Your AI Chatbot Knows More About You Than Your Friends Do

I had a document open that contained other people's personal details, and I pasted a section of it into a chatbot to summarise it. I did this in about four seconds and did not think about it once.

It summarised beautifully. I also spent the following weekend finding out exactly what I had done.

Nobody tells you this at the moment you need to know. Here is what I learned, minus the panic.

What actually happens to your chat

Three separate things, and people mix them up constantly:

• Your conversations may be retained and reviewed, in some cases, by a human, for quality and safety
• They may be used to improve models, depending on your settings and your account type
• They can be used against you legally in some jurisdictions, because a chatbot transcript is a document you created

That third one is the one that changed how I behave. A chatbot conversation is not a private diary. Treat every message as if a screenshot could appear in an email.

The seven things I no longer paste

1. Anything about other people — clients, patients, colleagues, your ex
2. Unredacted personal data — real names with addresses, phone numbers, ID numbers
3. Unpublished work — your employer's code, your unreleased designs, anything under NDA
4. Financial specifics — account numbers, card details, tax documents
5. Passwords and keys — an AI cannot and will not safely hold a secret
6. Medical or legal detail — about yourself or anyone else
7. Anything you would not screenshot

The pattern: if it is someone else's information, or it is secret, or it is legally privileged — it does not go in.

The eight-second habit that fixed it

I now paste this before the real text:

The following is for analysis only. Replace all identifying details with generic placeholders first. Do not guess at specifics.

Then I paste redacted text. It takes eight seconds and it has made the habit automatic. Once it is automatic, you stop relying on remembering to be careful.

What the model versus the platform distinction actually means

People say things get "used for training" as if that is the scary part. The platform-level setting matters more, because that is what determines who inside a company can read what you wrote.

Go and check your settings. Actually check them. Look for the control over whether your conversations are used to improve models, and set it the way you want. Then check whether you can delete your history, and delete anything sensitive. On some platforms you can turn training off, on some you cannot, and on some it is tied to your plan. Knowing which is true for you is the entire point.

The workplace question nobody asks

Your personal account and your employer's data are not the same thing. A lot of people are pasting customer records into a personal chat in breach of a policy they never read.

If you need to use AI for work, use the account your employer bought for it. It is less convenient and it is the one where the contract actually covers you. If your employer has not given you one, that is a conversation worth having in writing.

What I would do differently

Not "never use it". I use it constantly, and it is genuinely one of the best tools I have. The change is that I stopped treating it like a search engine and started treating it like a colleague who is sitting in an office where anyone might walk past and read over your shoulder.

Eight seconds of redaction. That is the whole habit. It costs less than the weekend I lost.