Everything you type into a chatbot leaves your device. That's usually fine — but there are a handful of things you should never hand over, and one simple habit that makes almost everything safe.
By the end, you'll know:
One short lesson that protects you for every lesson after it.
It helps to stop thinking of the chat box as a private notepad. When you type into a chatbot, your text travels to a server owned by a company, is processed there, and may be stored, logged, or reviewed.
So the test is simple: would you be comfortable emailing this text to a company you don't control? If not, don't paste it.
These shouldn't go into a consumer chatbot, ever:
Personal identifiers
Full names paired with addresses, phone numbers, dates of birth, ID or passport numbers.
Financial details
Card numbers, bank account details, account balances, tax IDs.
Health information
Diagnoses, prescriptions, medical records, insurance details — especially anyone else's.
Confidential work data
Client lists, contracts, unpublished financials, internal strategy, source code you don't own the rights to.
Passwords & keys
Passwords, API keys, login credentials, recovery codes. No exceptions.
Other people's data
Anything about a customer, patient, employee, or student that they haven't agreed to share.
Here's the good news: you can almost always get the same help without the sensitive details. Swap the specifics for placeholders.
You get the same writing help. The model never learns who your client is, what they owe, or that the invoice exists.
Policies differ by provider and change often, so check your tool's own privacy settings. But the usual picture is:
Free tiers often train on your chats unless you opt out. Paid or business tiers typically don't by default.
Chats may be kept for a period for safety, abuse review, or legal reasons — sometimes long after you delete them.
Human reviewers may read flagged chats at some providers. Another reason not to paste what you wouldn't email.
Deleting helps but isn't a guarantee. Best habit: never put it in to begin with.
Open your chatbot's settings and check these. The names vary slightly by product, but the options are usually there:
Turn off "improve the model" / training if you'd rather your chats not be used for training.
Check chat history & retention settings and clear old conversations you no longer need.
Review connected apps and extensions — revoke anything you don't recognise or use.
Turn on two-factor login so nobody else can read your history.
Safety isn't only about data. It's also about what you do with the answer — and this connects straight back to Lesson 5.
Medical, legal, financial, safety — treat AI as a starting point for questions, never the final answer. Talk to a qualified person.
Anything affecting another person — a rejection, a review, a decision — needs a human in the loop.
If your name goes on it, you own it. That's true of a colleague's draft too.
When a new employee asks for an example, you don't hand over the whole filing cabinet. You give the one document that does the job.
Minimum necessary information is both the safest and the most effective brief. Less clutter, better answers, nothing leaked.
Your feedback shapes what we build next.
The last mile: editing, finding your own voice, and making the output genuinely yours.
The broader set of habits that keep AI useful and honest.
Privacy protects your data; verification protects your decisions.
Back to the full tutorial library.
