How we ran this study

We wanted to know what ChatGPT, Gemini, Google AI Mode and Google AI Overviews say when a voter asks them about the 2026 New Zealand general election: how they explain the election, which parties they bring up, how they compare policies, where their information comes from, and whether what they say is right.

The study is neutral. No party is the focus, every registered party is treated the same way, and we take no view on any party or policy. We report what the assistants said.

Rather than asking general questions, we wrote each question as a particular person would ask it. We described 23 households by their circumstances: their housing, work, income, caring responsibilities and access to services. A retired homeowner and a retired renter face different questions, and so do a café owner and an exporting manufacturer. We added nine people in particular places, from a West Auckland renter to a Queenstown hospitality worker.

Each voter's priorities are our design assumptions for that situation. They are not findings about everyone in that group.

Every question was asked as a standalone conversation, so we captured the first answer a voter would get, including any offer to look further. The four assistants received identical wording.

The 31 local questions were asked twice: once with no location, and once from the voter's own town. Wellington stood in for Lower Hutt, the closest setting available. That gives 205 question-and-place combinations, and 820 planned answers.

NZBrands.ai runs its studies on Onsomble, the platform that asks AI assistants questions at scale and records what they say. We set the study up as a topic with no focal entity, so no party was singled out, and listed every party on the Electoral Commission's register as an entity to track. Onsomble asked each question through the assistants' ordinary consumer apps, saved every answer in full, and recorded for each answer the claims it made, which parties each claim was about, and the websites the assistant drew on.

We downloaded the complete record after the run and built every figure on this site from it. The scripts that do so are in the site's code, so the tables can be rebuilt from the saved files without running the study again.

Every ChatGPT, Gemini and AI Mode answer was saved. Five AI Overviews were not: four for the overseas voter considering a return, and one shared energy question. The record does not say whether Google showed no overview or the capture failed. We have not filled the gaps.

All 31 local questions came back from both settings on all four assistants, giving 124 pairs of answers that differ only in where they were asked from.

Assistants are compared on the 200 question-and-place combinations where all four answered, 800 answers in all. The five incomplete combinations are left out of every assistant's comparison, so a missing answer never changes which questions are compared. Whole-study counts, such as how many answers name a party, use all 815.

Places are compared only within the 124 pairs where the question was identical. We never rank towns against each other on different questions.

Voting guidance was read in full for all 32 answers to the eight questions that ask for a recommendation. Each of the 32 voter situations has a written analysis of its main question and selected others.

Each question was asked once. AI assistants vary their answers, so a repeat run would differ in places. The town comparison shows that answers differed; proving that location caused the difference would need repeat runs and verified access locations.

The voters are designed situations, not a sample of New Zealanders. The study shows what the assistants said and lets anyone check it. It does not show whether anyone changed their vote, or why the assistants answered as they did.

We read all 32 voting-guidance answers and the case-study questions in full, and checked 17 statements. We have not read all 815 answers in full or checked every policy claim.

The archive holds all 815 answers, the 174 questions with their tags, the 23,108 claims Onsomble extracted, the sources, the original requests and retrieval times, and file fingerprints so anyone can check the evidence has not changed since collection.