How we ran this study

We asked ChatGPT, Gemini, Google AI Mode and AI Overviews about the 2026 New Zealand general election. We counted party names, looked at how answers connected policies to voters’ priorities and checked selected factual statements.

This is Edition 8. We take no view on any party or policy, and no party pays to appear.

These 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.

We asked all 174 questions nationwide, then repeated the 31 regional questions from their relevant locations. That gave 205 planned answers from each assistant and 820 in total.

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.

We 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.

Google did not produce an AI Overview for 30 requests, and one ChatGPT request failed. These gaps are not counted as answers or as refusals to discuss politics.

All included answers were collected on 2 November, with no later replacement answers added.

We could compare 782 answers from each edition, collected on 31 October and 2 November. They answer the same questions from the same assistants and locations: 201 ChatGPT, 205 Gemini, 204 Google AI Mode and 172 AI Overviews answers per edition. Seven current answers and one earlier answer had no counterpart.

Read Edition 7 for all 783 answers and its original party counts. The before-and-after charts in this edition use only answers available for the same question, assistant and location on both dates.

The source comparison covers 782 answers from each edition. Citations appeared in 780 current answers and 779 earlier answers. A website counts once per answer, however many links point to it.

To compare the four assistants fairly, we used only questions and locations with an answer from every assistant on both dates. Each assistant’s chart therefore covers 168 answers per edition, or 672 across the four assistants.

The location comparison puts each question’s nationwide answer alongside the answer from its location. The party names differed in 79 of 122 current comparisons. Among the same 116 comparisons available on both dates, the figures were 75 previously and 77 now. One answer from each setting cannot show whether location caused the difference.

The voting-guidance comparison covers 31 answers from each edition. A preferred shortlist puts one party or a smaller group ahead overall as options to investigate. Comparing parties or suggesting one for a particular priority does not, by itself, count as a preferred shortlist. A shortlist is different from an unconditional instruction to vote for a party.

Each voter page keeps every current answer available to read. Its before-and-after topic table and source comparison use only answers available from the same questions, assistants and locations on both dates.

We read all 31 voting-guidance answers and Gemini’s nationwide answer for each of the 32 voter situations on both dates. We also read current answers selected for ambiguous party names and factual checks.

Nine statements were supported and 12 were wrong. Four needed an important qualification and four could not be verified; these eight are labelled unclear.

The original records cover 44 selected instances, including 15 where a statement from the earlier edition did not appear again. Leaving out an earlier error does not show that the assistant corrected it.

The four unresolved checks concern student-debt attribution, an electorate reference, welfare legislation and carer-allowance eligibility. They are not counted as verified errors.

The checks are dated 2 November 2026. A verdict applies to the selected statement rather than the entire answer, and does not give an accuracy score for all 789 answers.

AI 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 789 answers were included in the party-name counts. Detailed readings covered 165 distinct current answers and 63 earlier answers and were assisted by AI; this was not a factual review of every answer or an independent human 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.

The source rankings identify hosting websites, rather than the authors of every page or post. A Facebook citation can point to material posted there by another organisation, and citation frequency does not establish accuracy, quality or political bias.

Twenty-seven of the 789 current answers cited earlier NZBrands study pages. We record those links alongside the other sources without treating a citation as proof of influence.

Download the tables behind the findings, or open a question to read every saved answer in full.

Each earlier edition retains its own findings, voter pages and original answers. Read older answers in the context of their collection dates.