Here's a scene that plays out in plenty of small offices: a bookkeeper at an accounting firm gets a client's bank statement as a scanned PDF and needs the transactions in a spreadsheet. She pastes the whole statement, account number included, into a free chatbot on her personal account. It works beautifully. Nobody told her not to, and nobody at the firm knows it happened.
She wasn't careless; she was resourceful. That's the case for a written policy: your team is probably already using AI, and without a page that says what's fine and what isn't, each person draws the line somewhere different. The goal isn't to stop them; it's to make the safe way the easy way.
Why one page beats a handbook chapter
Long AI policies tend to go unread, and policies that ban everything push AI use onto personal phones where you can't see it. A single page that names real tools and real examples gets read, remembered and followed. Keep a longer document for your IT provider or security plan if you need one. The page is for the people doing the work.
The one-page template
Copy the seven sections below, replace the bracketed parts and delete anything that doesn't apply to your business.
1. Why we have this policy
“AI tools can save us time on drafting, summarizing and formatting. We use them where they help, we protect the information people trust us with, and a person is responsible for everything we send.”
2. Approved tools
| Tool | Approved for | Not approved for |
|---|---|---|
| [Business AI account with a no-training agreement] | Drafting emails, summarizing our own documents, reformatting data | Anything outside what our agreement with the vendor covers |
| [AI features inside our practice, tax or accounting software] | Work inside that software, under its existing access rules | Copying data out into other tools |
| Free or personal AI accounts | General questions with no client or business details | Any client, patient, employee or financial information |
3. What never goes into an unapproved tool
- Client, patient or customer names alongside anything about their matter, health, finances or account
- Social Security numbers, tax return details, bank and card numbers
- Passwords, access codes or details of our internal systems
- Employee records and pay
- Anything a client gave us in confidence or under an agreement
- Live client or patient calls and meetings: no AI note-taker or recorder joins unless the tool is approved and everyone on the call has been told. Recording-consent rules vary by state, and some, including California, require every party's consent.
4. Review before it leaves, and decisions AI doesn't make
“A person reads and approves anything AI helped write before it goes to a client, patient, court, agency or the public. Check names, numbers, dates and every citation or quote against the original. If you can't verify it, take it out. AI never makes hiring, firing, pay or discipline decisions; a manager does.”
5. Telling customers
“Whenever a customer is dealing with an automated assistant, we say so, and we don't pass off AI-written content as someone's personal work when that would mislead.” Add any disclosure your profession or state requires.
6. Who to ask, and what to do when something goes wrong
“Questions go to [name]. If client information went into the wrong tool, or an AI error reached a client, tell [name] the same day. Reporting quickly is what matters, and nobody is in trouble for reporting.”
7. Review date
“We review this page every [quarter]. Last updated [date]. Signed by [owner].”
Flowchart
“Can I use AI for this?” A one-minute check for staff
- StartYou want AI help with drafting, summarizing or reformatting
- DecisionDoes it include client, patient, employee or financial details?YesA personSwitch to an approved tool cleared for that data, or ask [name]; then check and approve as belowNo
- StepUse an approved tool and leave identifying details out of the prompt
- StepRead the result and check every name, number, date and citation
- DecisionWill it go to a client, patient, agency or the public?NoStepFine for internal use; save it where the team can find itYes
- A personThe responsible person approves it before it's sent
- DoneSent, with the approver's name on record
Before and after
No policy vs. one page
No policy
- 01Each person picks an AI tool, often a personal account
- 02Client data goes wherever is fastest
- 03AI drafts go out after a quick skim, or none
- 04Mistakes stay hidden because nobody knows the rules
ResultReal risk the owner can't see
One-page policy
- 01Firm pays for one business account with a no-training agreement
- 02Statements are converted inside approved software
- 03Drafts are checked against the source before sending
- 04Mistakes are reported the same day and fixed
ResultThe same time saved, and the owner knows where data goes
Add-ons for regulated professions
Some professions carry rules that deserve a line or two of their own on the page. Nothing below is legal advice, and rules differ by state and by profession, so confirm the specifics with your own advisers or professional body.
- Law firms. ABA Formal Opinion 512 treats generative AI under the rules lawyers already follow, and requires the client's informed consent before anything relating to the representation is entered into a self-learning tool. A boilerplate clause in the engagement letter isn't enough. Mata v. Avianca, where a federal court sanctioned lawyers in 2023 for filing a brief citing non-existent cases produced by ChatGPT, is why your review line should read “verify every citation in the original source.”
- Accounting and tax firms. Internal Revenue Code section 7216 restricts what preparers may do with return information, including sharing it. Under the FTC Safeguards Rule, a tax practice needs a written security program and has to keep watch over the vendors that touch customer data; IRS Publications 4557 and 5708 help small firms write that plan. List each approved AI tool in that plan.
- Medical and dental practices. Any AI vendor that handles protected health information must sign a Business Associate Agreement first. Your approved-tools table should say which tools have one, and staff should treat every other tool as off-limits for patient details. California physicians' offices (solo or group), clinics and health facilities have one more line to add: AB 3030 (Health and Safety Code section 1339.75) requires AI-generated patient communications about clinical information, written or spoken, to carry a disclaimer and tell patients how to reach a human, unless a licensed or certified health care provider has read and reviewed them first. Dental offices aren't among the settings the law names, so a dental practice should confirm with counsel whether it applies.
If you want a deeper framework later, NIST's AI Risk Management Framework, released in January 2023 for voluntary use, is free to download.
Rolling it out in one meeting
- 01Before the meeting, ask staff anonymously which AI tools they already use and for what. You'll learn where the real exposure is.
- 02Fill in the approved-tools table with at least one tool people can actually use. A policy with no approved option gets ignored.
- 03Walk through the page in 20 minutes using three real examples from your own work: one that's fine, one that's fine with care, and one that's never allowed.
- 04Have everyone sign and date it, and pin the one-minute check beside their screens.
- 05Put the review date in the calendar before the meeting ends.
If you need examples for step three, these tend to start the right conversation. Swap in situations from your own week.
| Situation | Verdict | Why |
|---|---|---|
| Rewording the holiday-hours notice for the website | Fine | No client data; a person reads it before it's posted |
| Summarizing a client's 30-page lease in the firm's business account | Fine with care | Approved tool; the summary is checked against the lease before anyone relies on it |
| Drafting a reply to a client email that includes their balance | Fine with care | Only inside approved software, and a person approves it before sending |
| Pasting a patient's visit notes into a free chatbot to draft a referral letter | Never | Patient information in a tool with no Business Associate Agreement |
| Reformatting a client's bank statement on a personal account | Never | Financial details leave the firm's control entirely |
Writing the policy often surfaces a bigger question: which tasks are worth automating properly rather than with copy-and-paste? Mapping that properly is what our AI consulting work is for. For a closer look at the data question, see whether it's safe to use AI with client data.
Sources
- American Bar Association: Formal Opinion 512, Generative Artificial Intelligence Tools (July 29, 2024)
- U.S. District Court, S.D.N.Y.: Mata v. Avianca, Inc., Opinion and Order on Sanctions (June 22, 2023)
- IRS: Section 7216 Information Center
- Federal Trade Commission: FTC Safeguards Rule: What Your Business Needs to Know
- IRS Publication 4557: Safeguarding Taxpayer Data
- IRS Publication 5708: Creating a Written Information Security Plan for Your Tax and Accounting Practice
- U.S. Department of Health and Human Services: Business Associates
- NIST: AI Risk Management Framework
- California Legislative Information: Health and Safety Code section 1339.75 (added by AB 3030, Chapter 848, Statutes of 2024)
- California Department of Public Health: AFL 25-07, Assembly Bill 3030, Health Care Services: Artificial Intelligence
- California Legislative Information: Penal Code section 632 (recording confidential communications)