Is this actually free, or do I hit a paywall at the end?

Free. The policy is complete on this page, you can copy or download it right now, and there is no email step. The paid kit is a separate thing you can ignore.

Is this legal advice?

No. It is practical business guidance written to be usable, not a legal document. Employment and privacy law differ by country, so have your final version checked against your local law before you rely on it.

Does this cover AI scribes and consultation recording tools?

It covers the decision about them. Those tools sit squarely in section 5: they need written approval before use, and the specific questions are what happens to the recording, whether the transcript is retained, whether it trains a model, and what the patient is told. This policy does not decide for or against them, it stops them arriving without anyone deciding.

Is de-identified clinical information safe to enter?

Less risky, but do not treat de-identification as a switch. A combination of age, location, condition and timing can identify a person in a small community even with the name removed. If your practice permits it, write down what de-identified means for you before someone applies their own definition.

Free generator

An AI usage policy built for a health practice.

Patient information is the most sensitive category your practice holds, and the tools that would save your staff the most time are the ones that want to see it. This writes the policy that separates the two, filled in for your practice.

No email required. No signup. Copy it, download it, use it.

Your business

Business name

Who staff ask when unsure

Who approves new AI tools

Approved tools

Tool 1

Tool 2 (optional)

Tool 3 (optional)

Settings

A breach is treated as

Reviewed every

Policy owner

Require staff to note AI use on customer-facing written work Include the bias and fairness section

AI Usage Policy

AI Usage Policy for [Business name]

A plain-language policy covering what staff may and may not do with AI tools at work. Read it, adapt anything that does not fit, then circulate it.

1. Why we have this policy

[Business name] uses AI tools because they save time and improve our work. They also carry real risks: leaking confidential or customer data, producing wrong or biased output, and creating work that nobody has checked. This policy sets out how we use AI so we get the benefit without the harm.

It applies to everyone: employees, contractors, and anyone acting on behalf of [Business name].

2. The one rule that matters most

Never put information into an AI tool that you would not email to a stranger.

If you are unsure whether something is safe to enter, it is not. Ask your manager before you proceed. Asking is always the right call and nobody will think less of you for it.

3. What you must never enter into a public AI tool

Unless the specific tool has been approved for it in writing, never enter:

Patient names, dates of birth, addresses, and health or national health identifiers
Clinical notes, histories, diagnoses, and test results
Prescriptions, medication lists, and treatment plans
Referrals, specialist correspondence, and discharge summaries
Images, scans, and anything from a patient record system
Billing and insurance details linked to a patient
Any detail that could identify a patient in combination, even where each part alone would not

4. Approved tools

Only use AI tools on this list for work. Using an unapproved tool for work is a breach of this policy.

Tool	Approved for	Not approved for
[Tool name]	Drafting practice-wide patient information material, rewriting general health guidance in plainer language, or administrative correspondence containing no patient detail	Any patient identifier, any clinical content, and any diagnostic, triage or treatment question

Free consumer versions of AI tools often train on what you enter and keep it. Paid business tiers usually let you turn that off. Only approve a tool once someone has checked how it handles your data.

5. Getting a new tool approved

Want to use an AI tool that is not on the list? Do not just start using it. Send a request to the operations manager with three things: what the tool is, what you want to use it for, and what data it would touch. The operations manager will check its data handling and security before approving or declining, and approved tools are added to the table above.

6. You are accountable for what AI produces

AI makes mistakes. It invents facts, gets numbers wrong, and is often confidently incorrect.

Check everything before it leaves the business. You are responsible for any AI-assisted work you send to a customer, publish, or rely on for a decision, exactly as if you had written it yourself.
Never send AI output to a customer or an external party without a person reading it first.
Do not use AI to make final decisions about people, including hiring, firing, discipline and pay, or about anything with legal, safety, or financial consequences. AI can assist your thinking. A person decides.

7. Be honest about AI use

If a customer or a colleague asks whether something was AI-assisted, tell the truth.
Do not present AI-generated work as if it involved professional judgement that it did not.
Customer-facing written work must note where AI was used in preparing it.

8. A clinician reads anything clinical

No AI-generated content is placed in a patient record, sent to a patient, or sent to another provider until a clinician has read it in full and taken responsibility for it.

AI is not to be used to reach or support a diagnostic, triage, prescribing or treatment decision. A tool may help a clinician write something down more clearly. It has no part in deciding what is written down.

Where a tool is proposed specifically for clinical documentation, it goes through the approval step in section 5 with particular care, and the question to answer is what the vendor does with the recording or the text afterwards.

9. Bias and fairness

AI reflects the data it was trained on and can produce biased or unfair output. Be especially careful using AI for anything involving people, and never let it be the sole basis for a decision that affects someone.

10. If something goes wrong

If you accidentally enter sensitive information into an AI tool, or you spot AI output that has caused a problem, tell your manager straight away. The point is to fix it fast, not to assign blame. Reporting something early is always treated better than a hidden problem that surfaces later.

11. Breaches

Not following this policy may be treated as a disciplinary matter under our normal procedures, because it can put the business, our customers, and our people at real risk.

Policy owner: [name and role]

Applies to: [Business name]

Version date: [date]

Review: every six months, because AI tools change fast

The risk that dominates in a health practice

Health information is different from other sensitive data in one specific way: the harm from disclosure lands on the patient rather than the practice, it is often irreversible, and it can be severe from a single record. Most data risks are about volume. This one is not.

The practical difficulty is that the useful applications sit right on top of the sensitive material. Summarising notes, drafting a referral, turning a consultation into a letter. These are real time savings and they all involve the record. That is why this policy leans harder on the written approval step than on a blanket prohibition: an outright ban pushes people to use tools quietly, which is worse than governing them.

The second risk is clinical. Any tool that produces something resembling clinical content, a summary, a suggestion, a plain-language explanation, produces it with the same confident tone whether it is right or wrong. Nothing generated goes near a patient or a record without a clinician reading it.

What this looks like in practice

A practice manager wants to cut the time spent writing referral letters. The tempting version is to paste the consultation notes into a chatbot and ask for a letter.

That sends an identified patient's clinical history to a third party the patient has not been told about, which this policy prohibits outright.

The version that survives the policy is a tool that has been through section 5, where someone has read what the vendor does with the text, confirmed it is not retained or used for training, and recorded the approval. Then the clinician reads every letter before it goes.

What this practice should not do is nothing. Refusing to approve any tool does not stop the letters being written faster, it just moves the activity somewhere nobody can see it. The approval route exists so that the convenient thing and the permitted thing can be the same thing.

Why this is free

An AI policy is the first thing a business needs and the easiest thing to put off. Charging for it would just mean fewer businesses have one. Handing it over, with no email wall in front of it, is also the honest way to show you what our work is like before you spend anything.

Use it, change it, put your own letterhead on it. There is no attribution requirement and nothing to sign.

If the policy was useful

The policy is document one of nine.

A policy tells people where the line is. It does not tell you which tools to trust, where AI is actually worth using in your business, or whether any of it paid off. That is the rest of the kit.

Data governance checklist, so you know what a tool does with your data before it touches it
Tool evaluation scorecard, with two real tools compared and the better product losing
Use-case grid, seven candidates scored including the high-value one worth refusing
90-day adoption plan, with the staff announcement script and the five failure modes
ROI tracker, a worked quarter and the three challenges a sceptic always makes
Prompt libraries for finance, operations, marketing, HR and customer service

Get the full kit$149 one time, nine documents

Other versions of this policy

The same document rebuilt around a different set of risks. If none of these is you, the general version is the place to start.

Questions

Is this actually free, or do I hit a paywall at the end?: Free. The policy is complete on this page, you can copy or download it right now, and there is no email step. The paid kit is a separate thing you can ignore.
Is this legal advice?: No. It is practical business guidance written to be usable, not a legal document. Employment and privacy law differ by country, so have your final version checked against your local law before you rely on it, particularly the breaches section.
Can I edit it and put my own branding on it?: Yes. Change anything, remove sections that do not apply, add your logo. There is no attribution requirement.
Does it work outside New Zealand?: The policy is deliberately written without country-specific law in it, so the substance travels. The one part to check locally is how breaches are handled under your employment rules.
How long should an AI policy be?: Short enough that people read it. A one-page policy that staff follow beats a twenty-page document nobody opens. This one is deliberately about one page once you delete what does not apply.
How often should we update it?: Every six months is a sensible default while AI tools are changing this fast, and immediately if you approve a new tool or something goes wrong.
Does this cover AI scribes and consultation recording tools?: It covers the decision about them. Those tools sit squarely in section 5: they need written approval before use, and the specific questions are what happens to the recording, whether the transcript is retained, whether it trains a model, and what the patient is told. This policy does not decide for or against them, it stops them arriving without anyone deciding.
Is de-identified clinical information safe to enter?: Less risky, but do not treat de-identification as a switch. A combination of age, location, condition and timing can identify a person in a small community even with the name removed. If your practice permits it, write down what de-identified means for you before someone applies their own definition.
Who wrote this?: Plainstart, a brand of Sypher Limited. We publish plain-language operational material for small and medium businesses adopting AI.