Why this matters
Someone in the office pastes a client’s email into ChatGPT to save five minutes. The email has the client’s full name, home address and new bank account numbers in it. Now those details sit in a chat on a personal account that may be used to train the AI.
Then the reply comes back sounding warm and confident, with a promise nobody made: "your payment details are updated", "our team will be there tomorrow at 9". It goes out unchecked. Either mistake can cost you a client or a complaint, and neither one was the AI’s fault.
In this module you learn the three habits that make AI safe to use at work: check one setting, take out what the AI does not need, and check what comes back before it goes out under your name.
What you’ll build
You are Jordan, who manages rentals at Lakeside Property Management (a made-up company). A tenant has emailed her new bank details and a complaint about a leaking dishwasher. You switch off model training, swap her name, unit and bank numbers for placeholders like [TENANT] and [NEW BANK DETAILS], and get a friendly reply drafted.
Then you catch the two things the reply made up: that her autopay is already updated, and a repair visit time nobody booked. You finish with a one-page checklist for your team.
What you need
- Nothing to start the practiceChatGPT is a practice copy inside the lesson. No account is needed, no setting really changes, and the bank numbers are made up.
- A ChatGPT account, to do it for realA free account at chatgpt.com is enough. On ChatGPT Business or Enterprise, your admin controls the data setting for everyone.
- The training settingIn ChatGPT, click your name at the bottom left, then Settings, then Data controls. The switch is called Improve the model for everyone.
- Your company’s AI policy, if it has oneAsk your manager or IT. If there is a policy, it comes first. If there is not, this module’s checklist is a good start for one.
Concepts
3 ideas you’ll use in the practice.
Training: whether your chats teach the AI
On personal ChatGPT plans (Free, Go, Plus and Pro), your conversations can be used to improve OpenAI’s models unless you switch that off. Switching it off covers new chats. It does not delete your history.
Business and Enterprise plans do not train on your business data by default. That is one of the main reasons companies pay for them instead of letting staff use personal accounts.
Placeholders keep the meaning, not the risk
To write a friendly reply, ChatGPT needs to know what happened, not who it happened to. Swap the real details for placeholders like [CLIENT], [ADDRESS] or [ACCOUNT NUMBER], and the reply comes back just as good.
Never paste passwords, full card or bank numbers, Social Security numbers, health details, or documents you promised to keep confidential. You put the real details back yourself when you send.
AI fills gaps with things that sound right
When the AI does not know something, it rarely says so. It writes the most likely sentence, and "I have updated your autopay" is a very likely sentence. That is how a promise nobody made ends up in a client’s inbox.
The fix is a quick check. Ask: "List every fact or promise in your reply that I did not give you." Then read it yourself for names, numbers, dates and promises, because the AI’s own check can miss things too.
Practice
Do every step yourself in a practice copy of the real tool. It checks your work, and “Do it for me” finishes any step you get stuck on.
Do it for real
Now repeat it in your own account. Tick each step as you go.
Downloads
- Safe AI at work checklist (.txt)The one-page checklist from the end of the lesson, plus the check prompt to paste after any draft.
Quick check
3 questions to check you’ve got it.
1. You want ChatGPT to draft a reply to a client who sent their bank details. What do you paste?
- The whole email, exactly as it arrived
- The email with the name, address and bank numbers swapped for placeholders
- Only the bank numbers, so it can check them
Show the answer
The email with the name, address and bank numbers swapped for placeholders. ChatGPT writes the same reply without the private details. They only add risk.
2. What does switching off Improve the model for everyone do?
- Deletes all your old chats
- Stops your new chats being used to train the AI
- Makes your chats completely secret
Show the answer
Stops your new chats being used to train the AI. It covers training on new chats. Your history stays, and chats are still processed to answer you, so placeholders still matter.
3. The draft says "Our team will be there tomorrow between 9 and 11." You never gave it a time. What happened?
- ChatGPT checked your calendar
- ChatGPT filled a gap with something that sounds right
- The tenant asked for that time
Show the answer
ChatGPT filled a gap with something that sounds right. AI fills gaps with likely sentences. Ask it to list what it added, and check every promise yourself.