Write AI usage guidelines for your team with ChatGPT

Most teams already use AI at work, on a mix of personal free accounts, personal paid accounts and maybe a corporate one. A one-page guideline turns that from a quiet risk into a set of habits. ChatGPT drafts the page in about 20 minutes, and the value depends on one thing: examples from your own team's work. A policy that says "do not share confidential information" changes nothing; one that says "do not paste the payroll export, paste the column headers and three rows with invented names" does.

Start with an honest inventory

Ask the team which AI services they use and on which accounts. Expect personal free accounts; that is where the risk sits, since chats there may be used for training by default. Then list the data your team handles and mark what is especially sensitive: salary data, client contracts, personal data of employees, unreleased financials. Keep the examples specific to your function. A sales team's sensitive data looks different from an accounting team's.

The prompt

Role: you are a manager writing AI usage rules for your team in plain language. Context: the team is {{department and what it does}}, it uses {{list of services and account types}}. Data the team works with: {{list of data types}}. Especially sensitive: {{what exactly}}. Task: write a one-page guide: 1) what can be given to AI, 2) what is forbidden, with three examples from our work, 3) how to anonymize data before pasting, with a before and after example, 4) how to turn off training on our data in our services, 5) whom to ask when in doubt, 6) what to do if sensitive data has already gone into a service. Format: up to 350 words, short sentences, no long dashes, no generic phrases. Every point starts with an action.

Point six is the one teams forget. People will make mistakes, and a guide that tells them what to do next (delete the chat, tell a named person) gets mistakes reported instead of hidden.

Push back on the first draft

The first draft usually has generic phrases about confidentiality and responsible use. Send it back with a specific request: "replace each prohibition with an example from an accounting department", or whichever function is yours. Check the anonymization example: it should show a real-looking line before and after, such as a payment reference with a client name turned into [Client 1].

Then verify the settings instructions yourself in the live interface. For ChatGPT on free and personal paid plans, the setting is in Settings, Data controls: Improve the model for everyone. Menus move and get renamed, so an instruction written from memory, by you or by the model, may already be wrong.

What to do

1. List the AI services the team uses and the account type for each.
2. List the data that must never go into public services, with examples from your work.
3. Run the prompt with your department, services and sensitive data types.
4. Replace generic prohibitions with examples from your function.
5. Check every settings instruction in the live interface of each service.
6. Approve the guide, send it out and include it in every new hire's onboarding.

Signs the guide works

It fits on one page, every prohibition has a real example, people know whom to ask, and the settings steps are verified. The best sign comes later: someone asks "can I paste this?" before pasting it.

Corporate accounts

If the guide shows that most people work on personal accounts with sensitive data, a team plan may be cheaper than the risk. ChatGPT Business costs 25 USD per seat with annual billing (30 monthly) and keeps team data out of training by default, with admin controls. Plus is 20 USD per month for individuals. See ChatGPT pricing. HR managers can use the version on the AI guidelines page for HR managers, and if the team also uses Claude, compare the two in ChatGPT vs Claude.