AI SEO Agency NZ

How getting cited by an LLM is different from ranking on Google

Ranking is a contest for position: you optimise a page to beat competing pages for a keyword, and the algorithm returns a list. Citation is a contest for trust and extractability: a language model is generating an answer in real time and deciding, sentence by sentence, whose material it can safely paraphrase or quote. It favours content that is unambiguous, well-structured, factually specific and easy to lift cleanly, over content that is merely keyword-relevant. A page can rank well and still never get cited, because it buries the answer in marketing language the model cannot extract with confidence.

The other difference is exposure. A page-one Google ranking is visible and auditable, you can see your position with a rank tracker. AI citations are opaque and variable: the same prompt can return a different set of sources on different days, on different platforms, and for different users, because the models are retrieving and re-generating rather than returning a fixed index. That means AIO is not a one-off optimisation you complete and walk away from. It is closer to an ongoing campaign of building the kind of source material that keeps getting selected, and watching, probe by probe, whether it is working.

There is also a reputational dimension classic SEO never had to deal with. When an LLM cites you, it is usually paraphrasing your claims in its own words, sometimes attributed, sometimes not. If your public content is vague, inconsistent or contradicts itself across pages, the model either avoids citing you or, worse, cites you inaccurately. Precision and consistency in what you publish about your own business become a search-visibility asset, not just a brand-hygiene concern.

What we actually do: liftable content, schema, entity building and off-site presence

We start with liftable answers. For each topic a buyer is likely to ask an assistant about, we write a direct, self-contained answer near the top of the relevant page: a clear claim, the specific number or fact behind it, and enough context that the sentence makes sense lifted out of the page entirely. This is a different writing discipline to persuasive web copy. It favours plain declarative sentences, specific figures, and named entities over broad claims and adjectives, because that is what a model can quote without hedging.

We wire that content with schema markup, structured data that tells an AI system explicitly what a page is, who the organisation behind it is, what service it describes, and how it connects to related pages. Organisation, Service, FAQPage and Article schema all narrow the gap between what a model has to infer and what it is told outright. Machines still read structure more reliably than they read tone.

We build the brand as an entity, not just a website. That means a consistent, unambiguous description of who you are, what you do and where you operate, repeated the same way across your site, your Google Business Profile, LinkedIn, industry directories and any press or partner mentions. Language models cross-reference multiple sources to decide whether a claim is trustworthy; a brand that says the same true thing about itself everywhere is easier for a model to trust than one that reads differently on every page.

We build off-site presence deliberately: citations, structured directory listings, partner and supplier mentions, and content published where LLMs already draw training and retrieval data from, such as Reddit threads, review platforms and industry publications. This is citation engineering in the literal sense, increasing the number and quality of places on the open web that name you correctly next to your category, so a model doing retrieval has more than your own website to go on.

How we measure it: weekly probes and citation rate

AIO cannot be measured with a rank tracker because there is no fixed position to track. Instead we run weekly probes: a set of real buyer-style prompts for your category, run against ChatGPT, Perplexity, Claude and Google AI Overviews / AI Mode, and logged for whether your brand is mentioned, whether it is cited as a source, what is said about you, and who else showed up instead. Over weeks this builds a citation rate, the share of target prompts where you are actually named, which is the AIO equivalent of a ranking report.

We track this alongside sentiment and accuracy, not just presence. Being named is the first bar; being named correctly, with your actual pricing, service area and positioning intact, is the one that matters for leads. Where a model is citing a competitor or a stale, third-party listing instead of you, that gap tells us exactly which page or piece of structured data to fix next.

Because assistant answers vary by platform, by phrasing and by day, we do not report a single number and stop. The probe set runs on a schedule, results get logged over time, and the direction of the trend, not any one week’s snapshot, is what tells you the work is landing.

The probe prompts themselves are built from real buyer language, not guesswork. We pull from the questions your sales team actually gets asked, your existing search-query data, and category-level research into how people phrase requests to an assistant versus a search engine, which is often more conversational and more specific than a typical Google query. A probe set built on invented questions tells you very little; one built on the phrasing real buyers use is what makes the citation-rate trend meaningful.

AI Search Optimisation for New Zealand businesses: a first-mover advantage in a small market

Most New Zealand businesses are currently invisible to AI answers, even ones that rank comfortably on Google. Local SEO has had over a decade to mature here; AIO has had barely two years, and almost no NZ SME has structured content or brand-entity data built specifically for it. When we run probes for New Zealand trade, professional-services and retail categories, the pattern is consistent: assistants either name a large, generic national or overseas player, or return a vague non-answer, because no local business has published the kind of clean, citable material a model can safely quote.

That gap is the opportunity. New Zealand is a small enough market that a handful of well-structured, well-cited businesses in a category can become the default sources an AI assistant reaches for, and once a model has learned to trust and retrieve from a source, that position tends to be sticky, the same way an early, well-earned Google ranking used to compound for years. Getting there now, while almost nobody else in most NZ categories is doing this work deliberately, costs a fraction of what it will cost once every competitor has caught on.

We do this work as the agency actually building the sites and systems, not as a reseller of a service we do not understand. The same structured-content and schema discipline we use across SpeaktoNotes, SpeaktoFinance and SpeaktoClients and our own client sites is what we apply here, so the AIO work sits on the same technical foundation as the rest of your online presence rather than bolted on as an afterthought.

The advantage compounds across the six main centres differently, because each has its own mix of industries and its own gap between offline reputation and AI visibility. A trades business in a fast-growing centre like Tauranga is competing against almost no AI-citable content at all; a professional-services firm in a research-heavy market like Wellington is competing against thin content rather than none. Either way, the New Zealand-wide pattern holds: the businesses publishing genuinely liftable, well-structured, accurately schema-marked content today are the ones an assistant will still be citing in two years, long after the rest of the market has caught up and the easy wins are gone.