Field Notes19 July 20268 min read
Are AI tools recommending your NZ business? How to find out
Most NZ businesses have no idea whether ChatGPT or Perplexity recommends them. Here's the measurement framework.
Most NZ businesses have no idea whether ChatGPT or Perplexity recommends them. Here's the measurement framework.
The first thing most NZ business owners ask when they hear about AI search optimisation is whether they should invest in it. The second question, usually about thirty seconds later, is how they would even know if it was working.
That second question is the right one, and it doesn’t have a clean answer yet because the tooling for tracking AI search visibility is still catching up to the behaviour itself. But the absence of perfect measurement tools doesn’t mean you’re in the dark. There are specific, repeatable checks you can run today that will tell you whether ChatGPT, Perplexity, Claude, and Google’s AI Overviews are recommending your business, and how that changes over time. Most NZ businesses haven’t run any of them, which means the businesses that start now have a genuine first-mover advantage in their categories.
Google rankings have a clear measurement infrastructure. Tools like Search Console, Ahrefs, and Semrush show your position for specific keywords, how that position has changed over time, and what queries caused your pages to appear. The data is granular, historical, and reliable.
AI search doesn’t work this way. ChatGPT, Perplexity, Claude, and Google’s AI Overviews generate their answers dynamically. They don’t publish citation data publicly, and the answers they return vary based on query phrasing, conversational context, and model updates. A business recommended prominently by Perplexity this week may receive slightly different treatment next week after an index refresh. There is no API for “which NZ businesses did you recommend for commercial fitout design in Wellington this month.”
What you can do is run structured manual checks and track them consistently over time. The consistency of the method matters more than the sophistication of the tools.
Not all AI search tools matter equally for NZ businesses. Which ones to prioritise depends on your market and buyer type.
ChatGPT is the most widely used AI search tool in NZ and Australia by a considerable margin. For B2C businesses, consumer product searches, and local service recommendations, it has the broadest reach of any single tool. If you’re only testing one surface, this is it.
AI Overviews appear at the top of standard Google search results for an increasing range of queries. Because most NZ searches still start on Google, AI Overviews have high exposure even for users who don’t think of themselves as AI search users. Testing here carries a high priority regardless of business type.
Perplexity has a smaller but growing user base in NZ that skews toward research-oriented buyers making higher-value decisions. For professional services, B2B, and technical categories, Perplexity is often cited by people who take longer purchase cycles. If your business serves that segment, it belongs in your regular testing set.
Claude is less commonly used as a standalone search tool but is embedded in an increasing number of business workflows and third-party applications. Tracking your visibility in Claude responses is lower priority for most NZ businesses, but worth including in a full baseline audit to understand where you stand across the field.
The queries most worth running fall into three categories: category queries, location queries, and comparison queries.
Category queries are the questions a prospective buyer might ask before they know which business they want. “Who does commercial photography in Wellington?” “Which Auckland accounting firms specialise in e-commerce businesses?” “What’s a good NZ web design agency for a hospitality business?” These test whether you’re being named in the category at all.
Location queries are the geographic variants. Adding your city, region, or suburb changes which businesses a model will cite and how confidently. “Wellington” and “lower North Island” and “NZ” can produce meaningfully different recommendations from the same underlying model. Testing each gives you a picture of where your geographic coverage is strong and where it’s absent, which is the kind of gap an AIO engagement is designed to close.
Comparison queries ask the model to recommend or evaluate specific options. “What’s the difference between [business type] X and Y?” or “Which option would work better for a NZ business in my situation?” These reveal whether you’re considered credible enough to include in a comparison, which is a higher bar than simply being mentioned in a list.
For most NZ businesses, running eight to twelve queries across these three categories, split across ChatGPT and Google AI Overviews at minimum, gives a meaningful visibility baseline. Run them on a fresh session without logged-in personalisation wherever possible, so the results reflect what a new user would see rather than what the model has personalised toward your own browsing history.
There’s a meaningful difference between being cited in an answer and being recommended, and the gap between the two shapes what work needs to happen next.
Being cited means the model names your business in a list. “Some options in Wellington include X, Y, and Z.” This is the floor, not the goal. A citation in a list of six businesses is useful visibility but carries less commercial weight than a direct recommendation.
Being recommended means the model has enough confidence in your business to say something specific: “For commercial fitout in Wellington, [business name] is worth looking at because they focus on hospitality operators and have recent project work in that segment.” That sentence requires the model to have read something specific and credible about you, not just encountered your name somewhere.
The distinction matters for AI search optimisation because it tells you which problem you’re solving. A business that isn’t being cited at all has a discoverability problem, typically a clarity gap on its website where the model can’t determine with confidence what you do, for whom, and where. A business that’s being cited but not recommended has a credibility gap: the model has found it but hasn’t encountered enough specific, verifiable content to make a confident recommendation.
An AIO audit distinguishes between these two situations and gives you different work to do in each case.
The approach that’s tractable for most NZ businesses: run the query set once a month, record the results in a spreadsheet, and track three things. Whether you appear at all. Whether you appear in the first two businesses named. Whether the recommendation includes any specific detail about what your business actually does.
Over twelve months, that data tells you whether content updates, website design changes, and SEO work are producing a visible shift in how AI tools describe your business. It also tells you if a competitor has moved up in recommendations, which is worth knowing early rather than late.
The record-keeping doesn’t need to be elaborate. Copy the AI response, paste it into the spreadsheet with the date and the query, and flag whether your business appeared and what the model said about it. After six months of monthly checks, you have enough data to see the direction of travel clearly.
One technical note: AI models can produce different answers to the same question depending on query phrasing, model version, and whether a live web search was triggered. Running the same query twice and getting slightly different responses is normal. Track patterns across multiple queries in each session and across months rather than reading individual responses as definitive verdicts.
The inputs that move AI search recommendations overlap with traditional SEO but are not identical.
Content specificity is the strongest single lever. A business whose website clearly states what it does, for whom, in which geographic markets, and with what evidence, gives AI models material for a confident recommendation. A website that says “we help businesses grow through innovative thinking” gives a model nothing to anchor to. The structural fix here almost always involves website and content changes before any AI-specific optimisation layer can work.
Third-party mentions matter more than most NZ business owners expect. Reputable publications, local directories, industry association listings, and customer review platforms all feed into the picture AI models form of who your business is. This is why the overlap between SEO and AIO is real: the external credibility signals that support search rankings also support AI search recommendations. Building one builds both.
Recency matters, but not in the same way as Google. A model that has indexed a credible piece of content about your business will continue citing it until a model update or index refresh changes what it has access to. AI search is less sensitive to publishing frequency and more sensitive to the quality and specificity of what exists. Posting more content that says nothing specific won’t move visibility. Publishing one specific, detailed piece that clearly articulates what you do for whom in your market often will.
The measurement observation worth taking from this: the majority of NZ businesses have never run these queries. They don’t know whether they appear in AI search recommendations, they have no record of what a model says about them when they do appear, and they have no baseline against which to measure change.
That situation is a meaningful short-term advantage for businesses that establish a baseline now. Once AI search visibility tracking becomes standard practice in NZ industry categories, the businesses that started early will have months of data and a clear picture of their gaps. The businesses that wait until this becomes obvious practice will be starting from zero in a more competitive field, and the window to establish AI search presence before competitors is compressing each quarter.
The practical first step: run eight queries across ChatGPT and Google AI Overviews today, record what appears, and set a monthly reminder to run them again. The data will make the next steps obvious.
The gains in NZ manufacturing aren't on the factory floor. They're in the admin layer that surrounds it.
Three areas where NZ construction firms are finding real returns from AI, and one where most are still leaving time on the table.
Local search, email, and social: where AI is changing the economics of NZ physical retail.