Generative Engine Optimisation NZ: The New Front Page

A builder I know rang me in a flat panic, unaware of generative engine optimisation NZ, because his enquiries had halved and his Google rankings had not moved. Same position, same traffic report, half the phone calls. So I opened ChatGPT and typed the exact thing his customers type: “who should I use for a kitchen renovation in west Auckland.” It named four companies. He was not one of them.

That is what this is really about. Not a new acronym to worry about, just a hard fact: a growing share of your buyers now ask an AI assistant instead of scrolling a results page, and the assistant gives them a short list. In this post, I will explain how those short lists get built, why the New Zealand market behaves differently from the US, the nine things I would fix first, and how to tell whether any of it is working.

Key Takeaways

• Generative engine optimisation in NZ is about being the answer an AI gives, not the tenth blue link on a page nobody scrolls to.
• AI assistants build recommendations from consensus across many sources, so third-party mentions often matter more than your own website copy.
• New Zealand’s small data footprint cuts both ways: thin coverage means one well-structured page can carry more weight than it would in the US.
• Short, factual, quotable answers under clear headings get lifted into AI responses far more often than long unbroken prose.
• Your Google Business Profile, NZBN record and review volume feed the models as much as your site does, so keep every detail identical everywhere.
• Track visibility by asking the models the questions your buyers ask, monthly, and logging who gets named. That log is your new rankings report.
• Blocking AI crawlers in robots.txt removes you from the answer set entirely, and most owners have no idea whether theirs are blocked.
• Winning the recommendation is worthless if nobody answers the phone. Response speed decides who gets the job once AI sends the enquiry.

What Generative Engine Optimisation Actually Is

Generative engine optimisation is the work of making your business the thing an AI system names when someone asks it a question in your category. ChatGPT, Claude, Gemini, Perplexity, Copilot, and Google’s AI Overviews all sit on top of search now. They read the web, weigh sources, and hand back one synthesised answer instead of ten links.

The shift matters because the interface changed shape. Old search gave you a page of options and let the buyer choose. New search gives the buyer three names and an explanation of why those three. If you are not in the three, you do not get a second look. There is no page two to be on.

GEO, SEO and answer engine optimisation

People use three terms and mostly mean overlapping things. Traditional SEO is about ranking a page. Answer engine optimisation is about being the source an assistant quotes for a specific question. Generative engine optimisation is the broader version: being the business the model recommends across a whole category of questions, whether it cites you or not.

They are not in competition. Every AI assistant that browses the live web is running search queries behind the scenes and reading what comes back. If your page cannot be found and read by a search crawler, it cannot be pulled into an answer either. Strong SEO fundamentals are the entry ticket. Generative engine optimisation is what you do once you have the ticket.

The part most owners get wrong

The instinct is to write more pages. That was the right instinct for 2015. The models are not counting your pages; they are checking whether independent sources agree about who you are and what you are good at. One page that clearly states what you do, backed by twenty places on the web that say the same thing, beats forty pages of thin content that nothing corroborates.

If you have been through the recent traffic drops, the reason is now obvious in hindsight. I wrote about the mechanics of that in why AI Overviews are eating your traffic, and the short version is that the click never happens because the answer already did.

Why Generative Engine Optimisation in NZ Is a Different Game

Everything written about this topic comes out of the US, and most of it does not transfer cleanly. New Zealand has roughly five and a half million people and a business population dominated by firms with under twenty staff. The data footprint for any given NZ category is thin compared to a US metro.

That is a problem and an opening at the same time.

The problem: models trained mostly on American web content have weak coverage of local categories. Ask about a service in Hamilton and the model has far fewer NZ sources to weigh, so it leans on whatever it can find. Often that is a directory page, a Reddit thread from four years ago, or the one competitor who happened to write a decent explainer.

The opening: in a thin category, a small amount of clear, well-structured, genuinely useful content moves the needle fast. In the US, you are fighting a thousand well-funded content teams for the same answer slot. In Tauranga you might be fighting three businesses, two of which have a one-page website built in 2019.

Local signals carry disproportionate weight

For NZ queries the models lean hard on a handful of anchors: Google Business Profile, the NZ Business Number register, Yellow, Neighbourly, industry association member lists, and regional council or trade body directories. In a data-thin market, those sources fill the gap, and they are frequently the deciding vote on whether the model believes your business exists at all.

I have watched a model refuse to name a perfectly good Auckland business because its Google Business Profile listed a different trading name from its website, and its NZBN record listed a third. The model could not resolve which entity was which, so it named someone simpler instead. That is not a marketing failure. That is an admin failure with a marketing consequence.

The reviews reality

NZ businesses habitually under-collect reviews. Thirty reviews here is often the top of a category that would need three hundred in Sydney. Because the models read review text as evidence, not just as a star rating, a handful of detailed reviews that name specific services and specific suburbs does real work. Two hundred reviews saying “great service, thanks” do almost none.

How AI Decides Which NZ Business to Recommend

There is no secret ranking factor to buy. What happens is closer to how a careful person would research a purchase if they read very fast and had no opinions of their own.

It looks for agreement across sources

When a model builds a recommendation, it gathers a set of documents and looks for things that repeat. If four independent sources describe you as an Auckland commercial electrician who does after-hours work, that is a fact the model will state with confidence. If only your own website says it, the model treats it as a claim rather than a fact.

This is why your own copy has less influence than you would like. You are one source, and you are the source with an obvious motive. Mentions on supplier sites, association member pages, local news, podcast show notes, partner websites and community pages all count as corroboration.

It prefers content it can lift cleanly

Models extract answers. Content that is already shaped as an answer gets extracted more often. A clear question as a heading, followed by a direct two-sentence answer, followed by the detail, is easy to lift. Three paragraphs of scene-setting before you get to the point is not.

The practical test I use: could someone screenshot forty words of this section and have it stand on its own as a complete answer? If yes, it is quotable. If it only makes sense after reading the previous six paragraphs, it will not get used.

It needs to know what you are

Entity clarity sounds abstract and is not. It means: is there one unambiguous answer to “what is this business, where is it, and what does it do?” Consistent name, address and phone number across every listing. Schema markup on your site that states your business type, service area and services. An About page that reads like a factual record rather than a mission statement.

Google’s own guidance on content performance in AI experiences is unglamorous on this point and worth reading. It comes down to unique, useful content that is technically accessible, with the structure making the meaning obvious.

It has to be able to read you at all

A meaningful number of NZ sites are actively blocking the crawlers that feed these systems. Sometimes a developer added it deliberately in 2024 when the training-data argument was hot. More often, a security plugin or a hosting default did it, and nobody noticed. GPTBot, ClaudeBot, PerplexityBot, Google-Extended and CCBot are the ones that matter.

Check your robots.txt today. If those agents are disallowed, none of the rest of this work will produce a thing, because you have removed yourself from the pool before the selection starts.

A Generative Engine Optimisation Checklist for NZ Businesses

Here is the order I would work through it. It is deliberately boring, because the boring parts are the ones that move.

1. Unblock the AI crawlers. Visit yoursite.co.nz/robots.txt. Remove any Disallow rules for the agents above. Ten minutes, and it is the only item on this list that can be worth nothing until it is done.

2. Fix your name, address and phone everywhere. Website footer, Google Business Profile, NZBN register, Facebook, Yellow, LinkedIn, industry directories. Identical strings, not close-enough variations. Same legal name, same trading name, same phone format.

3. Add FAQ content that answers real questions. Not the questions you wish people asked. The ones they actually type. Pull them from your inbox, your call logs, and the Google “People also ask” boxes for your category. Each answer stands alone in forty to eighty words, then expands.

4. Mark up the structure. Organisation schema, LocalBusiness schema, FAQPage schema on any page with a question and answer block. This is a job for whoever runs your site, and it takes an afternoon. It tells the machine what the human is reading.

5. Write the comparison and cost pages you have been avoiding. “How much does X cost in Auckland” and “X versus Y for NZ businesses” are exactly the queries people ask assistants, because they feel too awkward to ask a salesperson. If you will not answer them, someone with less experience will, and their answer becomes the one the model repeats.

6. Get mentioned somewhere that is not your site. Supplier case studies, association member profiles, local business awards, a guest slot on an industry podcast, a quote in a trade publication. Five genuine third-party mentions will do more for your AI visibility than fifty more blog posts.

7. Collect specific reviews. Ask customers to name the service and the suburb. “They rewired our 1930s bungalow in Grey Lynn and worked around our tenants” is worth ten generic five-star ratings, because it is extractable evidence about what you actually do.

8. Keep the important pages current. Recency is a real weighting factor for local and commercial queries. A page last touched in 2022 gets discounted. Reviewing your five core service pages every quarter is enough.

9. Publish something with a number in it. Original data, even small, gets cited disproportionately. Your average turnaround time. The percentage of jobs you complete in a week. What a typical project costs in your region. Facts nobody else has published are the fastest route to becoming a source.

None of that requires an agency retainer or a rebuild. It requires someone to own the list and work down it, which in most founder-led businesses is the actual constraint. If your bottleneck is that everything routes through your head, the checklist will sit untouched, and that is the same problem I write about in admin overload in small business.

How to Tell If Your Generative Engine Optimisation Is Working

This is where most people give up, because there is no rankings dashboard to open. There is a method; it is just manual at first.

Run a prompt audit every month

Write down the fifteen questions your buyers actually ask before they buy. Not keywords, questions. “Who does emergency plumbing in Christchurch after hours.” “What should I pay for a commercial fitout in Wellington?” “Best accountant for a small trades business in NZ.”

Ask each one in ChatGPT, Gemini, Perplexity and Copilot. Log who gets named, in what order, and what the model says about them. Do it the same day each month. Within three months, you have a trend line, and it is a far more honest picture of your market position than any keyword rank.

Use a fresh session each time and turn off any memory or personalisation; otherwise, the model quietly optimises for you rather than for a stranger.

Watch for AI referrals in your analytics

Referral traffic from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com shows up in GA4 as its own source. The volume is usually small, and the conversion rate is usually excellent, because someone arriving from an AI recommendation has already been pre-sold by a source they trust.

Segment it and watch the direction of travel rather than the absolute number. It is the leading indicator that the rest of the work is landing.

Add one question to your enquiry form

“How did you hear about us?” with a free-text field. When people start writing “ChatGPT said you were good” you have your answer, and you have it before it shows up anywhere else. I have seen that appear in NZ enquiry logs for over a year now, and it is climbing.

Watch branded search

If more people search your business name directly, something upstream is introducing you to them. Google Search Console will show it. A rise in branded search with flat category rankings usually means the assistants are doing the introducing.

Getting Found Is Only Half the Job

Here is the part the SEO conversation keeps skipping. An AI recommendation is a higher-intent lead than anything a search results page has ever produced. The buyer did not compare ten options; they asked a system they trust and got told to call you. They arrive warm, specific, and ready.

And then they ring on a Tuesday afternoon while everyone is on site, and nobody picks up.

Research on lead response has been consistent for years: roughly 78% of deals go to whoever responds first. Most businesses take hours. A lead that arrives via an AI recommendation decays faster than most, because the buyer has a list of three names and will simply go to the next one.

This is the pattern I see constantly. A business finally fixes its visibility, the enquiries lift, and the conversion rate falls, because the capacity to answer them was never there. One of the dental practices I work with, Dr Claire’s, was losing close to half its inbound calls before we put an AI receptionist on the line. Missed calls went to zero, and booked appointments lifted 44%. Not one dollar more in marketing. The demand had been there the whole time, ringing out.

So sequence it properly. Fix response first, or at least alongside, because visibility work that lands on a broken intake process just makes the leak more expensive. The follow-up side of this is covered in after-hours call answering if that is your weak point.

The Uncomfortable Bit About Timing

Every NZ business owner I speak to has the same reaction to this: it sounds like a lot, and there is no proof it is urgent yet.

Fair. Search referral volumes are still bigger than AI referral volumes for most categories here. But the mix is shifting quickly, and the reason to act now is not the current volume; it is the compounding. Corroboration takes months to accumulate. The third-party mentions, the review depth, the crawl history, the consistency of your records over time: none of that can be bought in a hurry when you finally decide it matters.

The businesses that will own the answer slot in their category in eighteen months are the ones doing the boring listing hygiene this quarter. That is the whole strategy. It is not clever; it is early.

There is a broader version of this argument worth keeping in view. MIT research last year found that around 95% of AI initiatives deliver no measurable return, and the ones that fail almost always started with a tool instead of a process. Generative engine optimisation fails the same way. Buying a GEO monitoring subscription is not a strategy. Deciding who owns the checklist, on what cadence, with what review process, is.

Conclusion

Generative engine optimisation NZ is not a new marketing discipline so much as a new front page. Buyers used to see ten options and pick. Now they see three names and a reason. Everything that decides which three comes down to whether independent sources agree about what your business is, whether your content can be read and lifted cleanly, and whether the basic records of your existence line up.

The work is unglamorous: unblock the crawlers, fix your listings, publish real answers to real questions, get mentioned by someone other than yourself, and check monthly whether the models have started naming you.

The bigger point is that visibility is one job among dozens sitting on the same overloaded founder. Fixing it permanently means the business has a brain that holds the context and a workforce that does the doing, so this stops depending on you finding a spare Sunday.

Ready to Talk?

If your enquiries have gone quiet and your rankings have not moved, that gap is worth thirty minutes. Book a Discovery Call, and I will run your top questions through the major assistants live, show you who is getting named in your category right now, and tell you honestly whether the fix is a two-week job or a bigger one. No pitch deck, no obligation.

Frequently Asked Questions

About Octavius