Large language models now sit between searchers and websites. A shopper types “Where can I hire a ute in Palmerston North today?” into ChatGPT or Perplexity and receives a conversational answer,often with phone numbers, prices, and opening hours stitched together from several sources. 

The user might never click through to the underlying sites; the model has already satisfied the request inside the chat window.

This trend toward “zero-click” discovery echoes what happened with featured snippets on Google, but the stakes are higher. LLMs summarise multiple pages, decide which facts to surface, and sometimes choose which brands to credit with a hyperlink. 

If your business information is messy, buried, or contradictory, the model may drop you from the answer entirely, handing the mention (and the sale) to a competitor.

Below is a step-by-step guide on structuring pages, feeding models the right data, and measuring your brand’s share of voice in these new answer boxes. Strategies are framed for New Zealand businesses, using examples that apply whether you run an e-commerce store, a local service, or a B2B exporter.

‍

1 Understand how an LLM decides what to surface

Before optimising you need to know the input pipeline. Public documentation is thin, but server-log forensics and academic papers reveal a three-step flow:

• Crawler fetch
  A lightweight bot grabs the first 100 kilobytes of HTML and any linked JSON-LD. If your key facts are further down the page or gated behind a script, they may be ignored.
  
  
• Index and de-duplicate
  Content is parsed into tokens and clustered with similar statements from other URLs. Conflicting data gets scored; the version with higher authority wins.
  
  
• Answer assembly
  When a user asks a question, the model retrieves relevant clusters and rewrites them into fluent prose. Citations may be appended if the engine quotes directly or if confidence thresholds are low.
  
  

Knowing this, we can influence each stage: get fetched, look authoritative, and earn the citation.

‍

2 Surface the non-negotiable facts at the top of every key page

LLMs love concise, unambiguous details. For New Zealand companies the must-have dataset fits inside a single “business card” block:

• Legal trading name
  
  
• Street address and suburb (NZ Post format)
  
  
• Phone in +64 format
  
  
• Region or nationwide coverage
  
  
• Opening hours in local time
  
  
• Primary product or service category
  
  

Place this block near the top of your HTML (not just visually in CSS). Many sites tuck it in the footer, but that risks falling outside the crawler’s intake slice. A table or definition list is fine; just avoid graphic images of text.

‍

3 Mark up pages with machine-friendly schema

Structured data is no longer an optional SEO extra; it is the LLM’s cheat sheet. Use JSON-LD with the LocalBusiness or Product type. Include these properties at minimum:

name: “Wellington Roof Repair Ltd”
address: Full object with street, region, postalCode, country “NZ”
telephone: “+64 4 555 5555”
geo: Latitude and longitude if customers visit you
openingHoursSpecification: Mon-Fri 08:00-17:00, Sat 09:00-12:00
areaServed: “Lower Hutt”, “Upper Hutt”, “Porirua”
sameAs: Links to your Facebook, Instagram, or Trade Me store

Two small details improve model trust:

• Use @id with a stable URL such as https://example.co.nz/#business to give the entity a unique anchor.
  
  
• Insert knowsLanguage with "en-NZ" and any other languages you serve (e.g. "mi" for Te Reo Māori).
  
  

‍

4 Keep citations consistent across the wider web

Perplexity, Gemini, and ChatGPT cross-check your onsite data against other sources. Mismatched phone numbers or trading names lower confidence, so the model may omit you. Run a quick audit:

• Google Business Profile
  
  
• NZBN register
  
  
• Yellow NZ listing
  
  
• Facebook “About” page
  
  
• Trade Me store profile if applicable
  
  

Update any discrepancies. Use exactly the same abbreviations (Ltd vs Limited) and punctuation.

‍

5 Answer the obvious questions in plain text

Schema handles the basics, but users ask follow-ups: “Do you do emergency call-outs?” or “Is the sunscreen reef-safe?” A well-structured FAQ section becomes the language model’s source of truth. Place FAQs directly on the relevant page rather than in a pdf; add a FAQPage schema block so the crawler maps each question-answer pair. Keep answers factual, under 60 words, and free of marketing fluff.

‍

6 Provide rich media that loads fast

LLMs are heading toward multi-modal reasoning, meaning images and soon video will inform answers. Optimise hero images to under 150 KB and provide descriptive alt text:

<img src="/images/kauri-sunscreen.jpg"     alt="Reef-safe SPF 50 sunscreen made in New Zealand in a recyclable aluminium tube">‍

If the crawler stops at 100 KB you still want that alt text captured.

‍

7 Use conversational headings that double as prompts

H2 or H3 headings such as “Do you offer same-day ute rental in Christchurch?” read awkward in traditional copy but act as retrieval triggers. LLMs look for semantic matches to the user query; a matching heading boosts relevance and can turn into a bolded line inside the answer box.

‍

8 Link outward to prove you are part of the wider ecosystem

An isolated site looks less trustworthy. Outbound links to authoritative local entities signal legitimacy:

• Industry associations (.org.nz)
  
  
• Regional councils (.govt.nz)
  
  
• Accredited supplier listings
  
  

Keep them to three or four per page. Too many outbound links dilute Pagerank.

‍

9 Leverage your own AI channels to reinforce authority

Text on a web page is not the only citation an LLM can ingest. Customer chat transcripts, voice call transcriptions, and answer logs also become part of the public web if published (with consent).

• Export top-asked questions from AI Chat Agents, clean up language, and publish them in a Q&A blog post.
  
  
• Use call summaries from AI Voice Agents to craft case studies that describe location-specific success stories.
  
  
• Track brand mentions with Visibility Tracker to see which new phrases the models attach to you—then mirror those phrases on service pages.
  
  

The more places your verified data appears, the higher the chance a model will rank it above scraped directory noise.

‍

10 Measure success the LLM way, not the SERP way

In a zero-click world traditional metrics such as organic sessions dwindle. Adopt new ones:

• Answer share
  Percentage of relevant LLM answers that mention your brand at all. The EnvokeAI Visibility Tracker provides this as “LLM Share of Voice.”
  
  
• Citation rate
  How often the mention includes a hyperlink. Perplexity emits citations 90 percent of the time if the source is trusted. ChatGPT citations are still rare but increasing.
  
  
• Mention sentiment
  Positive vs neutral vs negative phrasing. Voice-of-customer still matters inside generated text.
  
  

Set quarterly goals. A move from 3 to 6 percent answer share can translate into hundreds of extra direct enquiries even without traditional clicks.

‍

Putting it all together: a content checklist for 2025

Task: Add business card block near top of each service page
Tool or Location: CMS theme template
Status: 

Task: Insert JSON-LD with full address, phone, areaServed
Tool or Location: Site-wide header
Status: 

Task: Align NAP across Google/Yellow/NZBN
Tool or Location: Manual audit
Status: 

Task: Publish a concise FAQ with FAQPage schema
Tool or Location: Under product description
Status: 

Task: Optimise hero images to < 150 KB + descriptive alt text
Tool or Location: Media library
Status: 

Task: Create heading prompts for top three buyer queries
Tool or Location: Page editor
Status: 

Task: Add 3 outbound links to trusted NZ organisations
Tool or Location: Body copy
Status: 

Task: Repurpose chat transcripts into Q&A posts
Tool or Location: AI Chat Agents export
Status: 

Task: Publish location case study from call logs
Tool or Location: AI Voice Agents data
Status: 

Task: Schedule monthly answer-share report
‍Tool or Location: AI Visibility Tracker
‍Status:

Tick each box and you will have a site ready for zero-click discovery, one that is quoted accurately, credited visibly, and visited when the rare click does occur.

‍

Finishing Up

Zero-click answers are not the end of web traffic; they are an evolution. Much as bricks-and-mortar shops survive alongside e-commerce, detailed on-site content will still attract engaged readers. 

The goal is to ensure that when a prospective customer stays inside a chat window, the model’s answer includes your brand, your facts, and, when appropriate, your link.

By placing unambiguous data high in the HTML, reinforcing it with consistent citations, and measuring share of voice inside LLMs, New Zealand businesses can thrive in this new landscape. 

The work starts with structured data and ends with ongoing monitoring, a cycle made easier by tools like AI Chat Agents, AI Voice Agents, and Visibility Tracker.

Prepare now, and the next time a customer asks an LLM for “the best local provider,” the engine will have only one obvious answer: yours.

‍