# [BUSINESS NAME] — Employee AI use (one page)

**Last updated:** ________________  
**Policy owner:** [name] · [phone] · [email]  
**Related lesson:** A One-Page Employee ChatGPT / AI Policy You Can Adopt  
**Draft mill (optional):** `/tools/policy-generator` — generate, then cut. Do not publish the long draft as the shop rule.

This page is an internal rule. It is not legal advice. Have counsel review it before you treat it as a legal document, and before regulated work (health, children, finance, government).

---

## What this is

We use AI as a **draft tool**. We do not use it as a **source of facts**. The person who sends the message owns the names, numbers, hours, and promises.

---

## Traffic lights

**GREEN — paste freely**

- Public web copy and blog drafts
- Our published price list
- Grammar and rewrite of text that is already allowed to be public
- Brainstorm lists with no customer names

**YELLOW — company account only, strip names first**

- Internal process notes that are not a secret recipe
- Anonymized ticket themes (“three people asked about gate codes”)
- Drafts of public emails before they are sent

**RED — never paste**

- Customer names, phones, emails, addresses, gate codes
- Invoices, bank exports, payroll, IDs
- Passwords, API keys, alarm codes
- Medical, school, or children’s data
- Job photos or videos that show an address or a face we have not permissioned
- **Our extra red line:** ________________________________

If you are not sure, it is red until [policy owner] says otherwise.

---

## Approved tools (company login only)

Write the tools you actually have logged in. Do not list a plan you intend to buy.

As of **August 2026**, do **not** mark these as current: GPT-4o, Claude 3 Opus, Midjourney V6.

| Tool people say | Name we put on this page (current as of Aug 2026) | Rule |
|---|---|---|
| ChatGPT | ChatGPT (GPT-5.6; Terra is enough for most drafts) | Company login. Personal free tabs are not approved for yellow or red. |
| Claude | Claude (Sonnet 5 or Opus 5) | Company login. Same rule. |
| Images | Midjourney V7 or the image tool we already pay for | Never publish as a photo of real staff or jobs. |
| Other | [name] | Ask [policy owner] before any business text. |

A tool that is not in this table needs a yes from the policy owner before any business text goes in.

---

## Review rule

AI output is a first draft. Check numbers, names, prices, hours, legal claims, and anything a customer will act on against a source we already have (price list, calendar, signed quote). The model’s confidence does not count as a source.

Customer-facing bots and auto-replies must say they are not a person, and must hand off when they do not have an approved answer. See the FAQ seed if we run a site bot.

---

## If you pasted something you regret

Email [policy owner email] the same day. Say what tool and what kind of data (you do not need to paste the data again). Honest reports are not punished. Hiding them is.

---

## Adoption

This page is read out loud when it changes. We revisit it every 90 days or when we add a tool. If it no longer fits on one printed page in 11-point type, we cut it — we do not add a second page.

**Acknowledged by (optional sign-on):**

Name __________________ Date __________