Team leads should start with Use AI at Work Without Leaking Client Data, a 7-minute ChatGPT lesson, because your team is probably already using AI and the first risk is not a bad answer. It is someone pasting private details into a tool nobody has checked. This path takes you from the three habits that prevent most problems to a written team policy and a plan for the day something goes wrong.
Every lesson here uses a made-up office, so you can practice the settings and the wording without touching a real account. You do not need to be technical. You need to decide the rules, and the lessons show you how to write them down.
Which lesson should you take first?
Take Use AI at Work Without Leaking Client Data first. You play an office worker handling a tenant's email, with bank numbers and an address in it. You check ChatGPT's training setting, take out the details ChatGPT does not need, then catch the two promises ChatGPT made up in its reply before it goes out. You finish with a one-page checklist.
After this lesson you can give your team three habits in one sentence: check the setting, take details out, check the reply before it goes.
What comes next, and why in this order?
The order goes from what individuals do, to what the team agrees, to what you do when the rules are broken.
Step 2: How do you decide what never goes into AI?
Know What Never Goes into AI (7 minutes, ChatGPT) gives you a four-level sheet: Public, Internal, Confidential and Restricted, with one AI rule for each. You sort four items for Harbor & Pine Bookkeeping, take a Social Security number out by hand, then have ChatGPT redact a client email in a Temporary Chat and catch a mix-up it makes. The lesson's rule of thumb is that if you are unsure, treat the data as one level higher.
Step 3: Which ChatGPT settings should you change?
ChatGPT: Keep Client Data Safe (about 10 minutes) walks through the settings that matter: Improve the model for everyone, shared links, export, memory and Temporary Chat. It ends with a one-page team AI policy you can copy. See the plain-words companion, ChatGPT Data Controls: Which Settings to Change for Work.
Step 4: How do you clean up what is already in there?
Clear Client Data Out of ChatGPT (about 10 minutes) covers the cleanup that is easy to get half right. You delete one chat, archive a chat you keep, delete the one memory it left behind, set safe defaults and start using Temporary Chat. The lesson says to leave Delete all chats alone and to export before any bigger cleanup.
Step 5: How do you write the team policy?
Roll Out a Team AI Policy in One Afternoon (Claude) has you decide the rules: which tools, what never goes in, who checks the work and who to ask. Claude writes them as a one-page policy for the made-up Northfield Consulting Group. You check the never-paste list, the human-review rule and the no-blame reporting rule, then get an announcement email and a five-question quiz.
Step 6: What do you do when someone pastes client data?
Someone Pasted Client Data into AI (about 9 minutes, ChatGPT) is the plan. You follow a one-hour incident log. Delete the chat, delete the memory it left behind, check shared links and turn off training. Then you write the incident note, the client email and a new policy rule, and catch the two things ChatGPT got wrong when we tested it.
How fast should a team lead go?
Take the first two lessons in one sitting, then set aside an afternoon for the policy. Use the checklist from Step 1 with your team for a week before you move on. The whole path is under an hour of lessons, and the work that matters is the conversation with your team.
Costs to plan for when you do this for real:
- ChatGPT lessons: a free account works to start. On a work plan your admin controls some of the data settings.
- Claude lesson: a Claude account. Check Anthropic's current plans.
- Anything on a company card: get the owner's OK first and check each vendor's pricing page.
What should you do after the path?
Pick one rule and enforce it for a month before you add another. Compare your tools fairly too: Choose the AI Assistant Your Team Should Use tests ChatGPT, Claude, Gemini and Copilot on one office task and picks the one a company already pays for and allows.
Prompts that fit this role: draft an AI use policy, plan employee AI training, sort documents into data levels and score four assistants on the same task. Terms you will run into: AI acceptable use policy, data classification, Temporary Chat, shadow AI and human in the loop.
For the same rules in a small business setting, read the AI governance for small business explainer.
What should a team lead not hand to AI?
Do not hand AI the decisions about people or the rules themselves. Use it for drafts, summaries and checks, and keep the decisions with a person.
- Policy decisions. Claude can write the wording, but you decide which tools are allowed and what never goes in.
- Anything about a person. Pay, leave, discipline and complaints belong with HR or a qualified adviser.
- Restricted data. The four-level sheet says Restricted data never goes into AI. Remove it by hand.
- Legal or insurance requirements. The lessons are practical guidance, not legal advice. Check your contracts and your insurer.
More paths: the AI Learning Path for Office Managers covers customer replies and meeting recaps, and the AI Learning Path for Operations Managers adds a deck and an assistant comparison.
Key takeaways
- Team leads should start with Use AI at Work Without Leaking Client Data, a 7-minute lesson on three habits.
- The path moves from data levels and ChatGPT settings to cleanup, a written team policy and an incident plan.
- A four-level data sheet gives your team one AI rule per level, and if in doubt they treat data as one level higher.
- Settings only help if everyone uses the same ones, so write them down and review shared links monthly.
- Decisions about people and the rules themselves stay with a person.
Frequently asked questions
- Why start with what not to paste instead of the policy?
The policy is more useful once your team has felt the three habits. The first lesson is short and gives you the vocabulary for the rest.
- Do I need to be technical to take these lessons?
No. Every screen is a practice copy and you click and paste what the lesson gives you. The policy lesson asks for your decisions, not code.
- Can I use these lessons if my company uses Copilot or Gemini?
The settings lessons are about ChatGPT. The data levels, the policy and the incident plan apply to any assistant, and the assistant comparison lesson helps you choose.
- Is this legal advice?
No. It is a practical path. Check your contracts, your industry rules and your insurer.