SEOMarketing

How to Get Your NZ Business Cited by ChatGPT and Google's AI Answers

Kage Works9 September 20269 min read

Ask ChatGPT for a plumber in Hamilton and it names two or three, with links. Whether your business is one of them has turned into a product in New Zealand: AEO audits, GEO retainers, llms.txt files, "AI schema" add-ons. Two things have landed since those packages went on sale. Google published which of them its systems use, and one New Zealand study measured what local AI answers are built from. Read both before you sign anything.

What the New Zealand measurement found

Surfaced, a New Zealand AI-visibility company, published a state-of-the-market report in July 2026. The audit side covered 7,129 New Zealand business websites across 22 industries and 24 urban areas. The interesting half is the citation trace: 25 local-service questions phrased the way a customer would ask them, run three times each across ChatGPT, Gemini and Perplexity with New Zealand geolocation, giving 225 answers with every cited source logged (Surfaced).

A business's own website turned up in 80% of the ChatGPT answers, 93% of the Gemini answers and 100% of the Perplexity answers. Roundup pages of the "best plumbers in Hamilton" kind appeared in 33% to 68% depending on the engine. Reddit ranged from 12% to 63%. Directory aggregators, meaning Foursquare, Bing Places and Google Maps listing pages, appeared in none of the 225.

Take the ordering rather than the decimals. The authors say so themselves: it is a bounded sample, the engines are non-deterministic, and the trace is "a dated sample rather than a stable market-share estimate". It is also a vendor's own research, published by a company that sells work off the back of it. The ranking of source types survives all of that, and it points somewhere useful.

That direction runs against the global numbers most people have read. Semrush tracked more than 230,000 prompts over 13 weeks from 14 July to 12 October 2025 across ChatGPT search, Google AI Mode and Perplexity, logging over 100 million citations, and found Reddit, Wikipedia, LinkedIn and YouTube at the top (Semrush). Both results can be true together. Broad informational questions pull encyclopaedic and forum sources, because that is where general answers live. "Emergency electrician in Tauranga" pulls electrician websites and New Zealand roundup pages, because nothing on Wikipedia answers it.

For the searches that pay a local business, your own site is still doing the work, so most of the effort belongs where it already was. Our guide to what moves rankings for NZ small businesses covers that ground.

Google published the list of things you can skip

Google's optimisation guide for generative AI features is unusually blunt about the tactics being sold around it. On machine-readable files, including llms.txt: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them." On markup: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add." On the practice of splitting pages into small blocks for retrieval: "There's no requirement to break your content into tiny pieces for AI to better understand it" (Google Search Central).

The llms.txt evidence outside Google matches. Ahrefs checked 137,210 domains in May 2026 and found 38,360 of them, about 28%, serving a valid file. Of those, 97% received zero traffic that month. Among the 3% that saw any requests at all, 96% of the requests came from bots and 77% of those bots were not AI tools. No AI crawler went looking for a file that was missing (Ahrefs).

Writing one yourself costs an hour and harms nothing. Paying a retainer built around one is a different transaction. Ask whoever is pitching it for server log lines showing a named AI crawler requesting the file on a site they manage.

The one control that sits outside SEO

OpenAI runs separate crawlers for separate jobs, and this is where a real decision hides. In OpenAI's own words, GPTBot "is used to crawl content that may be used in training our generative AI foundation models", while OAI-SearchBot "is used to surface websites in search results in ChatGPT's search features". The consequence is stated plainly: "Sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers, though can still appear as navigational links" (OpenAI).

One robots.txt file governs both, and the two decisions pull in opposite directions. Plenty of owners blocked AI crawlers across 2024 and 2025 to keep their photography and writing out of training data, which is a fair call to make. Sweeping OAI-SearchBot into that block also takes the business out of ChatGPT's answers, and almost nobody who ticked the box was told that.

Squarespace draws that line in a more sensible place than the warnings circulating about it suggest. Settings then Crawlers gives you "Block known artificial intelligence crawlers", which writes disallow rules for a list of named agents including GPTBot, ClaudeBot, anthropic-ai and Google-Extended. OAI-SearchBot and PerplexityBot are not on that list, so the toggle acts as a training block rather than a search block. Squarespace still warns that enabling it "might negatively affect your site's visibility in AI-powered search and chatbot results" (Squarespace Help Center).

Google's side needs the same untangling. Google-Extended governs training and grounding for Gemini, and Google states that it "does not impact a site's inclusion in Google Search nor is it used as a ranking signal in Google Search" (Google Search Central). AI Overviews and AI Mode run off Googlebot instead, and Google's AI features documentation says "robots.txt directives for Googlebot is the control for site owners to manage access to how their sites are crawled for Search", with the granular options being nosnippet, data-nosnippet, max-snippet and noindex (Google Search Central). No setting leaves you in Search while keeping you out of AI features.

Open yourdomain.co.nz/robots.txt in a browser tab and search the page for OAI-SearchBot. If a disallow is sitting there, someone should choose it on purpose rather than inherit it from a default or a well-meaning developer in 2024.

The citations you do not control

Put two numbers from the New Zealand trace next to each other and your outreach budget should move. Roundup pages showed up in 33% to 68% of answers. Directory aggregator listings showed up in none. The single most-cited domain across the whole trace was moneyhub.co.nz with 76 citations, almost all of it roundup content.

A paid directory entry did not earn a single citation in that sample. Getting named on a New Zealand shortlist page did, repeatedly. That work looks like emailing the person who maintains the list with a real reason to include you, not buying a listing. It is slower and it is not something an audit tool can do for you.

Reddit sits in the same bucket, at 12% to 63%, and carries a rule attached. Posting recommendations for your own business while presenting yourself as an independent customer is misleading conduct in trade under the Fair Trading Act, and the Commerce Commission has taken it to court. Brand Developers, trading as The TV Shop, was convicted on 13 charges after staff wrote positive "customer" reviews on multiple websites without disclosing who they worked for, some of them for products the writer had never used. The District Court ordered the company to pay $1,104,000, with Judge Belinda Sellars KC describing the review conduct as deliberate and systematic (Commerce Commission). Turn up on Reddit as yourself, or stay out of it. Our local SEO guide for NZ trades covers where the same line falls for reviews.

What you can measure, and what you cannot

Google shipped a Search generative AI performance report in Search Console on 3 June 2026 and finished rolling it out to all sites worldwide on 31 August 2026. It covers AI Overviews and AI Mode, and it measures one thing: how many times links to your site were shown to a user in a generative AI feature on Google Search (Search Console Help). No clicks, no click-through rate, no query data, and Search Labs experiments sit outside it. Treat it as a presence check rather than a performance report.

Referral tracking is worse. ChatGPT visits often arrive without a referrer and land in your analytics as direct traffic, so the channel usually looks smaller than it is. On the quality side, Ruler Analytics added AI referral as a benchmark category in its 2026 report, drawn from more than 110 million sessions and 5 million conversions across 13 industries, published 26 May 2026. AI referral converted at 5.8%, against 4.9% for organic search and a 5.13% average across every channel they measured. Ruler notes the volumes behind that number are still modest (Ruler Analytics). Their spend data is UK-weighted, so treat it as a shape rather than a New Zealand benchmark.

A small stream of visitors converting above your organic average is worth keeping. Judging it on session count will make you switch it off.

An hour's work, in order

Open your robots.txt and settle the OAI-SearchBot question. Then ask ChatGPT and Gemini the five questions your customers ask you on the phone, with location set to New Zealand, and write down which businesses and which pages get named. Turn on the Search Console generative AI report so you have a baseline from today. Find the two or three New Zealand roundup pages that already rank for your service and city, and email the people who run them.

Everything after that is the ordinary work: pages that answer the questions customers ask, on a site that loads quickly. Google's guide lands in the same place after several thousand words of guidance, reducing it to one line: "focus on what your visitors would enjoy, find helpful, and feel satisfied with after visiting your website." If a proposal in front of you is priced mostly around files and markup that Google says it does not read, that is a fair thing to ask about before signing. The questions worth asking any web agency cover how to run that conversation.