AI Rules for Your Team
Goal. This template helps you write down how your business works with AI before someone pastes in something they shouldn’t. Your team knows which tools to use, what not to paste in and who checks the result. You have one document you can show a client who asks how you use AI.
In 15 minutes. (our estimate, not measured) To be honest: in 15 minutes you write a draft of the rules on your own, with no survey and no meeting. These are not yet the team’s rules: they become that once you go through them with people and collect signatures (the full version).
- (3 min) List the AI tools you use and the account for each: company, free personal or paid personal. Mark each ✅, ⚠️ or ⛔ (field 1).
- (4 min) Sort your data into colors, with 3 examples from your business. Not sure where something fits: put it in red (field 2).
- (3 min) List what a person always checks and sends, and who checks it (field 3).
- (2 min) Write whom you report an AI mistake or data in the wrong place to, and who approves connecting AI to email, the drive or systems (fields 5 and 6).
- (3 min) Write one line: “I decide that … Who … By when … I will check …”. For example: when you go through the rules with the team, and when the first review is.
Instructions (full version) (estimate: about 50 minutes, not counting the time for the team’s answers and the meeting; we have not measured this)
- Send your team five anonymous questions: which AI tools do you use at work, on what account (company, free personal, paid personal), what do you paste in, has anything from AI ever come out wrong, what would you show others. Add: “No one will face consequences for how they have used it so far.” With fewer than 6 people, skip the survey and just talk.
- List the tools from the answers and from the invoices. A person reads the terms of each plan on the provider’s website (whether the provider trains its model on your data, how long it keeps it) and writes down the link and the date.
- Sort your data into colors, with 3 examples from your business for each. If you don’t know where something fits, put it in red.
- Fill in the rest of the fields. Every rule ends with the word “because”.
- Go through the rules with your team in 30 to 45 minutes, collect signatures, enter the training in the record and set a review date.
Template fields
| Company | Version | Who is responsible for the rules | In effect from | Next review (in 3 months at the latest) |
|---|---|---|---|---|
1. Tools. For work, only tools marked ✅, on company accounts. A new tool: you ask [whom], and the answer comes within [2] business days. Because: a free personal account may keep what we paste in, and we can’t see that.
| Tool | Account | ✅ / ⚠️ only for… / ⛔ | Highest data color | Terms checked by (who, when, link) |
|---|
2. Data colors. Not sure: red. Removing names does not make data green. Because: the tool doesn’t know what is confidential. A person does.
| Green: any tool with ✅ | Yellow: tools with ✅, company account | Red: we don’t paste it into AI |
|---|---|---|
| 1. 2. 3. | 1. 2. 3. | personal data of clients and employees, contracts with a confidentiality clause, passwords, bank details, 1. 2. 3. |
3. Always a person. AI writes drafts. A person decides and sends: ☐ money ☐ contracts ☐ prices and offers ☐ decisions about people (AI does not assess people) ☐ [yours]. Because: whoever sends it answers for it.
| What leaves the company | Who checks it (can do it without AI) | What they check |
|---|
4. We say when AI helped: [when: results for a client, publications, AI talking to people on our behalf]. Because: trust breaks when it comes out later.
5. Mistakes. You report an AI mistake to [whom] within one business day, without looking for someone to blame. Red data in the wrong place: you report it right away. Review: 15 minutes, on [day] each month. Because: what we write down, we fix once.
6. Connections. No AI gets access to email, the drive, the bank or our systems, and none sends anything on its own, without written approval from [whom]. Because: broad access without a trace is a risk no one sees.
7. Changing the rules. At the review or after a second similar case, never after a single slip.
Training record (kept separately; only the responsible person has access)
| Date | Person | Rules version | What we covered | Passed / to repeat | Next refresher |
|---|
I have read these rules and follow them: [name, date]
Fictional example (the firm and the numbers are fictional, for this template)
A paint wholesaler, 14 people. Version 1. Responsible: Anna, the office manager. In effect from November 3, review on February 2. From the survey: 9 people use AI at work, 5 of them on personal accounts.
- Tools: an AI chat on a company plan ✅, yellow, terms checked by Anna on October 28 (link in the “AI” folder). The assistant in the company’s office suite ⚠️, green only, until Anna checks the terms. A free chat on a personal phone ⛔. A new tool: you ask Anna, answer within 2 days.
- Colors. Green: the price list from the website, manufacturers’ technical data sheets, product descriptions from the online store. Yellow: margins, the purchasing plan, a newsletter draft. Red: personal data, payroll, the contract with a retail chain that has a confidentiality clause, bank details, passwords.
- Always a person. Discounts outside the price list: Marek. Replies to complaints: Ewa. Contracts and decisions about people: the owner. Bank transfers: the accountant.
- Saying AI helped: in tender bids we add that part of the text was prepared by AI and checked by Ewa. Internal drafts: no label.
- Mistakes: to Anna within the day, review on the first Monday of the month.
- Connections: no AI has access to email or the warehouse system.
- Changing the rules: on February 2 or after a second similar case.
Record: November 6, a 45-minute training, 9 people, version 1, covered colors, tools and checking; 8 passed, 1 to repeat. Next refresher with version 2.
From us. The ideas come from books: Ethan Mollick (Co-Intelligence) says to always be the human in the loop and shows that people hide their AI use when they fear consequences; Elisa Farri and Gabriele Rosani (HBR Guide to Generative AI for Managers) warn that even seemingly anonymized data carries risk, because AI can re-identify people; Eric Lamarre, Kate Smaje and Rob Levin (Rewired) say to settle from day one who can access what, what needs a person’s approval and who is responsible. What we added is a one-page version for a small business: data colors with examples from your business, every rule ending in “because,” and a training record.
When to call us.
- The survey shows people use AI on personal accounts, and you don’t know what they paste into them.
- A client asks how you use AI, and you have no document to show.
- You can’t tell which data in your business is red, or no one has time to read the providers’ terms.
The first call (30 minutes) is free. We will see whether the template is enough or your team needs help putting the rules into practice.
Prompt for your AI. Paste it into your own ChatGPT, Claude or Gemini account; our own method, working from your company’s knowledge, is what we do in a Diagnosis.
You are a calm coach for the owner of a small business. Walk me through the
“AI Rules for Your Team” template: one question at a time, wait for my answer.
At the start, remind me: “Do not paste passwords, account numbers or personal data.”
Order: 1) the AI tools we use, the account for each, and a rating: ✅, ⚠️ or ⛔;
2) data in colors, three examples each from my business (unsure: red); 3) what a
person always checks and sends, and who; 4) whom we report an AI mistake to, and
who approves connecting AI to email, the drive and systems. For every rule ask:
“because…?”.
If an answer is vague, ask for a number or an example. Never invent facts about
my business or providers’ terms: tell me to read them on the provider’s site.
At the end, write one line: “I decide that … Who … By when … I will check …”.
The examples in the templates are fictional. The times given with the instructions are our estimates, not measurements.