Nobody decided it, but it already happened. Sales drafts replies in ChatGPT, someone in finance summarises contracts with it, the office manager translates supplier emails, and a developer pastes code in to find a bug. Employees using ChatGPT at work is now normal in most small teams, usually on personal accounts, and usually without anyone asking what goes into it.

The two usual reactions are to ban it or to ignore it. A ban pushes the use onto personal phones, where you see nothing. Ignoring it means customer data, prices and internal documents sit in accounts the company does not own. There is a middle path, and the first half of it is simply finding out what is happening, which takes an evening.

Why employees using ChatGPT without rules becomes a problem

Company data ends up in personal accounts. When someone pastes a customer email, a contract or a price list into a personal account, that text now lives under that person’s login and under the provider’s consumer terms, not under any agreement your company signed. The sign: nobody can say which accounts are in use or who pays for them.

Nobody knows what is safe to share. Without a rule, each person draws their own line. One never pastes names, another pastes whole customer files. The sign: when you ask two people what they would never put into ChatGPT, you get two different answers.

Answers go to customers unchecked. The assistant writes fluent, confident text, including when it is wrong about a price, a delivery term or a legal point. If a reply goes out without a human reading it with the facts in hand, the mistake goes out under your name. The sign: customer emails that sound unlike the person who sent them.

The knowledge leaves with the person. Prompts that work, templates, useful chats all sit in personal histories. When the employee leaves, the account and everything in it leaves too. The sign: the best practices in the team are known only to the one or two people who found them.

Every tool is chosen separately. One person uses ChatGPT, another Gemini, a third a browser extension that reads every page. Each tool has its own terms and its own access to your data. The sign: nobody has a list.

How to tell which one is yours

This is a conversation, not an investigation. People are usually glad to talk about how they use these tools once it is clear nobody will be punished for it.

  1. Ask the team which AI tools they use and for what. Make it a short, anonymous-friendly list: tool, task, personal or company account. If the list surprises you, the first cause is yours.
  2. Check expense claims and card statements. Personal subscriptions paid back through expenses show which tools are really in use.
  3. Ask what they paste in. Customer names and emails, contracts, prices, financial figures, code, internal plans. Anything personal or confidential in a personal account is the most urgent point.
  4. Pick five recent customer-facing texts and ask how they were written. If any went out drafted by an assistant and unchecked, the review step is missing.
  5. Look for browser extensions and plugins. Ask people to open their extension list. An assistant that can read every page in the browser can also read your CRM and mailbox in that browser.
  6. Check whether anything is written down. A policy, a message in the team chat, a line in the handbook. If nothing exists, every other problem follows from that.

The fix, in order

Cheapest first. The first three steps cost nothing and change the risk the same week.

  1. Write a one-page AI usage policy for employees. Approved tools, which accounts, what never goes in (personal data of customers and staff, passwords and access keys, confidential contracts and figures), which outputs need a human check before they reach a customer, and who to ask. Plain language, one page, pinned where people see it.
  2. Say clearly that using AI is allowed. The goal is to move the use into the open, not to catch anyone. People who are afraid of being caught hide their use, and hidden use is the risk.
  3. Stop the most sensitive pasting now. Before any purchase, agree that customer personal data and confidential documents stay out of personal accounts from today. This one rule removes most of the exposure.
  4. Move to business accounts. Choose one or two tools and buy them under the company, with an administrator who adds and removes people. Read the provider’s data-use terms for the business plan and check them against your own data protection obligations before letting sensitive data in. Choosing tools, setting access and writing these rules together is how our AI implementation projects usually start.
  5. Keep a shared library of prompts and templates. The instructions that work for replies, summaries and translations go into a shared folder, so the knowledge stays with the company and quality evens out across the team.
  6. Connect AI to company data only through controlled integrations. When the team wants the assistant to use customer history or deal data, connect it through the system that already holds permissions, not by copying exports into chats. In our CRM and analytics projects that means the assistant sees what the user is allowed to see, and nothing more. The same caution applies when an assistant answers customers directly; we covered that in whether an AI chatbot should answer your customers.
  7. Review the policy every few months. Tools and terms change often. Put a date in the calendar to update the list of approved tools and the rules.

What to measure

  • Share of AI use on company accounts rather than personal ones, from the team list, reviewed monthly until personal use disappears.
  • Customer-facing texts checked by a person before sending, spot-checked monthly.
  • Number of different AI tools in use, which should shrink to the approved list.
  • Questions people bring to the policy owner, as a sign the policy is read and trusted; silence usually means it is ignored.

Where we come in

An owner with an evening and an honest conversation can find out how the team uses AI and write the first page of rules. What we bring is the setup behind it: the choice of tools that fit your data, business accounts with proper access, a policy that people follow because it matches their work, and integrations that connect AI to company data without copying it around. If you would rather start with an outside view, a short brief with the tools your team uses today is enough.

Frequently asked questions

Should we ban ChatGPT at work?
A ban rarely stops the use; it moves it to personal phones where you see even less. It is usually safer to allow approved tools under business accounts, with a short written rule about what may and may not be pasted into them.
What should an AI usage policy for employees include?
Which tools are approved and under which accounts, what information must never be entered, which tasks need a human check before anything goes to a customer, who to ask when unsure, and who owns the policy. It should fit on one page so people actually read it.
Is it safe to paste customer data into ChatGPT?
Not into a personal account with no agreement between your company and the provider. With a business plan, read the provider's data-use terms and your own data protection obligations first, and keep personal and confidential data out unless the setup has been checked for exactly that use.
What is the difference between a personal and a business ChatGPT account?
A business account belongs to the company: an administrator adds and removes people, the provider's business terms apply to the data, and the history stays with the company when someone leaves. A personal account belongs to the employee, along with everything they have typed into it.

← All articles