Short answer: yes, your employees can use ChatGPT at work. And honestly, some of them already are, whether you know it or not.
That's not a disaster. AI is great at rewording an awkward email, summarizing a long PDF or turning bullet points into a decent paragraph. The problem is what people paste into it. A customer's full name and phone number. A patient's appointment details. Your pricing spreadsheet. The password someone was trying to "fix the formatting" on.
Nobody does that to be reckless. They do it because nobody ever told them where the line is.
That's what this article fixes. One page, three rules, and about ten minutes to roll it out to your team.
The three rules (this is the whole policy)
Print these big. Most people will only ever remember this part.
Rule 1: Go ahead
Use AI for wording, ideas and summaries of things that are already public or contain nothing private. Rewriting an email, brainstorming a sign for the window, summarizing a vendor's product sheet, drafting a social post.
Rule 2: Everyday customer messages, stripped, in an approved account only
A routine question like "what are your hours" or "can I return this" can go into your company's approved AI account once you remove anything that points to a real person: name, phone, email, address, order or account numbers. Swap them for [Customer]. If you can't strip it and still get help, don't paste it. And this only applies to ordinary messages. Anything covered by Rule 3 stays out, stripped or not.
Rule 3: Never, unless the tool is vetted and under contract for that data
Protected data never goes into an AI tool. That means health information about anyone, Social Security numbers, credit card and bank numbers, legal case details, employee HR and medical records, passwords and anything a law or contract says you must protect.
The only exception: a tool your business has specifically vetted and signed a contract for that covers that exact kind of data. For health information, that means a signed Business Associate Agreement (BAA) with the AI vendor, for the specific product your staff use. No BAA, no health data. No exceptions.
And removing the name doesn't make protected data safe. A patient's appointment details with the name cut out are still health information.
That's it. If your team follows those three, you've covered most of the risk.
Why public AI isn't private
This is the part most owners don't know, so here it is straight from the companies themselves:
Your chats can train the AI. OpenAI says that for its individual plans (Free, Plus and Pro), it may use your content to train its models unless you turn that off in settings (OpenAI). Google says Gemini activity is used to train its models when activity saving is on (Google). Anthropic changed its consumer terms in 2025 so that Claude chats are used for training unless the user opts out (Anthropic).
People may read them. Google's own Gemini privacy notice says human reviewers read some conversations and tells users not to enter anything confidential they wouldn't want a reviewer to see (Google).
There's no expectation of privacy. OpenAI's CEO said publicly in July 2025 that ChatGPT conversations don't have the legal confidentiality you'd get with a doctor, lawyer or therapist (TechCrunch).
A paid plan isn't automatically enough. OpenAI offers a BAA only for its API and for ChatGPT Enterprise and Edu accounts set up through its sales team, and says it does not offer one for ChatGPT Business (OpenAI). Federal health regulators treat cloud providers that store patient data as business associates who need a BAA (HHS), and an AI vendor holding your chats is in the same spot.
So the safe default: treat any public AI tool like a stranger's notebook. If you wouldn't write it there, don't type it there.
My rule of thumb is simple: if I'm responsible for protecting the information, I need to understand where it's going before I put it into another tool. Getting a better draft isn't worth losing control of someone else's information. When I'm unsure, I leave the sensitive details out and ask first.
At other workplaces, we ended up blocking AI tools on the network because of concerns about sensitive information leaving the business. A clear policy gives employees guidance before you reach that point. Technical controls may still have a place, but people also need to understand what they can use, what they can share and when to ask.
What to put in writing
The three rules are the poster. The one page policy adds a few lines that protect you if something ever goes wrong:
Which tools are approved, and for what. Name them. "Company ChatGPT account, for wording and ideas only" is better than "any AI." Free personal accounts are never approved for business data. If any tool is approved for protected data, name it, name the signed agreement that covers it (such as a BAA) and limit it to the people who need it.
Who to ask. One name. When someone isn't sure, they ask before they paste.
A human checks everything. AI writes drafts. A person reads every word before it goes to a customer, because AI will confidently make things up.
What to do if someone slips. Tell the owner right away, no punishment for reporting it. You want people to own up, not hide it.
The date and a signature. Everyone signs that they read it. Review it once a year.
The free template below has all of that filled in. You change the business name and the approved tools, and you're done.
How to roll it out in 10 minutes
Don't email it and hope. Do this at your next team huddle:
Hand out the one page and read the three rules out loud (2 minutes).
Give one real example of each rule from your own business (3 minutes). “Rewriting our holiday hours post: fine. A customer complaint email: strip the name first. Their card number: never.”
Ask, “Who's already been using AI for work?” Make it safe to raise a hand. You'll learn a lot (3 minutes).
Everyone signs the sheet. Pin the poster where people will see it (2 minutes).
Other places i have worked we ended up banning all AI tools at the network level to prevent anyone using them and leaking sensitive info, with this you wont need to apply a technical solution to an administrative problem.
Safe or not?
Wording, ideas and public information in an approved business account: safe.
Free personal AI accounts for any business data: no. They're built for individuals, can use your chats to train models by default and come with no privacy promise to your business.
Protected data (health, financial, legal, HR, IDs, passwords) in any AI tool: never, unless your business vetted that exact tool and signed a contract covering that data, like a BAA for health information. A paid business plan alone doesn't count, check whether the vendor actually signs one for that product.
Customer data that's already been pasted somewhere: don't panic, do act. Delete the conversation, turn off chat history and model training for that account, and if it involved health, financial or legal records, talk to whoever handles your compliance (or your lawyer) right away about whether it has to be reported.
5 minute win
Ask your team one question today: “What AI tools are you using, even once?” Write down every answer. That list is the first line of your policy.