books, web pages, conversations, code. From that material it picks up its patterns. of future models. Some companies use what you write to improve their next models. It doesn’t mean the next version will spit your conversation back out: your text gets mixed in with millions of others.
3. Safety review. Small samples can be read by designated staff, to catch abuse or improve the product.

Can it “repeat” what I wrote to someone else?

In practice, no. When a model is trained on millions of conversations, what you wrote becomes a drop in the training ocean. There are rare cases where a model reproduces chunks of its training data word-for-word, but they involve things published in many copies (the Wikipedia front page, lyrics of famous songs), not a unique conversation like yours.

Key point: these rules change depending on the plan you’re on (the type of subscription). You can see it at the top of your account settings: “Free”, “Plus”, “Pro”, “Team”, and so on.

• Free and consumer plans (ChatGPT Plus, Claude Pro, personal Gemini): even if you pay for a subscription like Plus or Pro, from a privacy standpoint you’re in the “consumer” bucket. By default, conversations can be used for training, but you can turn that off in the settings.
• Business plans (Team, Enterprise, Business): by default they are not used to train models, and the company makes contractual commitments to your employer.
• Developer APIs: different rules again, usually no training by default.

If you really care, look for it in the settings of the product you use: the option is usually called “Data controls”, “Privacy controls”, or “Privacy & security”.

What you should never give it

A simple mental test: “if this showed up in a Google search a year from now, would it create a problem for me?” If the answer is yes, think twice.

In practice, keep these out of free-plan and personal chats:

• Data about people other than you. If you paste a coworker’s email as is, you’re sharing her address, her topics, her writing style. Those aren’t your data to give away.
• Sensitive documents belonging to others. Family members’ medical reports, legal documents about relatives, third parties’ tax data.
• Professional secrets. Non-public company code, commercial strategies, confidential pricing, client lists. Many companies have explicit policies: before pasting something work-related into “personal” ChatGPT, check whether a dedicated business plan exists.
• Credentials. Passwords, API keys A kind of technical password developers use to let an application talk to a service. Not something you'd use unless you're coding, but it sometimes shows up in work files. , credit card numbers, PINs. Never, for any reason.

For your own personal data the calculation is different: if it helps you to have the AI explain your own bill, that’s your call. Just remember that the bill, with your name and address on it, leaves a trace.

How to share only what matters

Before pasting an email or a document, strip out what the AI doesn’t need to help you: names, addresses, phone numbers, codes. Replace “Dear Francesca” with “[name]”, replace an address with “[address]”. The AI still follows the context, and you’ve given away less. Works for coworkers’ emails, bills, medical reports, contracts.

Settings you can change

You don’t have to trust blindly: serious AIs give you real controls.

• Turn off use for training. On ChatGPT: Settings → Data controls → turn off “Improve the model for everyone”. On Claude: by default, standard-plan users’ conversations are not used for training. On Gemini: Settings → Activity → check the “Gemini Apps Activity” option.
• Temporary chats. A “throwaway” mode where the conversation isn’t saved and isn’t used for training. ChatGPT calls it “Temporary chat”; other platforms have equivalents.
• Delete conversations or your whole history. Always available, in your account settings.

Heads up: turning off training doesn’t delete the conversation. Your text still sits on the company’s servers for a while (typically 30 days, even after you delete it yourself). That’s the time required for fraud-prevention obligations, responding to legal subpoenas, and internal audits. After that period it actually gets deleted.

How do I check training is really off

After changing the option, close the settings page and reopen it: if the option is still off, the change was saved. If it flips back on by itself, the save didn’t go through (it happens, rarely). As an extra check, ChatGPT shows a small note next to the input box when you’re in a chat that won’t be used for training.

A mindset, not a checklist

Treat AI like a trusted colleague whose full information chain you don’t know. You trust it to hear about your day, to help with a hard email, to summarize an article. You wouldn’t give it your bank password, you wouldn’t let it read a friend’s private chats.

If you notice you’re about to give it something sensitive, pause for a second and ask whether you actually need to for what you’re doing. Most of the time the answer is yes and you go ahead without worrying. Sometimes it’s no, and you just saved yourself a headache.

Check what you understood

📝 Check-in with your AI

Four questions to close the module

5-7 min of check-in

Paste the text into your AI. It will ask you four questions to check whether you understood what you share when using it.

Copy the text below and paste it into your favourite AI (Claude, ChatGPT, Gemini, or another). Or download the file and give it to your AI as an attachment.

Copy the text Download Open

What comes next

You’ve finished the Fundamentals module. From here on, the manual moves into the Everyday use module: using AI to write better, summarize documents, understand complicated things, learn new topics, plan.