Guides · AI at work
What Not to Share With ChatGPT, and a One-Page Rule for Your Team
By Josh Croom · October 27, 2026 · 11 min read
Never share a name next to anything private, and keep out health details, passwords, card or bank numbers, door and gate codes, case notes and anything someone told you in private. Write those down as your team's never list, use one work account with training turned off, and swap real names for codes before you paste.
The email thread that went in whole
Tuesday, 4:50 PM. An office manager has one tricky reply to write before she heads home. A customer is upset about a bill, the thread runs long, and she copies the whole thing into ChatGPT with one line on top asking for a calm reply.
The reply comes back calm and clear, and she sends it. But the chat also took in the customer's full name, home address and cell number, the last four digits of a card, and the signature of everyone else on the thread. A good reply didn't need any of it. A short summary in her own words, with the customer written as the customer, would have gotten the same draft and kept every one of those details out.
Why a written rule matters
Two surveys from 2024 found that most people who use AI at work pick their own tools, and that even people whose job is guarding data have pasted private company facts into AI.
- 78%use AI tools they found on their ownof people who use AI at work, Microsoft and LinkedIn
- 48%put private company facts into AI toolsof privacy and security workers, Cisco
Without a written rule, each person on your team decides alone what's safe to paste. If people who guard data for a living paste private facts into AI, a busy front desk can too. A written rule and one work account give everyone the same line to stay behind.
The never list
Tick each item your team's rule already covers, and put anything left unticked on your one-page rule this week. Your ticks stay on this device.
0 of 9 done
Why each item is on the list
Each item on the never list could hurt a real person if it ended up somewhere you didn't plan, and once it's pasted, it sits in a tool you don't control.
A name turns a detail into a person. A name by itself is fine, but a name beside a bill, a complaint or a diagnosis points to one person. Write Person A or the caller instead, and the AI does the same job.
Health details can point to someone without a name. Add a town, an age or a date to a diagnosis, and someone who knows the area can often tell who it is. Leave health details out entirely.
No draft needs a password or a card number. Login codes and bank numbers don't help a draft either. If a password ever lands in a chat by mistake, change it that same day.
Codes open real doors. Anyone who reads a gate, alarm or door code can get into a home or office where people live or work. Keep codes on the job ticket and out of every chat.
Someone trusted you with case notes and private messages. They gave those to you and not to a software company, and the same goes for intake forms. Summarize what you need in your own words, with the names taken out.
Pay and reviews stay between two people. A leaked pay rate or job review can cost you a good employee's trust.
Swap names for codes before you paste
Both versions describe the same problem. The swapped one gives the AI everything it needs for a reply, and nothing that points to the customer.
Pasted as is
- Jane Carter at 412 Oak Street called about order 55812.
- Her card ending in 4471 was charged twice.
- She wants a call back at her cell, 555-0142.
Swapped for codes
- Customer A called about a recent order.
- Her card was charged twice for it.
- She wants a call back today.
Seven steps to a one-page AI rule
You can write the rule in one sitting, and teaching it takes one team meeting, which is the part that makes it stick.
Step 1
Write the never list
Start with the list above, add anything special to your work, and fit it on one page. A rule that runs longer than a page tends to sit in a drawer.
Step 2
Pick one tool and one work account
Choose one AI tool and sign in with a work email, so the chats stay with your team and not with one person. When someone leaves, the account and its history stay with you.
Step 3
Turn off training
In the tool's settings, turn off any switch that lets it learn from your chats. Check again after the tool updates, because settings can move or change names.
Step 4
Swap names for codes
Write Person A or the caller instead of a name, and leave out birthdays, addresses and account numbers. The AI writes just as good a draft without them.
Step 5
Use the stoplight test
Sort anything you plan to paste into green, yellow or red before it goes into a tool. Print the three colors and what each one covers, and keep the page beside every screen where your team uses AI.
Step 6
Name one person to ask
Name one person who owns the rule, and have everyone ask them before pasting anything they're unsure about. A quick question costs far less than a leak, and it shows the whole team where the line sits.
Step 7
Teach it on day one
Read the rule at a team meeting, and hand it to every new hire and volunteer on their first day. A rule that only lives in a shared folder rarely gets read.
The stoplight test
Run anything through three colors before it goes into an AI tool. If you can't tell which color it is, treat it as red.
| Color | What it covers | Where it can go |
|---|---|---|
| Green | Anything you'd post on your website, like your hours, your services and your public prices | Any AI tool |
| Yellow | Inside work with no one's private details, like a draft policy or meeting notes with the names taken out | Your work account only |
| Red | Everything on the never list | Never into an AI tool |
ChatGPT settings to check today
Check these four things in ChatGPT before your team uses it for work, and check them again after a big update.
Turn off model training. OpenAI says it may use content from its plans for individuals to train its models. On one of those accounts, open Settings, then Data controls, and turn off Improve the model for everyone. When you're signed in, that switch covers your account on the web and on your phone. OpenAI says it doesn't use content from ChatGPT Business, Enterprise or Edu workspaces to train its models by default. A work email on a personal plan doesn't get that default, so turn the switch off there too.
A temporary chat isn't gone right away. It stays out of your history and isn't used to improve OpenAI's models. OpenAI says it may still keep a copy for up to 30 days for safety, so the never list applies here too.
Memory carries details between chats. A detail you pasted last week can show up in a reply today. You can view, delete or turn off saved memories under Settings, then Personalization.
A shared link opens for anyone who has it. Read the whole chat for names and numbers before you share it, and delete old links you no longer need.
A second check before you send a prompt
Put this line above any yellow prompt that includes a message from someone else, after you've read it yourself. The AI still sees whatever you paste, so this check is a second net and never the first.
Before you answer, tell me if anything I pasted looks private, like a name, a phone number, a health detail or a money detail. If it does, stop and list it. (paste)
Before I start, I found two private details: a full name in the first line and a phone number in the signature. Do you want to take them out and paste it again?
Yes. Here it is with the name and number taken out.
Mistakes that slip past careful people
These four habits feel safe, and they let private details into a chat without anyone noticing.
- Thinking a first name alone makes a story safe, when the details around it, like a street, a job or a diagnosis, can still point to one person.
- Pasting a whole email thread, which can carry names, phone numbers and signatures you never looked at.
- Letting each person use a personal account for work, because the chats leave with them and you can't see what went in.
- Trusting a setting you checked once, even though tools update often and switches can move.
Add the details your kind of work handles
The never list covers every team, and each kind of business adds a few items of its own.
An online store should add order numbers, customer email addresses and anything from the payment screen, and our guide to AI for ecommerce shows how to draft product pages and support replies without them. A trade business should add gate codes and home addresses, which our guide to AI for contractors covers along with quotes and job write-ups. A nonprofit or church should add donor gifts, prayer requests and anything from a counseling or care conversation.
Does your team have an AI rule yet?
Answer five quick questions. Your answers stay on your screen.
Is your never list written down where everyone can find it?
Does everyone use one AI tool, signed in with a work account?
Is the training switch turned off in that tool?
Is there one named person who answers questions about the rule?
Do new hires and volunteers get the rule on their first day?
Your team has a real rule, and the job now is keeping it current. Check the training switch again after the next big update to your AI tool, and reread the never list together once a year.
You're partway there, so fix the first question you answered No this week. Each one closes a gap where a private detail could slip into a chat.
Your first step is the never list. Write it on one page tomorrow and read it at your next team meeting, and the other steps get easier from there.
How we use AI at Emprise
Our in-house AI helps with research, drafts and tracking checks, and a person checks every setup before it goes live. A person also approves any change to your budget or bid strategy. We run 64+ active Google Ads accounts for businesses and nonprofits from Conway, Arkansas, and you can see how that work runs on our Google Ads management page.
Talk it through with Miguel
If you'd like to talk through where AI fits in your marketing, or how your ads handle the calls and forms they bring in, book a call with Miguel. He'll ask how your team works today and where AI already shows up in it.
Questions people ask
What should you not share with ChatGPT?
Don't share a name next to private details, and leave out health information, passwords, card and bank numbers and Social Security numbers. Door codes, case notes, private messages and staff pay and reviews stay out too. Swap names for codes like Person A, and summarize in your own words instead of pasting whole threads.
Does ChatGPT use my chats to train its models?
On a personal account it can, unless you turn off Improve the model for everyone under Settings, then Data controls. OpenAI says it doesn't use content from ChatGPT Business, Enterprise or Edu workspaces to train its models by default.
Is a temporary chat in ChatGPT private?
A temporary chat stays out of your history and isn't used to improve OpenAI's models, but OpenAI says it may keep a copy for up to 30 days for safety. Treat it like any other chat and keep the never list out.
Is it safe to use ChatGPT for work?
It can be, with a written rule: one work account, training turned off, names swapped for codes and nothing from the never list. A person should still check every fact and read every message before it goes out.
What should I do if I already pasted something private?
Delete the chat, change any password or code that went in, and tell the person who owns your AI rule. Then add that item to your never list, so the next person on your team knows it's off limits.
See how we run Google Ads for small businesses and nonprofits
Read next
- Call Log Template: What to Write Down From Every CallA ready-to-copy call log, the one question to ask every caller, and how to sort wrong numbers.
- ChatGPT Custom Instructions Examples: Make Each Fix Only OnceEight custom instructions examples to copy, and a four-line card that carries your fixes into every chat.
- ChatGPT Prompt Templates: Write the Weekly Ask Once and Save ItCopy-ready ChatGPT prompt templates for the jobs you run every week, and how to save each one.