Methodology · NZ Election 2026 · Edition 6
How we ran this study
We repeated the election questions on 26 October 2026 and compared the answers with 23 October. This page explains what we asked, what came back and how we compared party names, sources and selected statements.
- questions
- 174
- voter profiles
- 32
- AI assistants
- 4
- answers recorded
- 783
- collected
- 26 Oct 2026
01
What we set out to learn
What AI tells a New Zealander asking about the election, and what changed when we asked again.
In the general method:The purpose of a studyWe asked ChatGPT, Gemini, Google AI Mode and AI Overviews about the 2026 New Zealand general election. We looked at the parties they named, how they connected policies to voters’ priorities and selected factual statements.
This is Edition 6. We take no view on any party or policy, and no party pays to appear.
02
Who we asked as
The questions describe 23 household situations and nine people in particular locations.
In the general method:Writing the questionsThese are made-up situations designed for the study, rather than a sample of real voters. They cover housing, work, income, caring responsibilities and access to services.
The priorities written into a question belong to that situation. They are not assumptions about everyone with the same background.
| Question set | Questions | Purpose |
|---|---|---|
| Household situations | 122 | Practical concerns of the 23 households |
| Shared national questions | 21 | Election questions without a particular voter described |
| Regional situations | 31 | Local concerns of the nine regional voters |
| Total | 174 | 20 topics across 32 voter situations |
03
How the questions were asked
The same words, asked separately, through four consumer AI assistants.
In the general method:The assistants we coverEach question was asked as a standalone conversation. Regional questions were asked nationally and from the relevant location, giving 205 question-and-location combinations for each assistant and 820 planned answers.
The nine collection locations were Ashburton, Auckland, Dunedin, Hastings, New Plymouth, Queenstown, Tauranga, Wellington and Whangārei. The regional situations also describe their local circumstances in the question.
04
How we used Onsomble
Onsomble asked the questions and saved the complete answers.
In the general method:Working with OnsombleWe used Onsomble to repeat the saved question set through the consumer versions of these four services. We checked that the wording and locations were the same before comparing the dates.
05
What came back
We saved 783 of the 820 planned answers on 26 October 2026.
Two planned ChatGPT answers and 35 planned AI Overviews answers were not saved. The records do not give an individual reason for each missing answer, so we do not treat these gaps as refusals or answers naming no party.
The collection finished on 26 October. No later replacement answers are included.
| AI assistant | Planned | Recorded |
|---|---|---|
| ChatGPT | 205 | 203 |
| Gemini | 205 | 205 |
| Google AI Mode | 205 | 205 |
| AI Overviews | 205 | 170 |
| Total | 820 | 783 |
06
How we count
The figures count party names in answers, rather than recommendations to vote for them.
In the general method:Counting rules- Answer
- The complete saved response to one question, from one assistant and one location.
- Party named
- A party counts once per answer when an unambiguous name or alias appears. Repeated names do not add to the count. We exclude web addresses but include visible citation labels and follow-up text.
- Names with other meanings
- We check context when a name can mean something else. National counts when it names the party; a reference to the National Library does not. The same principle applies to other party names.
- The seven displayed parties
- National, Labour, Greens, ACT, NZ First, Te Pāti Māori and The Opportunity Party. The downloads retain all 17 tracked parties.
- Tone
- The coloured bars summarise Onsomble’s saved assessments of statements about each party as positive, neutral, negative or mixed. Only statements about the party as their subject contribute to its bar. Strongly positive and negative assessments join their corresponding groups. These are automated assessments of how a party is portrayed, rather than factual verdicts or a measure of which party the assistant prefers. Supporting extracts are shown only when their wording appears in the original answer.
- Sources
- Source records are available for all 783 saved answers, including three with no recorded citations. A cited website counts once per answer. Other references stay in the downloads and are excluded from the citation rankings. Visible links in the answer are counted separately. A recorded citation does not establish that it supports every statement.
- Kinds of website
- Websites are grouped as party, government, news, comparison, advocacy, business, social or other. Other includes websites that do not fit a more specific supported classification, including websites we have not yet classified.
07
How we compared the answers
The main comparison uses the same questions, assistants and locations on 23 and 26 October.
In the general method:Making comparisonsComparisons use the 780 answers available on both dates: 202 ChatGPT, 205 Gemini, 205 Google AI Mode and 168 AI Overviews answers. Three current answers and six earlier answers are outside these comparisons.
All 786 answers and their party-name decisions from Edition 5 are preserved. Only the 780 with a current counterpart enter the two-date comparison.
The source comparison uses those same 780 answers per date. Citations were recorded in 777 of the current answers and 778 of the earlier answers.
The assistant comparison uses the same 166 question-and-location combinations for each assistant on both dates, giving 664 answers per date.
The location comparison uses 118 complete national-and-regional answer pairs from 26 October. It describes differences in those answers and does not establish that location caused them.
The guidance findings come from 31 answers read in full, alongside 32 selected national answers read in greater detail. Earlier recorded readings were reused after their text was checked. Each voter page’s extracts open the original answer from the date shown.
08
What we checked
We revisited 11 selected checks covering ten subjects.
Five assertions were stated in the new answers and six were not repeated. Of the five statements checked against primary sources on 27 October, three were incorrect and two were supported.
The six assertions not repeated remain in the download as absent, rather than being presented as corrections or current errors. A verdict applies to the statement checked, rather than the whole answer.
These selected checks do not estimate the accuracy of all 783 answers. The original research date and sources remain with each check.
09
What this can and cannot tell you
This is a dated record of answers, rather than a prediction of what every voter will hear.
In the general method:LimitationsAI assistants can answer the same question differently. One answer from each setting cannot show whether a change is lasting or whether location caused it. This study does not measure influence on votes or establish political bias.
All 783 answers were included in the party-name counts. The detailed readings covered 63 answers and were assisted by AI; they do not amount to a full factual review of all answers or an independent second review.
The study does not verify every policy statement or establish anyone’s personal entitlement. An earlier error disappearing from a later answer is not, by itself, a correction.
10
The records
Every saved question and answer stays available with the tables behind this edition.
The evidence pages include all 174 questions and 783 complete answers, recorded readings, source lists, sentiment assessments and selected checks. Downloads preserve the original research separately from these reader summaries.
Each earlier edition retains its own findings, voter pages and original answers. Read older answers in the context of their collection dates.