Date posted: 19/11/2026

Build your AI policy: a step-by-step method

AI does the work. You make the decisions. A human signs off.

By Inbal Rodnay

How to use this

  • Read What your AI policy has to cover first. This method assumes you know why an AI policy has to teach, make real decisions and stay current. It also assumes you know the six core areas (scope, confidential information, accountability, record-keeping, disclosure and agentic AI), plus training and what to do when something goes wrong.
  • The method fits in both situations:
    • if you have no AI policy yet
    • if you have one and want to find its gaps
  • Budget a couple of days of your own time, spread over a week or two. Most of it is step 2.

Use an AI tool your organisation has approved (or, if you work alone, one you have checked yourself). No approved tool yet? Use a business version signed in with your work account, and don’t upload anything confidential. Before uploading internal documents, check that you are allowed to share them with that tool. Remove client and staff names and other confidential details first.

For the research steps, use a tool that can search the web. Open the links it provides to check where the information comes from and when it was published. AI can invent sources and dates, so treat anything you can’t open and check as unverified. Not sure your tool can search the web? Ask it: “Can you search the web right now? Show me a link to one news story from today.”

The prompts include details for you to fill in. For example, replace [Australia / New Zealand] with the country you work in before using the prompt.

New to prompting? Copy a prompt, replace everything in [square brackets] with your own details, and paste it into your AI tool. To attach a file, use the paperclip or upload button. Then keep the conversation going: answer the AI’s questions, push back, and ask “what does that mean?” whenever you’re unsure. Save what you’ll need later as a document: the reading list from step 1 and your decisions from step 2.

Words you’ll meet. Agentic AI: AI that takes actions for you, such as sending emails or filing documents. Connector: a link that lets the AI read or act in another system, such as your email or accounting software. Skill: a packaged set of instructions you add to an AI tool. Sub-processor: another company your AI provider passes your data to. Privacy impact assessment: a structured check of how a new tool affects people’s privacy, done before you start. Defined terms: words your policies give a set meaning, such as “confidential information”.

Why this method

Many people start an AI policy by asking AI to draft one. They end up with something generic that they don't even fully understand.

I review many AI policies, and all too often my comments include “what does this mean?”

If I don’t get it, some of your staff won’t either.

The point of this method is to use AI to help you understand the issues, make the decisions, and then turn those decisions into a policy your people can actually follow. The drafting comes later. First, you build the judgement the policy depends on.

Build your AI policy

Image credit: Image supplied by Inbal Rodney.

Step 1: Research

Start by having your AI do the research: find reputable, current policy guidance. You are not going to copy any of it. You are going to understand it and take what fits.

Focus on sources that carry authority in Australia and New Zealand. For example:

  • CA ANZ’s AI Resource Centre and AI Fluency Playbook
  • in Australia, the National AI Centre’s Guidance for AI Adoption, with its AI policy guide and template and AI register template
  • in Australia, the OAIC’s guidance on privacy and the use of commercially available AI products
  • for Australian registered tax and BAS agents, the Tax Practitioners Board’s guidance on AI, TPB(GS) 55/2026
  • in New Zealand, the Office of the Privacy Commissioner’s guidance on AI, which recommends a privacy impact assessment before you start
  • in New Zealand, MBIE’s Responsible AI Guidance for Businesses
  • if you work in government, the Australian Government’s Policy for the responsible use of AI in government, or New Zealand’s Public Service AI Framework

Prompt, step 1.

“I am developing an AI policy for [describe your organisation and its work] in [Australia / New Zealand / both].

Find the most reputable and current guidance on AI policies relevant to this context, including, where relevant to my country, from CA ANZ, the National AI Centre, the OAIC, the Tax Practitioners Board, New Zealand’s Office of the Privacy Commissioner and MBIE.

Suggest additional reliable sources you find. Only list sources you have opened in this session, and prefer regulator, government and professional-body pages over commentary.

If I work in government, also include the Australian Government’s Policy for the responsible use of AI in government or New Zealand’s Public Service AI Framework.

For each source, give me the link, the publication or latest update date, and a list of the areas it covers. Say whether it applies in Australia, New Zealand or both, and flag anything that has been superseded. If you can’t confirm a detail, mark it “unverified” rather than guessing. Do not draft anything. We are building a reading list first.”

You should get: a list of sources, each with a link, a date and the country it applies to. Open a few of the links yourself to check them, then save the list as your reading list.

If it goes wrong: if a link doesn’t open or a date looks wrong, say: “That link doesn’t work. Only list sources you can open.” If the list leans towards the other country, say:

“I work in New Zealand only. Remove anything that applies only in Australia” (or the other way round).

Step 2: Understand

This is the step that makes the difference, so do not rush it. Take what the research surfaced and have AI explain all of it back to you in plain language, so you can make decisions. Not "summarise this", but "teach me this and show me my options".

For the prompt’s first line, write something like: “We are a six-person bookkeeping and tax practice. Staff can try AI for emails and summaries, but not with client information yet.” Or: “We are the 12-person finance team of a not-for-profit, and we want people to experiment within clear limits.”

Prompt, step 2.

"We are [size and type of organisation, and how much we want people to experiment].

Based on the sources you found (or the reading list attached), walk me through the key areas an AI policy should cover, one at a time, starting with scope.

Include at least scope, confidential information, accountability, record-keeping, disclosure, agentic AI, training, and what to do when something goes wrong.

For each area: explain in plain language what it means, why it matters for an organisation like mine, the realistic options I could choose between, and what each option would mean in practice on an ordinary working day.

Do not write the policy yet. Help me understand each area well enough to decide. Distinguish legal and professional requirements from choices our organisation can make. For each requirement, give the source and link. If you can’t, label it “confirm with an adviser”. Don’t present good practice as law.

Ask me questions where my context would change the answer, and wait for my answers before moving to the next area.
Keep the options proportionate to who we are."

You should get: a plain-language explanation of scope, your options, and a few questions for you. Then it should stop and wait for your answers.

If it goes wrong: if it covers every area at once, say: “Stop. One area at a time, please.” If it starts drafting the policy, say: “Not yet. Help me decide first.” To check an answer, open the link for anything it calls a law or a rule and find that sentence on the page. If you can’t find it, treat it as unverified.

Then have the conversation. Push back, ask what you do not understand, talk it over with whoever shares accountability (partners, board, leadership team or a trusted peer) and with your team. This is where you understand the capability, see the risk, and make your decisions. By the end you know what you want, and so do the people the policy will govern.

This step may span several AI conversations over a few days, with additional research and conversations with your team. This is the heart of the work. Starting a new chat? The AI won’t remember the last one. Attach your reading list and your decisions so far, then say: “We’ve covered [areas done]. Carry on with [next area], one area at a time.”

If you already have an AI policy, add it to the mix. Keep a running list of the gaps. That list is your decision set. Check in particular for three gaps that show up again and again: skills (packaged instructions you add to an AI tool), agents, and the assumption that approving one AI tool approves every feature, connector and automation inside it.

Add your existing AI policy (if you have one) and ask: "Now compare what we just discussed with our current policy. Help me understand what it already covers, what it doesn't, and where it is out of date. Check especially for skills, agents and connectors. Keep a numbered list of the gaps."

Write your decisions down as you go. One line per decision with some reasoning, in your own words. "Client information may go into our approved workplace AI tool, and only where our engagement terms allow it." "Tax file numbers and IRD numbers never go into an AI tool." "Every piece of AI-assisted advice is reviewed by a named person before it goes out." "AI notetakers may join internal meetings once attendees know; client meetings need the client's agreement first." "Skills and connectors from outside our approved vendors are reviewed before anyone installs them." "Payroll and staff records never go into an AI tool." "Everyone completes AI training before they get access." "If client information goes into the wrong tool, tell [named person] the same day." If you can't write a decision in one line, you haven't made it yet. These lines are the raw material for step 3.

Step 3: Draft

Now you are ready to draft or update your AI policy. Give AI three things: the decisions you wrote down, any existing policies you want it to match, and, if you have one, the structure you want. If you don’t, the prompt asks the AI to propose one. Keep the policy itself short. It often works better as a few pages of rules, with the explanations in a separate guidance note or a one-page chart of what can be used for what.

Prompt, step 3 (if you don't have an AI policy).

"Now draft our AI policy. Use the decisions I have made (attached). Propose a structure for the policy and wait for my approval before drafting.

Match the tone and defined terms of our existing policies (attached, if any).

Flag any overlaps or contradictions with those existing policies and suggest how to cross-reference them rather than duplicate.

Where I have not given you a clear decision, leave a marked gap with a question. Do not invent a position.

Write it so that a new staff member with no AI background can read any section and know what it means for their own work. Include an owner, an approval date and a next-review date. Do not add legal citations, dates or obligations that are not in my decisions or the reading list (attached); list any you would suggest separately."

Prompt, step 3 (if you already have an AI policy).

"Now suggest updates to our current AI policy using the decisions I have made and the gap list we created.

Keep every part of the existing policy that still works. Do not rewrite it just for style.

For each gap on my list, insert the decision I have made in the right place in the policy.

Where the existing policy says something that no longer matches how the tools work, suggest replacement wording and explain briefly why it needs to change.

Flag any overlap or contradiction with our other existing policies and suggest whether to cross-reference, remove duplication, or update both documents.

Where I have not made a decision, leave a marked gap with a question. Do not invent a position.

Show every change in a table: section, current wording, proposed wording and reason. List the open decisions separately."

You should get: a draft policy, or a table of changes to your current one, with marked gaps wherever you haven’t decided.

If it goes wrong: if it fills a gap with its own position, say: “Leave that as a marked gap. I haven’t decided.” If it adds a legal requirement you didn’t give it, ask: “Where does that come from? Give me the link.”

Because you fed it your decisions and your existing policy documents, what comes back is yours:

✓ It sounds like your organisation.

✓ It fits with what you already have.

✓ It flags the cross-references that often get missed.

Step 4: Review

Last step: human review. And probably more than one human.

After you review it and make sure you understand everything, ask someone who knows AI quite well to review it, such as your IT provider or a trusted peer. They check that the decisions hold up, that nothing important was missed, and that the capability claims are current. If your obligations are complex, or the policy changes how you handle personal information, also ask someone with legal or privacy expertise to check it, and consider a privacy impact assessment.

Then ask an ordinary member of your team (or, if you work alone, someone outside the business). Their job is simpler and just as important: read it and point to anything they do not understand. If a staff member cannot follow it, others will struggle too. Explain it better.

Finally, have the policy reviewed and approved through your organisation’s usual process. Then share the approved version with your team and explain what it means for their work. A guided reading session works well: it teaches the policy, and you can pick up any nuance that still needs handling. Then update the documents that depend on it: your privacy policy and collection notices, and your engagement letters, contracts or staff guidance. From 10 December 2026, organisations covered by Australia’s Privacy Act must also explain in their privacy policy when computer programs make, or help make, decisions that could significantly affect people. In New Zealand, check your notices against IPP 3A, in force since 1 May 2026.

Now you have a policy, and you have the knowledge behind it. That is a strong position.

The review rhythm

Put a date on the document and a reminder in the calendar. Every six months, give the policy back to your AI:

Prompt, review.

"Here is our AI policy, last reviewed on [date]. The AI tools we use may have changed since then.

Research what has changed in [the tools we use] in that period, especially anything that lets the AI take actions, make decisions or access more data.

Also check whether their terms on training their models, data storage or sub-processors have changed.

Also check for new laws, regulator guidance and professional standards in [Australia / New Zealand] since then. Give me a link and date for each change, and say if you find no evidence of a change.

Then tell me which sections of our policy no longer match how the tools work, and what decisions I now need to make. Do not rewrite the policy. Give me the list."

You should get: a list of changes since your last review, each with a link and date, and the sections of your policy they affect.

If it goes wrong: if it lists changes without links, ask for the source. If it says nothing has changed, check one or two vendor and regulator pages yourself.

A human decides what changes. Then go back to step 2 for any sections that need rethinking: have AI research your options and explain them to you, then draft the changes (step 3) and repeat the review and approval (step 4).

These are the kinds of changes it should catch. In 2026 alone: the Tax Practitioners Board’s AI guidance (July), New Zealand’s new privacy principle IPP 3A on personal information collected from someone else (in force 1 May), and Australia’s new privacy policy rules for automated decisions (from 10 December).

Your next step

An AI policy is not finished when the document is drafted. It is finished when your people understand what it means for their own work.

Use AI to help with the research, the explanations, the comparisons and the drafting. But keep the decisions with the people who are accountable for the work.

That is what makes the policy usable. And that is what makes it yours.

Sources

CA ANZ, AI Ethics & Governance, AI Resource Centre

CA ANZ, AI Fluency Playbook

National AI Centre, Guidance for AI Adoption

National AI Centre, AI policy guide and template; AI register template

Office of the Australian Information Commissioner, Guidance on privacy and the use of commercially available AI products

Office of the Australian Information Commissioner, New resources on transparency for use of AI and automated decision-making, September 2026

Tax Practitioners Board, TPB(GS) 55/2026: The use of Artificial Intelligence and the Code of Professional Conduct

Office of the Privacy Commissioner (New Zealand), Artificial intelligence and the Information Privacy Principles, September 2023

Ministry of Business, Innovation and Employment (New Zealand), Responsible AI Guidance for Businesses, July 2025

Ministry of Justice (New Zealand), New privacy information principle now in force

Digital Transformation Agency, Policy for the responsible use of AI in government, v2.0

New Zealand Government, Public Service AI Framework

This resource is a method for developing your own AI policy. It is not a policy, not a template, and not legal advice. Sources named in step 1 are external and are not reproduced here. Tool capabilities are correct as at November 2026 and will change, which is why the review rhythm exists. Align your policy with the legal and professional obligations that apply to your work, including APES 110 in Australia or the NZICA Code of Ethics in New Zealand, relevant privacy legislation (the Privacy Act 1988 in Australia; the Privacy Act 2020 in New Zealand) and, in Australia, Tax Practitioners Board, ASIC or APRA requirements where applicable.

Disclaimer
This guide provides general guidance only and is current as at October 2026. Tool names, features and capabilities were correct at that date but may change. The publication therefore focuses on general principles rather than specific feature lists.

This guide is not intended to be comprehensive and is not a substitute for professional, legal or other advice tailored to your circumstances. Use it alongside your organisation’s artificial intelligence policies and your applicable legal, regulatory, professional and ethical obligations. Check current vendor terms, product information and your own obligations before acting on or relying on the information.

While Chartered Accountants Australia and New Zealand (CA ANZ) seeks to ensure that the information is accurate and current, it does not guarantee its accuracy, completeness, currency or suitability for any particular purpose. To the fullest extent permitted by law, CA ANZ disclaims all warranties and liability for any direct, indirect, special or consequential loss or damage arising out of or in connection with any inaccurate statement, opinion, advice or information contained in this publication, any omission from it, or the use of or reliance on this [guide/document] or any material it contains.

© 2026 Chartered Accountants Australia and New Zealand ABN 50 084 642 571

Search related topics