Use ChatGPT at Work Without Leaking Client Data: Settings, Rules and a One-Page Policy in One Afternoon
What you’ll be able to do: After this course you can turn off model training and strip a client's details out of a prompt before it goes anywhere, sort your office data into four levels and know which one never goes into AI, set up ChatGPT's data controls, memory, Temporary Chat and shared links for your team in ten minutes, and clear a chat and the memory it left behind when someone has already pasted client data. Then you roll out a one-page acceptable use policy with an announcement and a staff quiz, contain a real incident in one hour with the note, the client email and a new rule, and catch the two attacks that target people who use AI: a poisoned email that talks to your assistant and a fake vendor bank-change request.
- Who it’s for
- Team leads, office managers, owners and anyone who answers for client data at a 5 to 200 person professional firm (accounting, law, insurance, agencies, consulting, healthcare admin). No coding.
- Time needed
- About 66 minutes of hands-on practice: 8 modules of 6 to 10 minutes each.
- Cost
- Free. No credit card, no trial, and no sign-up needed to start.
- Level
- Beginner
- Prerequisites
- None for the practice: a web browser on a desktop or laptop. Helpful first: AI for Executive Assistants and Office Managers: Inbox, Meetings, Vendors and Documents and AI for Nonprofit Staff: Board Minutes, Donor Numbers, Communications and a Safe AI Policy. To repeat the modules in your own tools: a ChatGPT account and a Claude account.
- Format
- Self-paced and online. Each module runs in a practice copy of the real tool that checks every step, then a checklist walks you through doing it for real.
- Tools you practice in
- ChatGPT and Claude.
Why this course
Picture Halvorsen & Reed, a made-up 25-person accounting and advisory firm. Half the staff already use ChatGPT for emails and spreadsheets, with the default settings, on personal accounts. A client has just asked, in writing, whether their files go into ChatGPT. A new hire pasted a payroll export into a chat last week. Nobody has written down what is allowed, and the managing partner has been sent a "new bank details" email that looks exactly like a real vendor's. None of this is unusual. It is what every small firm looks like one year into AI at work.
This course fixes it in one afternoon, in order. Step 1: the two habits that stop most leaks, training off and details stripped out of the prompt. Step 2: the four data levels, with the Restricted level taken out by hand. Step 3: the team's ChatGPT data controls, memory, Temporary Chat and shared links, set in ten minutes. Step 4: cleaning up after a paste, deleting the chat and the memory it left and setting safe defaults. Step 5: the one-page policy, written with Claude from your rules, with the announcement email and a five-question quiz. Step 6: the one-hour incident plan. Steps 7 and 8: a poisoned email that tries to instruct your AI assistant, and a fake bank-change request checked with three verify-before-you-pay rules.
The payoff: a written answer for the client, settings you can show an auditor, a team that knows the rules, and a plan for the day someone slips. You stay in charge the whole way: the modules use made-up firms and data, and nothing you set in the practice touches a real account. Each module is a practice copy of the real screens that checks your work, so you can start now and repeat it in your own ChatGPT in an afternoon.
Who this course is for
Every module is beginner level. You need a browser and a team that already uses ChatGPT or is about to. Do the modules in order; the settings in module 3 and the policy in module 5 are the two you will repeat in your own account the same day.
You can already
- Use ChatGPT, Claude or Gemini at work, or manage people who do
- Handle client, patient, employee or financial data that must not leak
- Nothing else: no code and no accounts to start (the practice runs in your browser)
After this course you can
- Turn off model training and strip personal and financial details out of a prompt
- Sort your data into four levels and name what never goes into AI
- Set ChatGPT data controls, memory, Temporary Chat and shared links for a team
- Clear a chat and its memory after a slip, and contain an incident in one hour
- Roll out a one-page AI policy with a quiz, and catch prompt-injection and bank-change scams
What you’ll build
- Module 1: 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.
- Module 2: You are Priya, office manager at Harbor & Pine Bookkeeping, a made-up six-person firm. You sort four everyday items into the four levels: a services brochure, a staff memo, a client’s payroll summary and a client’s Social Security number.
- Module 3: You are Priya, office manager at Hartwell Bookkeeping, a made-up six-person firm. Starting from the client’s email, you switch off model training, review shared links, find Export data and Delete all chats, decide on memory, and try Temporary Chat.
- Module 4: You are Morgan, a staff accountant at Calloway & Reyes CPAs, a made-up five-person firm. The partner wants the Okafor Family Trust’s details out of ChatGPT before Friday, without losing the chats the firm keeps for records.
- Module 5: For Northfield Consulting Group, a made-up 18-person firm, you send Claude one request with seven rules: the approved tools, what never goes in, what is fine to use, a person checks every output, be open with clients, who to ask, and a same-day, no-blame rule for mistakes.
- Module 6: You are Priya, office manager at Harbor & Pine Bookkeeping (made up). Jordan pasted Hale Dental Studio’s March payroll into ChatGPT and asked it to remember the client. You follow a one-hour incident log, sitting with Jordan at his screen.
- Module 7: You are Priya, office manager at Harbor & Pine Bookkeeping (made up). A vendor invoice arrives from Brightpath Supplies with a hidden line aimed at an AI. You find it, paste the email into a new Claude chat as plain text, and read what Claude does.
- Module 8: You are Priya, bookkeeper at Harbor & Pine Bookkeeping (made up). A vendor email with new bank details arrives, and a voicemail backs it up. You paste both into a new Claude chat as text and read the red flags.
- Module 1 · ChatGPT · Beginner · ~7 min
Use AI at Work Without Leaking Client Data using ChatGPT
You’ll be able to: Step 1, the two habits: turn off model training, strip a tenant's bank details and address out of a prompt, then catch the two promises ChatGPT made up before the reply goes out.
- Module 2 · ChatGPT · Beginner · ~7 min
Know What Never Goes into AI: a Four-Level Data Cheat Sheet using ChatGPT
You’ll be able to: Step 2, know what never goes in: sort office data into Public, Internal, Confidential and Restricted, take the Restricted part out by hand, then have ChatGPT redact a client email and catch the mix-up it makes.
- Module 3 · ChatGPT · Beginner · ~10 min
ChatGPT: Keep Client Data Safe, Set Up Your Team’s Data Controls in 10 Minutes
You’ll be able to: Step 3, the ten-minute setup: answer "do our files go into ChatGPT?" by setting the data controls, memory, Temporary Chat and shared links, then hand your team a one-page AI policy.
- Module 4 · ChatGPT · Beginner · ~10 min
Clear Client Data Out of ChatGPT Before It Becomes a Problem: Delete Chats, Clear Memory and Use Temporary Chat
You’ll be able to: Step 4, clean up after a paste: delete the one chat that has client details, archive what you keep, clear the memory it left, set safe defaults and start using Temporary Chat.
- Module 5 · Claude · Beginner · ~10 min
Roll Out a Team AI Policy in One Afternoon using Claude
You’ll be able to: Step 5, the policy in writing: have Claude write a one-page acceptable use policy from your rules, check the parts that matter, then get the announcement email and a five-question staff quiz.
- Module 6 · ChatGPT · Beginner · ~9 min
Someone Pasted Client Data into AI: Contain It in One Hour using ChatGPT
You’ll be able to: Step 6, the first hour after a slip: a new hire pasted a payroll export into ChatGPT. Delete the chat and its memory, turn off training, then write the incident note, the client email and the new policy rule.
- Module 7 · Claude · Beginner · ~6 min
Catch a Poisoned Email Before Your AI Assistant Obeys It using Claude
You’ll be able to: Step 7, the email that talks to your AI: spot the hidden line in a vendor email, test your assistant with a fake poisoned message, add a guard instruction and never let an AI approve a payment.
- Module 8 · Claude · Beginner · ~7 min
Stop a Fake Vendor Bank-Change Request Before You Pay using Claude
You’ll be able to: Step 8, the fake bank-change request: check an urgent "new bank details" email and a voicemail that sounds like your boss with Claude, apply three verify-before-you-pay rules, and write the bank-change policy and staff alert.
Prerequisites
- For the practice: a web browser. Each module runs in a copy of the real tool’s screens, so you need no account and no sign-up to start module 1.
- Helpful first: AI for Executive Assistants and Office Managers: Inbox, Meetings, Vendors and Documents, AI for Nonprofit Staff: Board Minutes, Donor Numbers, Communications and a Safe AI Policy.
- To do it for real: a ChatGPT account and a Claude account. Each module says which one it uses.
Time to complete
About 1 hour of hands-on practice across 8 modules (135 steps in all). Modules take 6 to 10 minutes each, so you can do one in a sitting and come back for the next.
Take it with you
- AI Governance for Small Business · The next layer: approvals, audit trails and testing before an AI agent touches customers.
- AI Security for Small Business · Connectors, agents and the scams that target people who use AI.
Questions about this course
Is it safe to use ChatGPT at work with client data?
It can be, with the right settings and rules. On a personal account with default settings, chats can be used to train models and memory keeps details across chats. This course shows how to turn training off, use Temporary Chat, set memory and shared links for a team, keep Restricted data (bank details, health, payroll, identity numbers) out entirely, and write the one-page policy that tells staff the same thing. For regulated data, check your contracts and your professional body's guidance as well.
Does ChatGPT train on my conversations?
By default on Free and Plus accounts, conversations can be used to improve models unless you turn that off in Data Controls or use Temporary Chat. Business and Enterprise plans do not train on your data by default. Module 3 walks through the settings on real screens; module 1 shows the single switch and why stripping details out of the prompt still matters.
Someone on my team already pasted client data into ChatGPT. What do I do?
Modules 4 and 6 cover exactly that: delete the chat, delete the memory it left behind, turn off training, archive what you keep, then write a short incident note, the client email if one is needed, and the rule that stops it happening again. Most of it takes under an hour.
Do I need ChatGPT Business or Enterprise to be safe?
Not to start. The habits in this course work on any plan: training off, Temporary Chat, Restricted data kept out, a written policy. A Business or Enterprise plan adds no training by default, admin controls and a data processing agreement, which matter once you handle regulated data or a client asks for one in writing.
Is there a free course on using AI safely at work?
Yes. This course is free: 8 hands-on modules, about 66 minutes in all, in practice copies of ChatGPT and Claude that check your work. No credit card and no sign-up to start, and a free certificate of completion for each module. It is written for small professional firms, not for IT departments.
How long does the course take?
The course takes about 66 minutes of hands-on practice across 8 modules of 6 to 10 minutes each. Most teams do it in one afternoon: the practice first, then the same settings and the policy in their own ChatGPT.
Want help doing this for real? Book a 1:1 session with Robert ($85/hour) →