Your employees' one-page AI guidelines should cover five things: what data never goes into AI tools, which tools are approved, that every AI output gets checked before it's used, when to say AI was involved, and who to tell when something goes wrong. Everything else belongs in your full policy. The one-pager exists because most people won't read the full policy at the moment they need it, but they will glance at a short list pinned in the team chat.
This post gives you a ready-to-adapt sheet of AI guidelines for employees, explains the reasoning behind each line so you can defend it, and covers how to tailor it by role and roll it out so it sticks. If you haven't written the full policy yet, start with our AI usage policy template; this sheet is the summary that sits on top of it.
Why have a one-page version at all?
A full AI usage policy does important work: it sets scope, assigns ownership, handles intellectual property, and records decisions. But it's written for the people who approve it, not for the person drafting a client email at 5:40 p.m.
A one-page sheet does a different job:
- It's findable at the moment of decision. Pinned in the team chat, printed by the desk, linked in the AI tool's sign-in page.
- It's memorable. Short rules, plain words, concrete examples.
- It reduces hidden use. When people know what's allowed, they stop using unapproved tools quietly. That quiet use, often called shadow AI, is where most of the real risk sits.
The rule of thumb: if a line on the sheet needs a paragraph of explanation, it belongs in the full policy, with a one-line summary on the sheet.
The one-page AI do's and don'ts sheet
Here's a version you can copy and adapt. Replace the bracketed parts with your own details.
| Topic | Do | Don't |
|---|---|---|
| Tools | Use [approved tools, e.g. the company workspace AI assistant] signed in with your work account | Use personal accounts or free AI tools for work tasks |
| Customer and personal data | Remove names, contact details and account numbers before pasting, unless the tool is approved for that data | Paste customer records, employee data, health or financial details into unapproved tools |
| Confidential business info | Ask [owner] before using AI with contracts, pricing, financials or unreleased plans | Upload confidential documents to tools that aren't approved for them |
| Checking outputs | Check facts, numbers, names, links and quotes against the source before using anything | Send, publish or act on AI output you haven't read and verified |
| Accountability | Treat AI output as your own work: you're responsible for what you send | Blame the AI if something you sent is wrong |
| Disclosure | Follow [rule, e.g. say when customers are talking to an AI assistant; flag AI-generated images] | Present AI-generated content as a person's first-hand experience or expert opinion |
| Decisions about people | Use AI to organize information; leave the decision to a person | Let AI decide on hiring, firing, pay, credit or discipline |
| Code and automation | Get review before AI-written code or automations touch live systems | Give an AI tool your passwords, API keys or admin access |
| Requests that seem off | Verify any urgent request for money or access through a known channel | Trust a voice, video or email just because it sounds like your boss |
| Mistakes | Report problems to [name or channel] straight away; no blame for reporting | Hide an AI mistake or try to quietly fix it alone |
Underneath the table, add three lines: who owns the guidelines, where the full policy lives, and the date it was last updated.
Why does each rule matter?
People follow rules they understand. Here's the reasoning to share when someone asks "why?"
Approved tools, work accounts. Business plans of AI tools usually come with better data terms (for example, inputs not used for model training) and admin controls. A personal account on the same tool may not have them. Work accounts also mean that when someone leaves, their AI history doesn't leave with them.
Data stays out unless approved. Once data is pasted into a tool, you depend on that vendor's retention, security and training practices. Data protection laws such as GDPR and India's DPDP Act also set rules on how personal data is processed and shared, and obligations depend on where you operate. Our guide to AI data privacy and security explains what's at stake.
Check every output. AI tools produce fluent, confident text that is sometimes wrong: invented figures, misremembered policies, citations that don't exist. That's the hallucination problem, and the only reliable defense is a person checking against the source.
You own what you send. This is the rule that changes behavior most. If the AI drafted it and you sent it, it's yours. Customers and regulators won't accept "the AI wrote that."
Disclosure. People generally expect to know when they're talking to a machine, and some jurisdictions are introducing rules on it. Equally, passing off AI-generated content as personal expertise damages trust when discovered.
People decisions stay with people. Hiring, pay and credit decisions carry legal and fairness risks, and AI tools can reflect biases in their training data. Use AI to summarize and organize; leave judgment to a person who can explain it.
Secrets stay secret. Credentials pasted into an AI chat can end up in logs. Keys belong in a secrets manager, configured by whoever manages your systems.
Verify unusual requests. AI makes impersonation cheap. A cloned voice or a perfectly written email asking for an urgent transfer is now realistic, which is why the sheet includes a line that isn't strictly about using AI.
Report, don't hide. You can only fix problems you hear about. Make reporting easy and safe.
How should the sheet differ by role?
One sheet for everyone is the right start. Then add a few role-specific lines where the stakes differ:
- Sales and customer support: never promise prices, discounts, refunds or delivery dates based on AI suggestions without checking; use approved reply templates.
- Marketing: check claims and statistics; respect image and music licensing; follow your disclosure rule for AI-generated visuals.
- Finance: no AI tool approves payments or changes bank details; double-check any AI-produced figures against the ledger.
- HR: no candidate or employee data in unapproved tools; AI never makes the hiring decision.
- Engineering and IT: review AI-written code like any other code; no secrets in prompts; follow your rules for AI agents that touch live systems.
Keep role additions to three to five lines each, or the sheet stops being one page.
How do you roll it out so people actually follow it?
A sheet that lives in a shared drive changes nothing. What works:
- Launch it in a 20-minute team session, not an email. Walk through real examples from your business: "Can I paste this customer complaint into the assistant to draft a reply?"
- Pin it where AI is used: team chat, intranet, the sign-in page of your approved AI tool.
- Make the approved route easy. If the approved tool is harder to use than a free one, people will drift back. Fix that before blaming anyone.
- Name a go-to person for "is this okay?" questions, and make sure they answer quickly.
- Review it every few months. Tools and risks change; so should the sheet.
- Praise good reports. The first time someone flags an AI mistake, thank them publicly.
Adoption is mostly a people problem, not a rules problem; our post on AI change management covers the wider picture.
What are the common gray areas?
Expect these questions and decide answers in advance:
- "Can I use AI to rewrite an email to a customer?" Usually yes in an approved tool, as long as personal details are removed or the tool is approved for them, and you check the result.
- "Can I record and transcribe a meeting with AI?" Only with participants' knowledge and consent, and in line with local law. Make the rule explicit.
- "Can I use AI on my phone for work?" Only approved apps, signed in with your work account.
- "A client asked if we use AI. What do I say?" Have a short, honest standard answer ready.
Add answers to the most common questions as a short FAQ beneath the sheet, or in the full policy.
Frequently Asked Questions
Is a one-page AI guideline enough on its own?
For a very small team it can be a reasonable start, but most businesses should also have a fuller policy covering ownership, approved tools, data categories, intellectual property and review. The one-pager is the summary people use day to day; the policy is the reference behind it.
Should we ban AI tools instead?
Bans tend to push usage out of sight rather than stop it, and they give up real productivity gains. Approving a small set of tools with clear rules usually reduces risk more than a ban, because you can see and manage how AI is used.
Do employees need to tell customers they used AI?
It depends on the context and your jurisdiction. Customers should generally know when they're talking to an AI assistant, and AI-generated content shouldn't be presented as first-hand experience. Staff-reviewed drafts often don't need a label, but set a clear rule so people don't guess. This isn't legal advice; check rules that apply to you.
How often should the AI do's and don'ts be updated?
Review it every three to six months, and whenever you approve a new tool, change vendors or have an incident. Date each version so people know they're reading the current one.
Who should own the AI guidelines in a small business?
One named person, often the founder, operations lead or IT manager, who approves tools, answers questions and updates the sheet. Ownership matters more than seniority; the owner needs to respond quickly when people ask.
Want help turning AI rules into everyday practice?
Rules work best when the approved way is also the easiest way. We set up approved AI tools, private assistants and automations with the right guardrails built in, as part of our AI automation work. Talk to us about how your team uses AI and we'll help you make the safe route the default.