AI Search Measurement in New Zealand: Start With Evidence

AI search has created a measurement problem for New Zealand marketers. A visibility score can move sharply when the prompt set is small, the geography is narrow or one answer changes. Treat that score as a diagnostic, not a business outcome.

Google has now rolled out dedicated generative AI performance reports in Search Console worldwide. They show impressions, pages, countries, devices and trends for appearances in AI features. The data also remains part of overall Search performance. That gives teams a first-party baseline for Google, but it does not measure every answer engine or prove commercial impact.

Start with the decision

Before buying another dashboard, ask what the result would change. Are we deciding which product information to improve, which market deserves content investment, or whether brand demand is translating into qualified enquiries? Set one decision and one accountable owner for each measurement view.

Build three layers of evidence

1. Visibility

Track Google’s generative AI impressions by country and landing page, alongside ordinary Search Console clicks and impressions. Keep branded and non-branded demand separate where the data allows. For third-party answer engines, use a fixed, documented prompt set that reflects actual customer questions. Record engine, locale, date, answer, cited sources and whether the brand appears. Repeat the same sample before interpreting change.

2. Behaviour

Look at referral sessions, engaged visits and the paths people take after landing. Google’s Search Console and Analytics answer different questions; a citation, impression, visit and application are separate events. Check tagging and landing-page quality before attributing a fall in traffic to AI answers.

3. Outcomes

Connect the view to enquiries, applications, sales or other qualified outcomes where those signals are reliable. Compare against a prior period and other demand indicators. If volume is thin, aggregate by month or by meaningful theme and report the underlying counts. Do not turn a handful of sampled answers into a precise market-share claim.

How to read a NZ result

A useful report shows the numerator, denominator and method. Say “appeared in 8 of 40 tracked answers for these ten prompts across four runs”, rather than “20% AI market share”. Note prompt selection, location, model and collection dates. A change from eight to ten appearances is a lead for investigation; it is not proof that two more customers found or chose the brand.

Search Console itself has limits. Google omits some low-volume query rows for privacy while totals can include them. In a smaller market, that makes country and query cuts especially easy to overread. Keep a visible “insufficient data” state and a qualitative example alongside the chart.

What we would put on the first page

One page is enough for a pilot: the business outcome and its owner; Google AI visibility and organic traffic by NZ/AU market; a small, repeatable answer sample; the pages and claims that were cited; and the two or three changes the team will make. Put uncertainty beside each number. Review monthly, with an earlier check for major site or platform changes.

This approach lets D3 distinguish a discoverability problem from a conversion or product problem. It also gives clients a practical decision rather than another score to admire.

Sources

Google Search Central, “Introducing Search Generative AI performance reports in Search Console” (3 June 2026; worldwide rollout noted 31 August): https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports

Google Search Central, “Optimizing your website for generative AI features on Google Search” (updated 10 July 2026): https://developers.google.com/search/docs/fundamentals/ai-optimization-guide

Google Search Central, “A deep dive into Search Console performance data filtering and limits”: https://developers.google.com/search/blog/2022/10/performance-data-deep-dive