evidoria

← Back to browse

Good practice Imported

AI-Assisted Submission Triage — New Zealand's Regulatory Standards Bill Consultation Under Scrutiny

New Zealand · Wellington · See the New Zealand profile · See the Wellington profile

Evidence: Descriptive / self-reported Top 97% 20/100 · Ask Evidence Copilot about this practice

New Zealand's Ministry for Regulation used an LLM to triage 22,821 submissions on the Regulatory Standards Bill (88% opposed) into support/oppose categories, drawing MP and academic criticism over transparency, disputed 'bot' claims and confidence in the process.

22,821
Total submissions received (November 2024 – January 2025)
~88%
Submissions opposing the bill (November 2024 – January 2025)
~0.33%
Submissions supporting or partially supporting (November 2024 – January 2025)
939
Submissions manually reviewed for thematic analysis (November 2024 – January 2025)
605
Additional submissions manually reviewed separately (November 2024 – January 2025)
99.5% (disputed)
Disputed claim: submissions alleged bot-generated (2025)
AI-Assisted Submission Triage — New Zealand's Regulatory Standards Bill Consultation Under Scrutiny

Details

Promoter
New Zealand Ministry for Regulation with Public Voice
Period
Nov 2024 – Jun 2025
Keywords
public consultation, regulatory policy, NLP triage, transparency

Context

Between November 2024 and January 2025, New Zealand's Ministry for Regulation ran a public consultation on its proposed Regulatory Standards Bill and received 22,821 submissions, of which roughly 88% opposed the bill and only about 0.33% supported or partially supported it.

Objectives

Given the volume, the Ministry worked with the research organisation Public Voice and a large language model to classify every email and Citizen Space submission as supporting, partially supporting, opposing or unclear, following a logic model the Ministry designed.

Activities

Staff then manually reviewed a subset — 939 submissions for thematic analysis plus 605 more separately — and translated te reo Māori submissions.

Results

The process drew sustained criticism: Labour MP Duncan Webb argued that citizens who took time to submit 'deserve more than a computer reading their submission,' Victoria University's Dr Eddie Clark warned that routine AI analysis of submissions 'risks undermining people's confidence in the democratic process,' and Minister David Seymour separately made a widely disputed claim that 99.5% of submissions were bot-generated.

Conclusions

The New Zealand Council for Civil Liberties had to file an Official Information Act request to seek the AI vendor's identity, privacy-impact assessment and Algorithm Charter compliance detail, none of which the Ministry had proactively published.

Implementation
Indicative cost
Low (< €50k) — No budget figures published for the AI/LLM classification work or the Public Voice engagement.
Time to results
Short (< 1 year) — Submissions processed November 2024 to January 2025; public scrutiny and OIA disclosure continued into mid-2025.
Staffing & skills
Ministry for Regulation staff (manual review of subsets), Public Voice (external research organisation, LLM classification)

Conditions for success

  • a logic model designed by the Ministry to guide submission classification
  • manual review of a subset for thematic analysis and quality assurance

Common failure modes

  • the AI vendor's identity, privacy-impact assessment and Algorithm Charter compliance were not proactively published and required a formal OIA request from a civil-liberties group
  • a disputed 'bot-generated' claim by a minister undermined public trust
  • MPs and academics warned the process risks undermining confidence in the democratic process
  • reports emerged of thousands of submissions not read by the ministry

Where it fits

Governance type
national government ministry, public consultation
Scale
single legislative consultation, 22,821 submissions
Income level
high-income (New Zealand)

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful