---
title: "Immigration study: methodology"
publisher: "Onsomble Limited"
publication: "NZBrands.ai"
edition: 1
collected_on: "2026-10-15"
html_url: "https://nzbrands.ai/research/hospitality-tourism/immigration/method/"
---

# Immigration study: methodology

Edition 1 · answers collected 15–20 October 2026.

[Read on NZBrands.ai](https://nzbrands.ai/research/hospitality-tourism/immigration/method/)

How we chose the questions, collected the answers, counted the results and checked selected advice.

## Study figures

| Measure | Value |
| --- | --- |
| questions | 143 |
| captured answers | 797 |
| missing responses | 27 |
| answers collected | 15–20 October 2026 |

## The people and questions

We asked 143 questions for nine migrant profiles.

Researchers wrote the people, circumstances and priorities. The profiles test how advice changes when the facts change; the profiles do not represent the size or preferences of real migrant groups. The study does not follow households over time.

Questions cover whether to move, drawbacks, visa options, locations to live, work and study, money, settling in and preparation. Each topic was asked in at least two ways. Questions about where to live were asked in three ways. Work, study and business questions apply to eight profiles; settling-in questions apply to five.

[Read all questions](https://nzbrands.ai/research/hospitality-tourism/immigration/questions)

## Available and missing answers

We collected 797 of 824 planned answers, including every answer using the New Zealand setting.

Answers were collected from 15 to 20 October 2026 and downloaded on 21 October. No further collection is underway.

All 27 missing answers are from the nine China questions about where to live on ChatGPT, Gemini and Google AI Overviews. A missing answer is not counted as advice against a place.

Answers collected

| Question group | Planned | Collected | Missing |
| --- | --- | --- | --- |
| Main questions using the NZ setting | 360 | 360 | 0 |
| Additional ways of asking where to live | 72 | 72 | 0 |
| Where-to-live questions across eight countries | 288 | 261 | 27 |
| Questions about work, study and settling in | 104 | 104 | 0 |
| Total | 824 | 797 | 27 |

## How we counted locations

A place counts once when the name appears in an answer.

The place comparison uses 108 New Zealand answers about where to live: twelve for each profile, from three question versions on four platforms. Names found only inside web addresses do not count. Cities and surrounding regions are counted separately.

A place named in a warning, rejected option or comparison still counts. The figures do not show the best places to live or how often a place was a first choice. Cambridge is excluded because student answers also use Cambridge as the name of an education qualification.

Auckland, Wellington and Christchurch appear in all twelve answers for the technology professional. Waikato, Taranaki, Canterbury and Southland appear in all twelve answers for the dairy family.

## How we compared countries

The country setting used to collect an answer is not the applicant’s citizenship.

The same nine questions about where to live were asked using eight country settings. Comparisons across all four platforms use New Zealand, India, the Philippines, the United Kingdom, South Africa, Fiji and Australia, with 36 answers per country.

China has nine answers from Google AI Mode only. The eight-country comparison therefore uses Google AI Mode alone. Differences in one round of research do not establish that geography caused the differences or that the answers will stay the same.

## How we assessed tone toward places

Every named place received one label in each of the 108 answers about where to live.

We read all 108 New Zealand answers about where to live and assessed the tone toward each named place. Positive means benefits without drawbacks; negative means drawbacks without benefits; mixed means both; neutral means a factual or incidental mention. Each place receives one label per answer. The labels describe AI wording, not verified destination quality.

The review covers 630 place mentions across the ten cities and regions in the comparison. A name found only in a university, hospital or source title is neutral unless the answer also evaluates the place. General visa or job-search cautions do not make every place negative or mixed. The original appearance counts remain unchanged.

The coloured bar divides answers naming a place into the four tone categories. Supporting passages and the reason for every label are available in the download. The review was completed on 27 October 2026 using the original answers; no new answers were collected.

[Download the place-by-place review](https://nzbrands.ai/research/hospitality-tourism/immigration/downloads/place-sentiment-ledger.csv)

## What we read in full

We counted names and selected phrases across the complete answer set, then examined 41 selected answers in detail.

The selected answers cover all nine profiles, all eight topics, all four platforms and several country settings. The answers were chosen for useful contrasts, not at random. The findings do not give an overall error rate or a complete ranking of recommendations.

Each example links to the full original answer. Counting a word or place name is separate from judging what the answer recommends.

## How we counted sources and links

Sources attached to an answer and web links in the answer text are separate measures.

Each website recorded as a source counts once per answer. “Had sources” counts answers with at least one recorded website. Website rankings count how many answers have the website as a source. Percentages show the mix of website types across those sources, not a share of answers. Websites without a confirmed type appear under Other.

“Showed you a link” counts answers containing a web address beginning with http:// or https://, including image and file links. A separate downloadable table excludes image and file links. A recorded source does not prove that the page supports every statement, and an answer without a link may still have used web search.

The source records were retrieved on 26 October 2026 for the same 572 New Zealand answers. The source update did not collect new answers or change the collection and policy-check dates. A downloadable table separates answer totals from the number of question styles that produced those sources; repeated appearances in one style do not establish consistency across styles.

## How we checked selected advice

We checked 47 statements covering 27 policy questions.

The six original checks were made on 22 October 2026. We expanded the review on 28 October, using 30 complete saved answers across all nine profiles and four platforms. The reviewed statements include correct guidance, errors, outdated rules and wording that needs qualification.

Both versions of the visa questions were read for every profile. Thirteen of the 27 policy questions have checked statements from only one wording. The downloadable coverage table shows which wordings support each check; a single wording does not establish consistency.

Effective dates are assessed against the 15–20 October collection period. A future rule announced before collection is not treated as already in force. Two questions with missing context remain outside the verdict tally.

The checks cover the statements listed, not every statement in every reviewed answer. The selection is not random and does not give an overall accuracy rate. The original AI wording remains unchanged, including mistakes.

[Read what we checked](https://nzbrands.ai/research/hospitality-tourism/immigration#checks)

## What the questions actually said

We interpret each answer using only the facts stated in its question.

A review of all 143 questions corrected 57 descriptions of the facts supplied. Twenty-one questions do not name New Zealand. The country setting and the intended profile do not supply facts missing from the question.

Student questions do not explicitly state international fee status. Some also omit the destination, the parents’ location or earlier qualifications. Domestic fees, funding and support advice cannot be judged as though those missing details were supplied.

The original questions, answers and name counts remain unchanged. The downloadable review shows the corrected facts and interpretation limits for each question. Future questions need the relevant missing context before collecting new answers.

[Read the question context review](https://nzbrands.ai/research/hospitality-tourism/immigration/downloads/prompt-reporting-eligibility.csv)

## Following the evidence

Every published example links to the original answer.

The archive contains 797 different, non-empty answers. The saved questions were checked against the approved wording, and no duplicate or unexpected answers were found.

Download the full answers, questions, planned and missing answers, place counts, comparisons between countries and explanations of the counting rules.

[Read and download the evidence](https://nzbrands.ai/research/hospitality-tourism/immigration/evidence)

[Evidence guide (Markdown)](https://nzbrands.ai/research/hospitality-tourism/immigration/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/)