---
title: "How we ran this study"
publisher: "Onsomble Limited"
publication: "NZBrands.ai"
edition: 1
collected_on: "2026-10-11"
html_url: "https://nzbrands.ai/research/insurance/home-contents-insurance/method/"
---

# How we ran this study

Edition 1 · answers collected 11 October 2026.

[Read on NZBrands.ai](https://nzbrands.ai/research/insurance/home-contents-insurance/method/)

What we asked about home and contents insurance, the consumer situations and regions we covered, and how we counted recommendations, sentiment, sources and checked statements.

## Study figures

| Measure | Value |
| --- | --- |
| questions | 372 |
| consumer situations | 12 |
| AI assistants | 4 |
| answers saved | 1,476 |
| collected | 11 Oct 2026 |

## What we set out to learn

What AI tells a New Zealander choosing home or contents insurance or seeking help with an existing policy or claim.

We compared ChatGPT, Gemini, Google AI Mode and Google AI Overviews: which insurers they recommend, how they explain cover and trade-offs, what help they offer after purchase, where their information comes from, and whether selected factual statements are right.

Home insurance concerns the building. Contents insurance concerns belongings. Each question has one product tag, and the product-specific results remain separate throughout the study.

[What a study is for](https://nzbrands.ai/methodology#purpose)

## Who we asked as and what we asked

Twelve designed consumer situations, sixteen named regions and 372 exact questions: 189 home and 183 contents.

The situations include an everyday household, a first-time buyer, someone under budget pressure, a rural household, a household with valuable assets and someone who prefers help over the phone. They are researcher-authored situations, rather than a population sample.

Questions cover understanding the need for insurance, exploring options, choosing cover, comparing insurers, checking the details, buying and switching, and ongoing support. These groups do not track real people over time. Buying factors describe question priorities and answer reasoning, rather than how common a preference is among consumers.

Regional questions name where the consumer lives. Other questions describe claims, policy decisions and particular priorities. The evidence retains the exact saved wording.

[How we write questions](https://nzbrands.ai/methodology#questions)

## How we collected the answers

1,476 of 1,488 planned answers were saved. Twelve captures are missing: nine AI Mode and three AI Overviews.

Onsomble collected the answers through the four consumer applications. Collection completed on 11 October 2026 at 7:48 pm New Zealand daylight time. Every question was planned for each application.

All collection settings were New Zealand. Region names appear in the questions; these are not captures taken from sixteen different regional access locations. Failed executions do not enter the collected-answer denominator.

Planned and saved answers

| Assistant | Planned | Saved | Missing |
| --- | --- | --- | --- |
| ChatGPT | 372 | 372 | 0 |
| Gemini | 372 | 372 | 0 |
| Google AI Mode | 372 | 363 | 9 |
| Google AI Overviews | 372 | 369 | 3 |
| Total | 1,488 | 1,476 | 12 |

[The assistants we cover](https://nzbrands.ai/methodology#assistants)

## How we count insurer recommendations

An insurer counts when an answer favourably includes it as an option, once per insurer per answer.

Conditional recommendations count with their conditions retained. An insurer in a shortlist need not be the first or strongest recommendation. Ties remain ties. Collected answers with no qualifying provider stay in the denominator.

Reviewed provider roles distinguish insurers and customer-facing brands from brokers, underwriting groups, citations, neutral references and escalation organisations. Bank insurance brands remain distinct distribution options when offered as choices. AA Roadservice is distinct from AA Insurance. Incidental names do not count as recommendations.

| Term | Definition |
| --- | --- |
| Recommended answers | Distinct collected answers that favourably include the insurer for the requested task. Several insurers can count in one answer. |
| Recommendation percentage | Recommended answers divided by collected answers in the relevant comparison group. This measures shortlist inclusion, rather than market share, quotation prices or measured claims performance. |

[Counting rules and definitions](https://nzbrands.ai/methodology#definitions)

## How we distinguish shopping from ongoing support

The actual-support group contains 71 answers to 18 questions across eleven consumer situations.

The questions concern an existing policy or claim: renewal, moving home, water and fire damage, replacement value, reopening a flood claim and getting practical help or an external review.

None of these answers prefers a named insurer to provide support. Provider references, service examples and alternatives for a future renewal are recorded separately. A renewal shortlist does not become a recommendation for another insurer to handle an existing claim.

The rural consumer has no eligible home or contents support question in this group. The digital-first storm-photo question is tagged contents but refers to a home insurer and home claim; its wording and original product tag are preserved.

[Counting rules and definitions](https://nzbrands.ai/methodology#definitions)

## How we compare consumers, platforms and regions

Each comparison uses its own specified questions, products and platforms.

Consumer selection contains 94 collected answers to 24 product-and-situation questions: 47 home and 47 contents. Most questions have four answers. The digital-first home question and family contents question have three because AI Mode did not collect them. Guidance, benchmarks, priority variants and support do not increase this denominator.

The consumer platform comparison uses 22 identical selection questions answered by all four applications: eleven home and eleven contents questions per platform. The two incomplete questions are omitted from all four platforms.

There are 127 observed regional selection answers. Equal regional comparisons use 112: four home and three contents answers in every region. AI Mode contents selection is omitted from every region because the Hawke’s Bay answer is missing. Home and contents grids retain their separate four-answer and three-answer denominators.

The separate regional platform comparison uses 31 identical questions per platform, 124 answers in total. Additional regional scenarios have different facts and inform interpretation rather than equal insurer rankings.

Comparison groups and denominators

| Group | Questions | Saved answers | Comparison unit |
| --- | --- | --- | --- |
| Consumer selection | 24 | 94 | 47 home and 47 contents answers |
| Consumer platform comparison | 22 | 88 | 22 identical questions per platform |
| Actual consumer support | 18 | 71 | Eligible questions across 11 situations |
| Equal regional comparison | 32 | 112 | 4 home and 3 contents answers per region |
| Regional platform comparison | 31 | 124 | 31 identical questions per platform |

[Comparison rules](https://nzbrands.ai/methodology#comparing)

## What the sentiment bars measure

The number measures recommendation inclusion. The bar shows the stance of claims about the insurer.

The insurer sentiment dataset contains 3,828 saved claims in consumer and regional selection answers. Each record is a unique claim about one insurer in one answer, with an exact passage and its original-text offsets.

Bars use saved native claim stances, rather than inferring sentiment from recommendation counts. Positive, mixed, neutral and negative claims remain distinct. Different claims in one answer can express different stances about the same insurer.

Each grid uses only claims from its own product and answer membership. A matrix cell opens the sentiment breakdown and original passages; the complete captured answer remains available.

[Counting rules and definitions](https://nzbrands.ai/methodology#definitions)

## How we count sources

A website counts once per answer. Recorded sources and visible links are separate measures.

The source archive contains 13,048 answer, domain and URL associations and indexes all 1,476 answers, including those without recorded sources.

A domain counts once for each answer associated with it, however often it appears. An answer can contribute to several websites. Publisher types identify insurer websites, comparison tools, government, consumer groups, news and other websites.

“Had sources” counts answers with recorded source associations. “Showed you a link” counts answers containing a web URL in their captured text. An associated source need not appear as a literal link in the answer.

[Counting rules and definitions](https://nzbrands.ai/methodology#definitions)

## How we checked factual statements

Fifty factual topics in 28 matched questions were reviewed across 112 complete answers.

The audit checked the selected topics across the four platforms, producing 200 claim-and-platform cells. Primary sources include policy wordings, insurers, award publishers, statutory bodies and legislation. The audit was completed on 12 October 2026, with source dates and policy versions retained.

Seventy-seven statements were supported, five were inaccurate, eleven missed a material qualification and four remained unresolved. The other 103 cells did not state the selected assertion or a direct conflicting version. These are statement-and-platform counts; several checked claims can occur in one answer.

The accuracy section shows the 97 stated assertions: 77 correct, five wrong and fifteen unclear. Unclear includes missing material qualifications and unresolved checks. Omitted facts are not counted as incorrect answers.

Claims were selected for their consequences for insurer choice, cover, payouts and next steps. This is not a random sample of every statement and does not estimate accuracy across all 1,476 answers. Original recommendation counts retain what the applications said even when an accompanying factual claim is wrong.

[Limitations](https://nzbrands.ai/methodology#limitations)

[Read the checked statements](https://nzbrands.ai/research/insurance/home-contents-insurance#accuracy)

## What this can and cannot tell you

The results describe saved answers to designed questions collected for this study.

Consumer situations are not a population sample. Regional differences describe answers to region-named questions collected with a New Zealand setting.

Recommendation frequency does not measure prices, consumer uptake or claims performance. The factual audit covers selected statements. The original questions, full answers and reviewed comparison memberships are available on the evidence page.

[Limitations](https://nzbrands.ai/methodology#limitations)

[Evidence guide (Markdown)](https://nzbrands.ai/research/insurance/home-contents-insurance/evidence.md)

## Publication and corrections

Published and funded by Onsomble Limited.

[Independence](https://nzbrands.ai/about/independence/) · [Corrections](https://nzbrands.ai/about/corrections/) · [Contact](https://nzbrands.ai/about/contact/)