What information does generative AI surface when New Zealanders research election policy, and which sources influence the answers?
CiteCompass has launched a 12-week independent longitudinal study examining how leading generative AI platforms represent and cite New Zealand’s parliamentary political parties and their publicly available policies.
As New Zealand heads toward the 2026 General Election on Saturday 7 November 2026, the way people discover information is changing.
The research journey is no longer always: Question → Google search → Website → Answer
Increasingly, it can look like: Question → AI Search → Synthesised Answer → AI-Cited Sources
That shift creates an important new research question:
When New Zealanders use generative AI to research major election issues, what political information do AI systems surface — and where does that information come from?
The CiteCompass NZ Election 2026 AI Visibility Monitor has been established to observe that information environment over time.
Research commenced: 25 August 2026
Study duration: 12 weeks
Methodology version: 1.0
Research status: Week 1 — study launch and baseline establishment
Why This Research Matters
Generative AI is becoming an increasingly important layer between a question and the information people ultimately see. Instead of presenting a list of links, AI systems can synthesise information from multiple sources into a single answer.
That raises questions that matter well beyond politics:
- Which organisations and sources are visible to generative AI?
- Which sources are cited or relied upon?
- Does authoritative first-party information reach AI-generated answers?
- Do different AI platforms represent the same subject differently?
- How stable are those answers as the underlying information environment changes?
- What happens when new information, media coverage and policy material enter the public domain?
A national election provides a dynamic, highly visible environment in which to observe these changes longitudinally. The purpose of this research is to study how AI-mediated information discovery behaves, not to assess the merits of political policies.
What CiteCompass Is Monitoring
The study monitors the six parliamentary political parties represented in Parliament at the commencement of the research across the same five policy domains.
What We Measure
CiteCompass will monitor quantitative and descriptive measures including:
- AI presence — whether a party or its policy information appears in relevant AI-generated responses;
- AI share of voice — relative visibility within defined research activity;
- answer prominence — how prominently relevant information appears;
- policy visibility — the extent to which published policy information is surfaced;
- citation share — which sources appear in AI citations and references;
- official-source penetration — whether authoritative party policy material reaches AI-generated answers;
- third-party source influence — the role of media, government, institutional and other external sources;
- differences between AI platforms — whether different systems produce materially different information environments; and
- change over time — how visibility, citations and source influence evolve during the 12-week study.
The research focuses on observable AI behaviour and source visibility. It does not evaluate whether a policy is good, bad, effective or preferable.
A Longitudinal Observational Study
This research is intentionally designed as a longitudinal observational study. The election information environment is dynamic. During the research period:
- political parties may publish or update policies;
- media coverage will change;
- political events will occur;
- online information will change;
- AI systems and models may change;
- the sources available to AI systems may change; and
- the information people seek may change.
CiteCompass therefore monitors how AI visibility, responses, citations and sources evolve under a consistently applied research framework.
Dynamic Research Queries
CiteCompass uses its existing platform architecture to generate research queries based on the relevant profile and information need.
- Individual research queries are not artificially frozen for the entire 12 weeks.
- They may change over time to model changing voter information needs within an evolving election information environment.
- This is not intended to represent polling or direct measurement of voter sentiment.
The important control is equivalent process.
All six parties are monitored using the same:
- five policy domains;
- research architecture;
- profile framework;
- dynamic-query methodology;
- AI-platform framework;
- monitoring cadence;
- measurement framework;
- data-retention process;
- QA process; and
- reporting methodology.
Differences in the resulting AI responses, sources and citations are therefore observations generated by the study — not adjustments made to favour a political party.
Political Independence
CiteCompass is politically independent.
This study does not assess which political party has the best policies.
- It does not recommend how New Zealanders should vote.
- It does not predict the election result.
- It does not rank politicians.
- It does not claim that AI visibility determines voting behaviour.
The study has one central purpose:
To measure how publicly available political information is represented, surfaced, cited and sourced within generative AI systems.
CiteCompass will report observed results even where:
- one party records a higher result than another;
- the outcome changes from week to week;
- a finding is unexpected;
- a result is politically inconvenient; or
- there is no meaningful difference between parties.
Our governing principle is simple:
Observe. Measure. Report. Do not advocate.
How Results Will Be Reported
Where comparative results are published, CiteCompass will use neutral, descriptive research language.
We will report what was:
- measured, observed, surfaced, cited, appeared, increased, decreased or changed.
- We will not turn visibility data into political conclusions such as a party having:
- won, lost, the best policy, the worst policy, the strongest policy or the weakest policy.
Where practical, if a publication compares parties using a particular metric, the equivalent result for all six monitored parties will also be made available.
Scheduled valid research runs will not be discarded or repeated simply because a result appears surprising or politically inconvenient.
Research Integrity and Auditability
CiteCompass is retaining sufficient research evidence to reconstruct published findings.
The research record includes, where applicable:
Date → Party → Policy Area → Research Profile → Generated Query → AI Platform → AI Response → Mentions → Sources → Citations → Official-Source Presence → CiteCompass Metrics
We will retain relevant:
- generated research queries;
- AI responses;
- timestamps;
- source URLs;
- metric calculations;
- platform/model information where available;
- technical exclusions;
- corrections;
- methodology changes; and
- published results.
This provides an audit trail for the longitudinal research.
Factual Accuracy and Corrections
Generative AI systems can produce inaccurate or unsupported statements.
CiteCompass will not knowingly republish a materially adverse AI-generated assertion about an identifiable person as established fact without appropriate verification.
Where an AI error is itself relevant to the research, it may be documented as an observation of AI behaviour while clearly identifying that CiteCompass was unable to verify the assertion.
All six monitored political parties will be given equivalent opportunity to identify objective factual matters such as:
- an incorrect official policy URL;
- superseded policy information;
- incorrectly attributed material; or
- an objective factual error in CiteCompass’s description of published policy information.
Parties will not be able to alter the research methodology, prompts, metrics, weighting, interpretation or comparative results.
Material corrections will be logged.
Commercial Independence During the Study
CiteCompass is a commercial technology platform, but the political research is being conducted independently.
For the duration of the programme, CiteCompass will not provide AI-visibility optimisation services to any political party or candidate monitored by the study.
This separates the measurement of political AI visibility from any commercial activity intended to improve that visibility.
The broader commercial question demonstrated by the research is generic:
Organisations increasingly need to understand whether generative AI knows about them, how it describes them, which sources it relies upon and whether their own authoritative information is being surfaced.
Individual political-party results will not be used as product endorsements.
Publication and Electoral Compliance
CiteCompass has proactively requested an advisory opinion from the New Zealand Electoral Commission regarding the proposed research publication framework and associated unpaid social-media communications.
The purpose of seeking that guidance is to operate conservatively and transparently during the regulated election period.
Until that process is sufficiently clear, CiteCompass will distinguish between:
- announcing and explaining the research programme; and
- publishing substantive comparative political findings.
CiteCompass does not intend to purchase paid or boosted social-media promotion for individual political findings.
Distribution is intended to occur through:
- the CiteCompass website;
- unpaid organic social-media communications;
- voluntary sharing by readers;
- earned media coverage; and
- independent references or links from third parties.
The 2026 General Election regulated period runs from 7 August to 6 November 2026, with Election Day on Saturday 7 November 2026.
CiteCompass will apply an Election Day publication blackout to election-related promotional activity in accordance with applicable Electoral Commission guidance.
View Electoral Commission election advertising guidance
What Happens Next?
Week 1 — Launch and Baseline Establishment
During Week 1, CiteCompass is:
- establishing the research profiles;
- confirming relevant authoritative policy sources;
- applying the common monitoring framework;
- establishing the baseline dataset;
- documenting Methodology v1.0; and
- completing publication and QA controls.
Week 2 — Baseline Research
Subject to the applicable publication framework, the first research release is intended to examine:
What policy information and sources are generative AI platforms surfacing for NZ Election 2026?
Further releases are expected to examine:
- source influence;
- official-source penetration;
- differences between AI platforms;
- visibility across major policy domains;
- changes from the baseline;
- longitudinal volatility;
- a midpoint analysis; and
- a final 12-week research report.
Frequently Asked Questions
1. Is CiteCompass conducting political polling?
No. The study does not survey voters, predict voting intention or claim to measure real-world voter sentiment. It observes how generative AI systems respond to dynamically generated research queries designed to model changing information needs.
2. Is CiteCompass ranking the political parties?
No. The study may report that one party has a higher or lower measured AI-visibility result for a defined metric, but it will not interpret that result as meaning the party has a better policy or deserves electoral support.
3. Are all political parties treated the same way?
Yes. All six parliamentary parties included in the study are subject to an equivalent research architecture, policy framework, measurement approach, monitoring cadence, QA process and publication standards.
4. Are the research questions exactly the same every week?
No. CiteCompass uses dynamic research queries because the information environment changes throughout the election period. The methodology is designed to observe that changing environment rather than create a laboratory experiment in which every question is frozen.
5. Does CiteCompass work for any of the political parties being monitored?
CiteCompass will not provide AI-visibility optimisation services to any monitored political party or candidate for the duration of the research programme.
6. Will CiteCompass publish the underlying methodology?
CiteCompass will publish sufficient methodology to explain the research framework, enable informed interpretation of the results and support transparency.
Proprietary platform architecture, scoring logic and implementation methods that are not necessary to interpret the research may remain confidential.
7. Where will the findings be published?
The principal research record will be published here on CiteCompass.com, with unpaid social-media communications used to alert interested readers when new research is available.
Follow the Research
The CiteCompass NZ Election 2026 AI Visibility Monitor will develop over the full 12-week research period. Follow CiteCompass for upcoming baseline findings, weekly observations, source analysis, AI-platform comparisons and the final longitudinal report.
Next planned research release:
Week 2 — NZ Election 2026 AI Visibility Baseline
Media and Research Enquiries
Journalists, researchers and other interested parties are welcome to contact CiteCompass for background on the research framework and published findings.
Media enquiries: [email protected]
Press Release: https://citecompass.com/citecompass-launches-nz-election-2026-ai-visibility-study/
Research enquiries: [email protected]
Website: https://citecompass.com
About CiteCompass
CiteCompass is an AI visibility platform that helps organisations understand how they are represented within generative AI systems, which sources influence AI-generated answers, and whether their authoritative information is being surfaced and cited.
The NZ Election 2026 AI Visibility Monitor applies that measurement approach to a dynamic public information environment in order to better understand how generative AI discovers, synthesises and presents political policy information.
Research launched: 25 August 2026
Methodology version: 1.0 – 25 August 2026
This page describes the research programme and methodology. It does not constitute an endorsement of, or opposition to, any political party or candidate.