Why NZ Law Firms Are Looking at ChatGPT Now

Every partner I speak with in Auckland, Wellington, or Christchurch is asking the same two questions. Can ChatGPT actually help us, and can we use it without breaking our obligations. The honest answer to both is yes, with the right guardrails. Around the country we are seeing sole practitioners, mid tier firms, and a few large practices quietly start to use generative AI for the parts of legal work that drain time without adding judgement.

The reason is simple. NZ legal work is being squeezed. Conveyancing margins are tighter than they were five years ago. Discovery and document review for litigation keeps growing. Compliance work tied to AML and the FIC Act keeps getting layered. Partners want to spend their time on advice and client relationships, not on first drafts of memos that a capable assistant could do faster with the right tool.

The tension is that law firms carry higher ethical and confidentiality duties than most businesses. The Law Society’s Professional Conduct rules, client privilege, and the Privacy Act 2020 all apply. So the question is not whether to use ChatGPT. The question is how to use it without putting your practising certificate or your clients at risk.

What “Use” Actually Looks Like in a NZ Practice

When NZ lawyers say they are “using ChatGPT”, they usually mean one of four things. The first is drafting assistance, where ChatGPT produces a first draft of a letter, a memo, a board paper, or a contract clause. The second is research support, where the lawyer asks the model to summarise case law or statute, then verifies the citations themselves. The third is document review, where ChatGPT reads a stack of discovery documents and flags the relevant ones. The fourth is client comms, where the model rewrites a partner’s notes into a clean client update.

None of those uses are exotic any more. The pattern we see across our network is a firm starts with one lawyer, the news travels, and within a few months there are five or ten people on the tool. That is the moment where the lack of a written policy becomes a real problem. The conversation moves from “should we” to “how do we govern this properly”.

One thing worth saying upfront. ChatGPT is a drafting and research aid, not a legal advisor. Anything it produces needs to be checked by a human with the right qualifications. Treat it the same way you would treat a junior lawyer’s first draft. Useful, often impressive, never relied on unchecked.

The Privacy Act 2020 Question You Can’t Avoid

This is the area where most NZ firms stumble. The Privacy Act 2020 sets out 13 Information Privacy Principles, and several of them bite when you paste client information into a tool hosted offshore.

The short version. When you put personal information about a client, a witness, or a third party into ChatGPT, that information is being sent to a server, most likely in the United States. That triggers IPP 8 on disclosure of personal information, and the new cross border disclosure rules that came in with the 2020 Act. Under IPP 12, if you send personal information offshore you have to take reasonable steps to ensure the recipient is subject to comparable privacy obligations, or you have to get authorisation from the individual.

For most NZ law firms the practical answer is twofold. First, talk to OpenAI about their enterprise offering and the data processing terms. Second, build a firm policy that says what categories of information can and cannot be entered. Names, addresses, IRD numbers, and any identifying case details usually sit in the “do not enter” bucket. Anonymised facts and statute references usually sit in the “fine to enter” bucket.

This is also the area where you should get a privacy lawyer involved. The detail here matters, and it changes. Verify the current state of the cross border rules with your lawyer before you roll ChatGPT out at scale.

Where ChatGPT Helps and Where It Hurts

The high value uses for a NZ law firm are the repetitive, low judgement ones. We typically see the strongest return in three areas. First, drafting correspondence and contract templates. Second, summarising long documents like lease schedules, terms of trade, or discovery bundles. Third, rewriting complex legal English into plain English for clients who are not lawyers.

The areas to be careful with are anything involving legal advice, anything that influences a court filing, and anything that touches on credit, employment, or insurance decisions about a person. Those are areas where you need a human sign off and a clear audit trail. The other place to be careful is citation. ChatGPT will confidently invent case references that do not exist. We have seen it happen. Always verify any citation against a primary source like the NZ Courts database or a LexisNexis check.

A useful mental model is to split work into three buckets. The “ChatGPT does it” bucket is research, summarisation, and first drafts. The “ChatGPT helps, lawyer finishes” bucket is correspondence, contract clauses, and client updates. The “lawyer does it alone” bucket is anything that goes to a court, anything that constitutes legal advice, and anything that affects a person’s rights.

A Realistic NZ Cost Picture

Pricing for the relevant ChatGPT tiers is published in USD and shifts periodically, so the figures below are a rough guide only. As a working conversion, one USD is roughly NZD 1.65 at the time of writing, but treat these as approximate and check current rates before you commit.

The free tier is fine for a partner to experiment with. For a team, the ChatGPT Team plan is published at around USD 25 per user per month, which works out to roughly NZD 41 per user per month before GST. For a five lawyer firm that is around NZD 205 per month, or about NZD 2,500 per year. For a firm that wants stronger data controls, the Enterprise tier is custom priced and typically starts in the USD 60 per user per month range, which is roughly NZD 99 per user per month before GST. Enterprise gives you an admin console, SSO, and clearer data processing terms, which is the version most NZ law firms should be looking at once they have more than a handful of users.

Compare that to a junior lawyer’s fully loaded cost. Even at a conservative NZD 70,000 to NZD 90,000 per year for a first year solicitor, an AI tool that frees up ten percent of their time pays for itself many times over. The economics are not the obstacle. The governance is.

Practical Workflows That Hold Up Under Review

If you want workflows that will survive a Law Society review, here are three we see working well in NZ firms.

The first is the “anonymise, draft, verify” workflow for client letters. The lawyer strips identifying information, asks ChatGPT to produce a first draft based on the underlying facts, then reviews and adds the specifics back in. Time savings on a standard letter run from around 30 minutes to closer to 10.

The second is the “summarise then cite” workflow for due diligence or discovery. The lawyer uploads a redacted document set and asks for a summary of the key issues, then does the actual legal analysis themselves. We typically see document review time drop by 30 to 50 percent on standard commercial matters, with the caveat that a senior lawyer still has to verify the output.

The third is the “plain English rewrite” workflow for client updates. The partner writes their notes, ChatGPT turns them into something a client can read, the partner reviews and sends. This is particularly useful in litigation and family law, where clients are paying for clear guidance at stressful moments.

The common thread is that the lawyer stays in the loop. ChatGPT does the first 60 percent. The lawyer does the last 40 percent, which is the part that actually requires judgement.

Building a Firm Policy That Protects You

A good AI policy for a NZ law firm does not need to be 30 pages. Two pages is enough if it covers the right things.

The policy should say who is allowed to use ChatGPT and on which tier. It should set out what categories of information can and cannot be entered, with examples rather than principles. It should require verification of any citation, any figure, and any legal claim before it leaves the firm. It should require a human reviewer for anything that goes to a client, a court, or the other side’s lawyer. It should say where prompts and outputs are stored, and how long they are kept. And it should name a partner who is responsible for the policy and reviews it every six months.

You also want to align this with your existing systems. If you run practice management software like LEAP or AffinityLive, your policy should say how ChatGPT sits alongside it. If you use Xero or MYOB for trust accounting, the policy should be clear that AI tools do not touch the trust ledger. The point is to make the rules boring and repeatable. Boring policies get followed. Exciting policies get ignored.

How to Start Without Burning Trust

If you are a partner reading this and you have not started yet, the path of least regret is to pick one lawyer, one workflow, and one month. Give that lawyer access to the Team or Enterprise tier with a clear written scope. Pick a workflow that does not touch privileged content, so summarising publicly available statutes or rewriting your own internal memos. Measure the time saved and the quality of the output. At the end of the month, decide whether to expand.

The reason to start small is not caution for its own sake. It is that you cannot write a good policy until you have seen how the tool actually behaves in your firm. After a month of real use, you will know where the sharp edges are, and you can write a policy that reflects how your people actually work rather than how you imagine they work.

It is also worth telling your clients. The Lawyers and Conveyancers Act and your professional conduct obligations do not require you to disclose every internal tool, but a short note in your engagement letter or terms of engagement that says “we may use AI assisted tools to support our work, with human review at every step” is a sensible practice. Clients who hear about AI from you first are relaxed. Clients who hear about it from someone else get nervous.

Common Mistakes We’ve Seen

A few patterns come up over and over across the firms we work with.

The first is using a personal ChatGPT account for client work. Personal accounts have weaker data controls and the prompts are used to train the model unless you opt out. For any client work, you want an Enterprise or API tier with a data processing agreement.

The second is treating ChatGPT as a search engine. It is not. It produces plausible text, which is not the same thing as accurate text. The hallucination problem is real and not going away. Always verify.

The third is letting unsupervised use spread. Once one lawyer has access, every other lawyer wants access. That is a feature, not a bug, but you want the rollout to be controlled so the policy is in place before the wave hits.

The fourth is forgetting the regulatory perimeter. If you do litigation work, the High Court Rules around filing signed documents still apply. If you do conveyancing, the anti money laundering rules under the AML/CFT Act still apply. AI does not change your underlying obligations. It just changes how you fulfil them.

The Next 12 Months for NZ Law Firms and AI

The direction of travel is clear. AI tools are getting cheaper, the New Zealand judiciary is publishing more guidance on AI use in court processes, and clients are starting to ask whether their lawyers use these tools. The firms that work this out in 2026 and 2027 will be the firms that win the next decade of work. The firms that bury their heads will find themselves explaining to a client in 2028 why their competitor delivered the same work in half the time at a lower cost.

You do not need to become an AI expert. You do need to become a literate buyer and a sensible governor. That means a written policy, a named responsible partner, an appropriate tier of ChatGPT, and a habit of human review on anything that matters. That is the whole job.

If you want a practical starting point, the next step is a short working session where we look at your current workflows, your data handling rules, and your client mix, and we map out the two or three places where AI can give you the biggest lift without putting your obligations at risk.

If this is the kind of problem agents can help with, the free Working With Claude field guide is the practical next step. Thirty-two pages, no fluff. Get the free guide.