# AI for New Zealand law firms.

URL: https://levystreet.com/industries/legal/
Updated: 2026-10-19

How New Zealand law firms use AI for intake, contract review and chronologies, what the Law Society and courts expect, and where generative AI fails lawyers.

## Short answer

AI for lawyers in New Zealand is most useful for reading and organising documents: triaging new enquiries, reviewing contract clauses against a playbook with citations to the source, and extracting chronologies from case files. The NZ Law Society is clear that lawyers remain responsible for advice or documents produced with generative AI, so every output needs a source a lawyer can check.

- The Law Society says lawyers are not absolved of responsibility because work came from generative AI.
- NZ courts are already dealing with citations to cases that do not exist.
- Test AI systems on fictional data, not client information, as the Law Society advises.
- Chronology extraction and clause review with citations are the strongest first projects.

## Where AI helps a New Zealand law firm

AI helps law firms with the reading that sits underneath legal judgement: sorting new enquiries, comparing contracts with a playbook, pulling dates out of a document bundle, and finding what the firm has already written on a point. It does not replace the judgement itself, and it is unreliable on New Zealand authority when it answers from memory, because most of what models were trained on is American, English and Australian law.

The more immediate issue for most firms is unmanaged use. In Datacom's 2025 State of AI Index, a vendor survey of 200 senior leaders at NZ organisations with 100 or more staff, [52% said "shadow AI"](https://datacom.com/au/en/insights/articles/datacom-releases-2025-state-of-ai-index-research-report), staff using unapproved tools, was a problem. In a law firm that means client confidences in personal chatbot accounts. A sanctioned tool that answers from the firm's own documents is usually the fastest fix.

## Legal workflow map

| Workflow | What AI does | What a person still decides | Data involved | Difficulty |
|---|---|---|---|---|
| [Legal intake and matter triage](/industries/legal/legal-intake-and-matter-triage/) | Reads enquiry forms and email, extracts parties, matter type and urgency, and drafts a file note | Whether to act, conflict clearance, the fee estimate | Enquiry emails, web forms, call notes | Medium: urgency and conflicts |
| [Contract clause review with source citations](/industries/legal/contract-clause-review-with-source-citations/) | Compares each clause with the firm's playbook, quoting the clause and citing its number for every flag | Negotiating position and advice to the client | Contracts, playbooks, precedent clauses | Medium: depends on playbook quality |
| [Chronology extraction from case documents](/industries/legal/chronology-extraction-from-case-documents/) | Builds a dated event list, each entry citing document and page | What is relevant, disputed or privileged | Correspondence, briefs of evidence, records | Medium: dates and pages drift |
| Conflict check pre-screening | Lists every party, related entity and former name in the intake documents for the conflicts search | Whether a conflict exists | Intake documents, practice management records | Low: produces lists only |
| First-pass disclosure review | Tags documents for likely relevance and privilege, for lawyer review | Relevance, privilege and what is disclosed | Document sets in the DMS | High: privilege mistakes are costly |
| Internal precedent search | Finds the firm's own precedents and past advice on a point, with links | Whether the precedent fits this matter | DMS, knowledge bank | Medium: matter permissions |
| Time entry narratives | Drafts narratives from calendar entries, emails and documents worked on | What is billed and how it is described | Time records, activity logs | Low: lawyer edits first |
| Client care and engagement letters | Drafts letters from templates and intake details | Scope, fees and terms | Templates, intake notes | Low: template-bound |

Three of these have worked guides: [intake and matter triage](/industries/legal/legal-intake-and-matter-triage/), [contract clause review with citations](/industries/legal/contract-clause-review-with-source-citations/) and [chronology extraction from case documents](/industries/legal/chronology-extraction-from-case-documents/).

## What the Law Society and the courts expect

New Zealand has the clearest professional guidance of any sector on this site.

- **NZ Law Society guidance.** Issued in March 2024 and [updated on 24 February 2026](https://www.lawsociety.org.nz/professional-practice/rules-and-maintaining-professional-standards/generative-ai-guidance-for-lawyers/), it says privacy and fair trading requirements apply on top of the Lawyers and Conveyancers Act (Conduct and Client Care) Rules 2008, and that "a lawyer is not absolved from responsibility for legal advice or defects in an end-product (such as a contract) because it is derived from Gen AI." It flags that inputs may be transferred to AI companies overseas, with Privacy Act implications, and says client information should not be used for testing AI systems or generating templates.
- **The December 2025 warning.** The Law Society [said on 17 December 2025](https://www.lawsociety.org.nz/news/newsroom/lawyers-are-responsible-for-the-contents-of-court-documents-prepared-using-generative-ai/) that New Zealand courts "are already grappling with references to cases that do not exist", and that firms should have clear policies for all staff and quality-assurance monitoring.
- **The courts' guidelines.** Published on 7 December 2023 and [applying to all courts](https://www.courtsofnz.govt.nz/assets/7-Publications/OCJ-Media/20231207-MR-publication-guidelines-for-use-of-Generative-AI.pdf), the Waitangi Tribunal and 28 other tribunals, the [lawyers' version](https://www.courtsofnz.govt.nz/assets/6-Going-to-Court/practice-directions/practice-guidelines/all-benches/20231207-GenAI-Guidelines-Lawyers.pdf) makes you responsible for the accuracy of everything you file, does not require routine disclosure of AI use unless asked, and notes that chatbots generally do not account for New Zealand's cultural context or the values of Māori and Pasifika.

## The legal software stack AI has to fit

Most NZ firms run a practice management system such as LEAP or Actionstep, a document management system such as iManage, NetDocuments or SharePoint, and Microsoft 365. An AI system should read from these with the same permissions lawyers have, not from a copy dumped into a chatbot. That matters most for information barriers: if an AI search index ignores matter-level access controls, it can surface one client's documents to a team walled off from them. For statutes, retrieve the current version from legislation.govt.nz rather than trusting a model's memory. Our guides to [private AI search](/engineering/building-private-ai-search/) and [SharePoint](/integrations/sharepoint/) cover permission-aware retrieval, and [RAG development](/services/rag-development/) is how we build it.

## A first project: chronology extraction you can check

Chronologies suit AI because every entry can be verified against a cited page, and the Law Society's test-data rule is easy to respect.


1. **Build a fictional bundle.** Create 150 to 300 realistic documents for a typical matter type, with no client information, and have a lawyer prepare the answer-key chronology.
2. **Fix the output format.** Date, event, people involved, source document and page, plus a confidence flag for inferred dates.
3. **Score it.** Measure recall against the answer key, date accuracy, and whether every citation opens to a page that supports the entry.
4. **Move to live matters carefully.** Use a provider whose terms keep inputs confidential and untrained-on, and have a lawyer verify every entry on the first matters.
5. **Decide on evidence.** Keep the system only if citation accuracy holds on live bundles.


Illustrative example: a 12-lawyer litigation practice builds chronologies from 2,000-page bundles. A junior working at about 80 pages an hour needs roughly 25 hours. If the junior instead verifies an AI draft at about 250 pages an hour, that is 8 hours, a saving of about 17 hours per bundle, or around 50 hours a month at three bundles. Our [AI document processing](/services/ai-document-processing/) work starts with exactly this kind of scored test.

## Where generative AI fails lawyers


- **Invented authorities.** Never file a case the system did not retrieve from a real source, and check that each one says what the draft claims.
- **Wrong jurisdiction.** Models mix in Australian, English and American concepts. Restrict retrieval to New Zealand sources and say so in the instructions.
- **Out-of-date law.** Since February 2026, an employer dismissing someone who earns $200,000 or more a year (unless they have opted back in) does not need to show good reason or follow a fair process, as [Employment New Zealand explains](https://www.employment.govt.nz/starting-employment/rights-and-responsibilities/good-faith). A model trained earlier will not know. Check every statement against the current statute.
- **Broken information barriers.** The search index must inherit document permissions from the DMS, tested before launch.
- **Page drift.** OCR and document splitting shift page numbers. Cite document and page together and spot-check them.
- **Missing cultural context.** Matters involving tikanga, whenua or whānau need a lawyer's judgement; the courts' guidelines say chatbots generally miss it.


Firms that want help with policy as well as tools can start with [AI governance advisory](/services/ai-governance/).

## Questions people ask

### Can New Zealand lawyers use ChatGPT or Claude for legal work?

Yes, within limits. The NZ Law Society's guidance says there is no overarching AI regulation in New Zealand but lawyers remain responsible for the services they provide, and the conduct rules, privacy law and fair trading law all apply. In practice that means a business plan that does not train on your inputs, no client data in personal accounts, and a lawyer checking every output against its source.

### Do we have to tell the court we used generative AI?

Not as a matter of course. The courts' guidelines for lawyers, published in December 2023, say you do not need to disclose use of a generative AI chatbot unless the court or tribunal asks. They also say you are responsible for ensuring everything you put before the court is accurate, which includes every citation an AI tool suggested.

### Can we test an AI tool on our own client files?

The Law Society's guidance says personal or client-related information should not be used for testing AI systems or generating templates, and that fictional data should be used instead. Build a realistic fictional bundle with a lawyer-made answer key, prove the system on that, and only then consider live matters under proper confidentiality terms with your provider.

### What should a law firm's AI policy cover?

In December 2025 the Law Society said firms should have clear policies for all staff on using generative AI and on how quality will be monitored. A workable policy names approved tools, sets which information can go into them, requires source checking for every authority, covers billing and client disclosure, and assigns someone to review outputs. Our AI use policy template is a starting point.

### Should AI change how law firms bill?

Possibly. The Law Society's guidance notes that lawyers may need to review billing practices and the information given to clients at the start of a retainer as AI takes on tasks. If a chronology that took 25 hours now takes 8, time billing passes the saving to the client, which may be right, but the engagement terms should say how AI is used and reviewed.


## Sources

- [Law Society releases generative AI guidance for lawyers](https://www.lawsociety.org.nz/news/newsroom/law-society-releases-generative-ai-guidance-for-lawyers/)
- [Generative AI guidance for lawyers](https://www.lawsociety.org.nz/professional-practice/rules-and-maintaining-professional-standards/generative-ai-guidance-for-lawyers/)
- [Lawyers are responsible for the contents of court documents prepared using generative AI](https://www.lawsociety.org.nz/news/newsroom/lawyers-are-responsible-for-the-contents-of-court-documents-prepared-using-generative-ai/)
- [Guidelines for use of generative AI in courts and tribunals: lawyers](https://www.courtsofnz.govt.nz/assets/6-Going-to-Court/practice-directions/practice-guidelines/all-benches/20231207-GenAI-Guidelines-Lawyers.pdf)
- [Media release: guidelines for use of generative AI](https://www.courtsofnz.govt.nz/assets/7-Publications/OCJ-Media/20231207-MR-publication-guidelines-for-use-of-Generative-AI.pdf)
- [Datacom releases 2025 State of AI Index research report](https://datacom.com/au/en/insights/articles/datacom-releases-2025-state-of-ai-index-research-report)
- [Good faith](https://www.employment.govt.nz/starting-employment/rights-and-responsibilities/good-faith)

Published by Levy Street (https://levystreet.com), Wellington, New Zealand.