Research · July 2026
The State of AI Visibility in New Zealand 2026
We audited 7,129 New Zealand business websites and traced the citations in 225 real AI answers to find out who ChatGPT, Gemini and Perplexity named and which sources appeared. This is a bounded sample, not a census of every New Zealand query or model response.
Free to cite with attribution: “Surfaced, State of AI Visibility New Zealand 2026” + a link to this page.
7,129
unique NZ business domains in the website benchmark
7,411
industry-domain observations; a domain may occur in more than one category
225
AI answers in the separate citation trace
0
Foursquare/Bing Places citations observed in this 225-answer sample
Finding 1
Business websites appeared frequently across all three engines in this sample.
In our 225-answer trace, at least one business’s own website was cited in 80% of ChatGPT answers, 93% of Gemini answers and 100% of Perplexity answers. Many cited pages were ordinary service pages that stated practical facts such as prices, areas served or review counts in plain language. Foursquare, Bing Places and Google Maps were cited zero times in this sample; that observation should not be generalised beyond these questions, engines and run dates.
table scrolls sideways →
| Source type · share of answers citing it | ChatGPT | Gemini | Perplexity |
|---|---|---|---|
| Business websites | 80% | 93% | 100% |
| Roundup / “best of” pages | 33% | 52% | 68% |
| 28% | 12% | 63% | |
| NZ directories | 8% | 5% | 44% |
| Foursquare / Bing Places / Google Maps | 0% | 0% | 0% |
Finding 2
A small set of New Zealand roundup sites recurred in the sampled answers.
The single most-cited domain in the entire study was moneyhub.co.nz (76 citations) — its “best plumbers in Hamilton”-style pages are Gemini’s favourite source. A small layer of roundup sites (TopChoice, 10best, top5.nz) also appeared repeatedly. Inclusion on a cited page can create a relevant source opportunity, but this trace does not prove that inclusion causes a recommendation. ChatGPT returned 3.1 cited domains per answer on average in this sample, compared with 8.9 for Perplexity.
Finding 3
The industry league table: who’s ready for AI search, who isn’t.
Average historical website-readiness score by industry across 7,129 audited websites. These July 2026 benchmark scores use rules v1.0; live checks now use rules v2, so the two versions should not be treated as one time series until the benchmark is re-scanned.
Finding 4
The most common failures are the cheapest to fix.
- 31.8% of category observations had no detected structured data — a domain may appear in more than one industry category, so this is not a unique-domain national rate.
- 4.1% of category observations triggered the historical crawler flag — this rules-v1.0 result predates the current separation of discovery, user-fetch, training and model-control crawlers. Domains may repeat across categories, and the dataset must be re-scanned under v2 before the figure is interpreted as current discovery access.
- A recurring pattern in the trace: plain service pages with concrete facts, service areas and direct answers were present among the cited business pages. This is an observation from the sample, not a causal ranking formula.
What this means for your business
- Publish citable pages. Use clear service and area pages, answer the main question near the start, and include only facts you can substantiate.
- Get onto the roundup pages AI already cites in your industry and city — they’re a short, knowable list.
- Keep public review information accurate. Ratings and counts appeared in some sampled answers; monitor whether that evidence is used for your own questions.
Methodology
Website audits: 7,129 NZ business websites across 22 industries and 24 urban areas, discovered via Google local results (13,101 listings) and audited with Surfaced’s deterministic six-pillar engine (July 2026). Aggregates are published only where a segment has 15+ audited sites; no individual business is named. Citation trace: 25 local-service questions phrased as real customers ask them, run against live ChatGPT, Gemini and Perplexity with New Zealand geolocation, three repeats each (225 answers), with every cited source logged and classified. Engines are non-deterministic; shares are reported at the answer level across repeats. The 225-answer trace is a dated sample rather than a stable market-share estimate, and the published aggregates do not let a reader reproduce every raw response. Full method notes: surfaced.co.nz/methodology.
Cite freely with attribution: “Surfaced — State of AI Visibility New Zealand 2026” with a link to this page. Media enquiries: hello@surfaced.co.nz.