ChatGPT, Gemini and Perplexity mention some businesses over others, and how stronger third-party corroboration improves AI visibility. 

You can do everything right on your own website, publish good content, hold real credentials, and still never be named when someone asks ChatGPT, Perplexity, Claude or Gemini who the credible providers in your category are. The reason sits off your site. AI systems build their answers from what the wider web says about you, and they weigh agreement across independent sources far more heavily than the quality of any single page you control. If the sources around you are thin, inconsistent or silent, the AI has little to work with, however strong your site is.

The underlying shift is that Google and AI assistants are doing different jobs. Google ranks documents. It asks which page best answers this search, and institutional domains usually win that contest. An AI assistant does an entity job instead. It has to decide which businesses it's confident enough about to name, and that confidence gets built from repetition and agreement across many sources rather than from one authoritative page. An Ahrefs study of around 75,000 brands found branded web mentions correlated with AI visibility more strongly than backlinks or domain authority, a pattern that held across ChatGPT, AI Mode and AI Overviews. That's a correlation across a large dataset, worth taking seriously without inflating it into a law.

We looked at how one New Zealand health and safety training provider, which we'll keep anonymous, shows up for a search like "health and safety courses NZ". In the AI-generated answer for that query, the provider was named among the top options, alongside recognised industry bodies. In the normal Google results for the same search, universities, polytechnics and government sites led the page and the provider sat well below them. AI answers like this vary from one search to the next, so the fair description is that the provider is reliably in the cited set rather than that it ranks number one. A business losing the document-ranking race was being surfaced prominently in the AI answer for the identical query.

The specific-query behaviour reveals the mechanism further. On narrower queries like "confined space training NZ" or "working at heights training NZ", the provider drops back or disappears, and those queries often don't trigger an AI answer at all. When we checked the provider's accreditations against independent sources, the pattern held. Its NZQA registration is confirmed on NZQA's own government provider register. Its NEBOSH Learning Partner status is corroborated publicly by a New Zealand safety industry body. Its IOSH-approved courses are stated consistently but lean more on the provider's own material. The signals the AI is most likely reading are the ones with independent corroboration, a government register, an industry association and a polytechnic partnership all describing the same entity the same way. In the AI overview for that query, studywithnewzealand.govt.nz appeared as a cited source. To be precise, the government directory was a cited source in that answer. We didn't trace whether it was what surfaced the provider, and inferring that would dress an assumption up as an observation.

Building corroboration in a thin ecosystem

Most of the AEO advice you'll read was written for the US and UK, and it assumes a depth of independent review sites and industry press that New Zealand doesn't have. There's no local Trustpilot equivalent with critical mass, industry press is thin and getting thinner, and regional news has hollowed out. A Koi Tū report by Dr Gavin Ellis, as reported by RNZ, notes New Zealand employs around 1,400 journalists in total, fewer than the New York Times employs alone. Stuff and NZME have shed most of their community papers over the past decade, NZME alone closing 14 titles, and credible NZ voices are now using the term "news deserts". You can't wait to be written about here. You build the corroboration deliberately.

Start with entity consistency, because it's the lever fully in your control. Get your name, address, phone number and one-line description identical everywhere you appear: your own site, Google Business Profile, LinkedIn company page, the Companies Register and every directory. Then work through the sources a legitimate NZ business can genuinely reach. The .govt.nz layer carries weight, so make sure your NZBN and any sector-specific register are accurate. Join your industry or professional association and take the member directory listing, which delivers the same corroboration signal as an accreditation and is reachable in weeks. Complete your Google Business Profile properly, take your regional Chamber of Commerce listing, and start collecting genuine reviews now. This works because it's corroboration.

For a newer business, the established incumbent owns the broad category query, and that's honest. On narrower, high-intent queries the confidence threshold appears lower and the field is largely open, and the consideration sets for most NZ category queries are still forming. Expect no dashboard for AI citation, answers that shift between runs, and six to twelve months before consistent signal. You cannot make an AI cite you, only improve the odds. The pattern we're observing is that the field is still open in New Zealand, and the businesses building this foundation now appear to be establishing positions before the category settles. You are not trying to become an institution. You are trying to be the clearest, most consistent, best-corroborated answer to the specific questions your customers ask. The technical groundwork that sits underneath all of this, building for the AI agents that read and act on your behalf, is where we go next.

Frequently Asked Questions

How do I check what AI currently says about my business?

Ask your main category question across ChatGPT, Perplexity, Claude and Gemini, and note whether you appear, how you are described, and which sources get cited. Run it a few times, because answers vary between sessions. Then check whether your accreditations, reviews and business details read consistently across the web.

I run a newer business without years of accreditations. Can this still work for me?

Yes, though not on the broad category questions, where established, well-corroborated providers tend to win. Your opening is the narrower, specific queries your customers actually type, where competition is thinner and the confidence AI needs is lower. Consistent details, real reviews and credible listings can get you there in months.

Which New Zealand listings or registers are actually worth getting on?

Start with the ones AI already treats as credible: government and sector registers on .govt.nz, your industry or professional association's member directory, a complete Google Business Profile, your LinkedIn company page, and your regional Chamber of Commerce. Accuracy and consistency across them matter more than how many you join.

How long before any of this shows up in AI answers?

Longer than on-site changes, and not on a fixed timeline. Entity details you can fix in days, but the corroboration they enable builds over months as other sources are indexed and the models catch up. Expect six to twelve months for consistent signal, not weeks, and treat it as a foundation.

If you want this checked properly, ADMATIC runs a free AI visibility check for Marketing Association members. It looks at how ChatGPT, Perplexity, Claude and Gemini currently describe your business and where you are absent, and whether the AI crawlers can reach your site in the first place. It is a one-page snapshot of where you stand, and a starting point for a fuller Foundation Audit.

Source: Rod Russell, Managing Partner, ADMATIC ,  17th July 2026