Introducing the Lex Praxis AI Output Auditor — a free tool that reads AI-generated employment law text so you can see exactly what the AI got right, what it got wrong, and what requires a practitioner.
A client sent me a submission last month to review. It read well. The tone was measured, the logic appeared sound, and the structure followed the right sequence. Then I looked at the cases it cited.
One of them did not exist.
It wasn't a real case with the wrong citation or with the wrong date but one that had never been decided, never been filed, and never been heard. It was a neutral citation assembled by an AI from the materials it had absorbed in training, offered with the same confidence as every other sentence in the submission.
The client had not noticed but luckily had not yet submitted the document. That was the only reason we were having a conversation rather than preparing a response to explain the oversight.
AI generates text. It does not verify it. That is a different function — and until now, there has been no easy way to perform it.
Today that changes, for Lex Praxis clients and for anyone working with AI-generated employment law content in New Zealand.
The Lex Praxis AI Output Auditor is now live. It is free. It takes less than a minute to use. And it will tell you more about what your AI actually produced than any amount of re-reading or prompt engineering the text yourself.
What the auditor does
Paste any AI-generated employment law text — whether it's a policy clause, a dismissal letter, an investigation finding, or a redundancy rationale — into the auditor. It reads the text, extracts every factual and legal claim, and classifies each one.
The classification runs across four categories:
- Verified — the claim is supported by an identified primary source.
- Needs Checking — the claim appears plausible but cannot be confirmed without further verification.
- Unsupported — the claim is not grounded in any identifiable legal authority.
- Likely Wrong — the claim appears to be incorrect based on known NZ employment law.
The auditor also surfaces invented elements separately — fabricated case names, non-existent statutory provisions, fictional regulatory bodies — and flags risk areas that a practitioner or employer would want to examine before the document goes anywhere.
Every finding is tagged to a step in the Lex Praxis AI Verification Framework — the seven-step methodology that underpins all Lex Praxis work product. That tagging is deliberate. It makes the methodology visible and connects each finding to the specific verification layer it represents.
At the top of the results, you get a reliability score between zero and one hundred. It is not a grade, and it is not a pass/fail. It is a signal: how much of what the AI produced is actually anchored in verifiable authority?
What the auditor does not do
The auditor surfaces the problem. It does not fix it.
When a claim is flagged as Likely Wrong or Unsupported, the auditor does not produce a corrected version. That is intentional. Rewriting an employment document — understanding why the AI got it wrong, identifying the correct legal position, and producing text that will hold up — is practitioner work. It is the responsibility of the employer or their counsel to fix it before it goes anywhere. Automation cannot do it, and it would be dishonest to suggest otherwise.
Every 'Likely Wrong' badge is a call to action, not a resolution.
This distinction — between identifying a problem and solving it — is the core of the Lex Praxis model. The auditor makes the problem visible. What you do with that information is where the practitioner comes in.
Why this matters for NZ employers and law firms
The LEAP Profitability in Law: Global Report 2026 found that New Zealand has the highest verification and trust concerns of any region surveyed globally. One in four NZ legal professionals report low or no trust in how AI is being deployed. That scepticism is professionally correct. What has been missing is a tool that gives it somewhere to go.
In New Zealand employment law, the stakes of getting this wrong are concrete. The Employment Relations Act 2000 imposes a justification standard under section 103A that requires an employer to have acted as a fair and reasonable employer would in all the circumstances. A redundancy rationale built on an invented legal principle does not meet that standard — and the ERA will not be impressed by the explanation that an AI wrote it.
The NZ Supreme Court made this position clear in Jones v Family Court at Whangārei [2026] NZSC 1, the Court's first ruling on AI-generated legal submissions. The Court found that a self-represented litigant had filed submissions citing AI-hallucinated authorities — real case names paired with invented neutral citations, referring to cases of no relevance to the application. The Court endorsed the principle already established in the Courts of New Zealand GenAI Guidelines: responsibility for accuracy rests with the person filing, not the tool. That principle applies with equal force to lawyers, and it applies to HR professionals and employers drafting employment documents.
The auditor is the first step in discharging that responsibility properly.
The AVF connection
The AI Output Auditor is not a standalone tool. It is a public expression of the AI Verification Framework — the seven-step methodology Lex Praxis applies to all work product and offers as a standalone client product.
The auditor runs Steps 3 through 6 of that framework:
- Step 3 — Primary Source Anchoring: every claim is traced to legislation, case law, or official guidance.
- Step 4 — Specificity Test: generic or jurisdiction-agnostic claims are flagged.
- Step 5 — Temporal Check: outputs are examined against current NZ law.
- Step 6 — Confidence Signal Prompting: areas of uncertainty are surfaced rather than smoothed over.
Steps 1, 2, and 7 — the Challenge Loop, Cross-Model Check, and Expert Filter — remain practitioner functions. The auditor handles the verification layer. The practitioner handles the judgment layer. That division is not a gap in the tool. It is the point.
How to use it
The auditor is at lexpraxis.co.nz/auditor.html. Paste your AI-generated text. Run the audit. Read the findings.
If the results show a clean document with a high reliability score and no invented elements, you have reasonable grounds for confidence in the text — though practitioner review remains the only complete answer.
If the results show unsupported claims, likely errors, or invented authorities, you have a decision to make. The auditor will point you toward the next step: a fixed-fee document review, an AI Verification Framework engagement, or an ongoing governance conversation if your firm or HR team is running AI outputs regularly.
The audit identifies what the AI got wrong. It cannot tell you what the ERA would make of the underlying decision. That requires a practitioner.
— · —
