How this works
Transparency is core to this project. Here is exactly how the NZ Election Policy Checker works, where the data comes from, and what the limitations are.
The scene you see on the homepage is a stylised New Zealand Beehive. Each party stands in front of its own patch of lawn and holds up a banner with its short response. The Benevolent Dictator AI stands on top of the Beehive to make clear it is not one of the parties. Click any banner to read the full response and the source it draws from.
Where the policy data comes from
Party positions are sourced from each party’s officially published manifesto and policy pages. Every response carries a link back to that party’s own policy page, so you can check the source directly.
To keep answers current during a live campaign, a scheduled job runs every night at 3am NZT. It scans the previous 24 hours of New Zealand political news from RNZ, Stuff, NZ Herald, Newsroom, and beehive.govt.nz, and summarises any party-relevant items into a briefing. That briefing is committed to the project’s public git history, so anyone can see exactly what news informed a given day’s answers. The footer shows the date of the most recent scan.
We do not editorialise, rank, or score party positions. Each party’s response is a factual summary of what they have publicly stated, with recent news added as context only where it clarifies or updates that position.
Policy positions change during election campaigns. We aim to reflect the latest, but always check the original party sources for the most current information.
The Benevolent Dictator AI
The eighth panel is not a political party. It is an AI model (Claude, by Anthropic) with a single mandate: suggest evidence-based policy approaches that would deliver the greatest good for the greatest number of New Zealanders. On the homepage it stands on top of the Beehive, visually apart from the seven parties on the lawn, to reinforce that it is a separate kind of voice.
Its responses are generated in real time based on peer-reviewed research, established economic principles, public health data, environmental science, and international policy examples. It is instructed to present trade-offs honestly and acknowledge uncertainty where evidence is unclear.
Important limitations:
- AI models can produce errors and may reflect biases present in their training data.
- Research has shown that large language models, including Claude, can be perceived as having a left-leaning bias on political topics. We have instructed the AI to present trade-offs rather than recommendations, but this is not foolproof.
- AI analysis is not a substitute for your own judgement, values, and priorities.
- The AI has no access to information published after its training data cutoff and may not reflect the very latest policy developments.
How the alignment matching works
When you give a thumbs-up to a party’s policy response, that counts as one point of agreement. The results page shows what percentage of your thumbs-up votes went to each party.
This is a simplified measure. It treats every policy topic as equally important, which may not match your priorities. It also only captures agreement, not the strength of your agreement.
Your results are stored locally in your browser. We do not collect, store, or transmit your voting choices or results.
Our neutrality commitment
This tool is built by Ten Past Tomorrow (an AI consultancy) and Payper (a PR agency). Neither organisation is affiliated with any political party.
Our goal is to help New Zealanders make more informed voting decisions. We are not telling you how to vote.
If you believe any policy summary is inaccurate or unfair, please contact us at mark@tenpasttomorrow.com.
Privacy
This tool does not require you to log in or provide any personal information. Your preferences and results are stored only in your browser and are never sent to our servers or any third party.
When you ask a question, it is sent to our server to retrieve policy data and generate AI responses. We do not log or store your questions.