The upside is real, so a ban will fail
Your staff are already using AI chatbots. They draft emails with them, summarize documents, rewrite awkward paragraphs, and get unstuck on spreadsheets. Used well, these tools genuinely save hours a week, and I use them in my own business.
Which is why an outright ban is the wrong move. Ban the tools and people will quietly use them anyway, on personal accounts, on personal phones, where you have zero visibility. The realistic goal is not to stop AI use. It is to draw a clear line around what company information is allowed to go into it.
Where what you paste actually goes
When someone pastes text into a chatbot, that text leaves your business and lands on another company's servers. Depending on the account type and settings, it may be stored, reviewed, or used to train future versions of the tool. Training means the text can influence what the model learns, which is not something you can undo later.
On free consumer accounts, the default settings are usually the most permissive ones, and the account belongs to your employee, not to you. If that person leaves, their chat history, including whatever company information ended up in it, walks out the door with them.
None of this makes the tools evil. It makes them like any outside service: fine for some information, wrong for other information.
What should never go into a chatbot
Here is the list I give clients. Do not paste:
- Customer names, phone numbers, addresses, or anything identifying a specific client
- Card numbers, bank details, or payroll figures
- Passwords, API keys, or anything that looks like a credential
- Health information of any kind
- Contracts, bids, or documents covered by a confidentiality agreement
- Employee records, complaints, or disciplinary notes
The simple test I teach: if you would not email it to a stranger, do not paste it into a chatbot. Staff can still use AI on all of this work by stripping the specifics first. Rewrite this email works just as well when the customer's name says a customer.
A one-page acceptable-use policy
You do not need a lawyer or a twenty-page document. One page covers it.
- Approved tools: name the two or three AI tools the business allows, and on which accounts.
- The never-paste list from above, printed in full.
- The stranger test as the tiebreaker for anything not on the list.
- A rule that AI output going to a customer gets read and corrected by a human first, because these tools state wrong things confidently.
- Who to ask when unsure, with no penalty for asking.
Walk the team through it in one fifteen-minute meeting. The goal is not compliance theater. It is making sure the person at the front desk knows the difference between summarize this public article and summarize this client file.
Business accounts versus consumer accounts
If the team uses these tools weekly, pay for business-tier accounts. The major AI vendors' business plans typically commit in writing not to train on your data, and they give you an admin view of who is on the account. Expect to pay in the neighborhood of 20 to 30 dollars per user per month.
That also solves the offboarding problem: when someone leaves, you disable their seat and the history stays with the business. A consumer account can never give you that.
One more quiet risk worth naming: free AI browser extensions and note-taking bots that join your meetings. Treat every one of them as a vendor that can read whatever it touches, and approve them the same way you would approve giving a stranger a key.
Next steps this week
- Ask the team, without any blame attached, which AI tools they already use and for what. You will learn more in ten minutes than any audit would find.
- Write the one-page policy using the structure above and go over it in your next staff meeting.
- If usage is real and regular, move those people onto business accounts and turn off training in the settings.
- Revisit the list twice a year, because the tools change fast.
Handled this way, AI is like any other power tool in the shop: genuinely useful, and safe exactly as long as everyone knows where the guard rails are.