Large language models have turned into discovery engines. A growing share of shoppers no longer open Google first. Instead they type natural language questions into ChatGPT, Gemini or Perplexity and expect a tidy answer complete with product suggestions or a short list of providers. 

That shift has huge upside for New Zealand businesses, provided the models can recognise that your company is relevant to a user in Auckland or Queenstown rather than Atlanta or Manchester.

To find out which local signals actually land, we ran a controlled experiment in April 2025. We asked the same ten location-sensitive queries inside Perplexity.ai, Google Gemini Advanced and OpenAI ChatGPT-4o. We used a fresh account with no previous history, set the location to Auckland, cleared cookies between sessions, and recorded the top three business references each model returned. Then we changed the apparent location to Melbourne to see how answers shifted.

Below we break down what we found, what the results mean for Kiwi operators, and, most important, how to optimise the signals that move the needle first.

‍

Why local signals matter more in conversational search

Traditional SEO relies on pages ranking for keywords. Local SEO adds proximity and Google Business Profile data. Conversational search folds both into one answer box that may include citations but rarely shows ten blue links. If your address, phone, or .nz domain fails to surface during the LLM’s reasoning, you risk disappearing from the short summary.

Local signals help the model answer questions like:

• “Best gluten-free bakery near me”
  
  
• “Trusted electrician in Karori”
  
  
• “Where can I rent snow chains between Christchurch and Tekapo”
  
  

Each request contains either an explicit place name or an implicit “near me” intent. The model must infer distance, service coverage and authority, job done only if it can parse structured data or authoritative citations.

‍

Our test setup

• LLM versions
  
  
  
  • Perplexity Pro
    
    
  • Gemini Advanced
    
    
  • ChatGPT-4o
    
    
• Queries (abbreviated for space)
  
  
  
  • Best podiatrist in Wellington
    
    
  • Where to buy kawakawa balm in Hawke’s Bay
    
    
  • Local courier that ships same day Auckland to Hamilton
    
    
  • Top roofers near Nelson with five-star reviews
    
    
  • Gluten-free bakery Wellington CBD
    
    
  • Cheapest kayak rental Lake Taupō
    
    
  • Eco cleaning service Dunedin
    
    
  • After-hours emergency dentist Christchurch
    
    
  • Cat boarding on the North Shore
    
    
  • Small-batch coffee roaster delivering to Queenstown
    
    
• Metrics captured
  
  
  
  • Business names surfaced
    
    
  • Distance or suburb mentioned
    
    
  • Citations shown or linked
    
    
  • Whether phone, address or website appeared
    
    
  • Changes after switching user location to Melbourne
    
    
• Exclusions
  We ignored generic advice answers (“Try searching Google”) and counted only explicit business references.
  
  

Headline results

Model: Perplexity
Average Mentions: 2.7 per answer
Mentions with Citations: 92%
Shift toward Australian Business: Strong shift (83 % switched to AU results)

Model: Gemini
Average Mentions: 1.9 per answer
Mentions with Citations: 48%
Shift toward Australian Business: Moderate shift (45 % to AU)

Model: ChatGPT-4o
Average Mentions: 2.3 per answer
Mentions with Citations: 58%
Shift toward Australian Business: Weak shift (22 % to AU)

Perplexity clearly leads on surfacing local New Zealand providers and backs most references with a source link. Gemini trails, often returning general information or a single brand. ChatGPT sits in the middle: it finds local businesses but is less consistent with citations.

‍

How each model parses locality

Perplexity

• Geographic bias
  Parses the user IP or manual location hint aggressively. When we typed “best podiatrist in Wellington” it placed three clinics within 10 km of the CBD. After switching to Melbourne, answers flipped to Victorian clinics.
  
  
• Structured data dependency
  Perplexity snippets almost always drew from pages with JSON-LD LocalBusiness markup. In four tests it skipped a highly rated provider whose site lacked schema even though the name appeared in online reviews.
  
  
• Citation first
  The engine includes footnote links by design. Pages on .nz domains and official associations (e.g. nzpodiatrist.org) counted more than social media profiles.
  
  

Gemini

• Blend of web and Google index
  Because Gemini has access to Google’s Knowledge Graph, we expected strong local output. Instead its model favoured well-optimised Google Business Profiles but returned fewer hard citations.
  
  
• Preference for suburb names
  It often included suburb rather than street addresses (“a popular bakery in Te Aro”) which can feel vague to users.
  
  
• Local shift moderate
  Changing location to Melbourne replaced only half the businesses. Gemini retained Kiwi examples if the brand also had shipping or service coverage in Australia.
  
  

ChatGPT-4o

• Reasoning over structure
  ChatGPT frequently referenced businesses mentioned only in long-form articles or forum posts, suggesting it weighs unstructured citations.
  
  
• Schema still helps
  Pages with correct address, areaServed, and openingHours fields surfaced more often in our coffee roaster and eco-cleaner queries.
  
  
• Cautious distance filter
  When user location was set to Melbourne it sometimes hedged: “While I am seeing many options in Melbourne, you might also consider shipping from New Zealand roasters like Flight Coffee.”
  
  

‍

The four strongest local signals (ranked)

Collecting all ten queries, we tallied which page or profile attributes appeared alongside a business name across the three engines.

Signal: LocalBusiness JSON-LD with full address object
Share of appearances where signal present: 68%
Notes: Critical for Perplexity and ChatGPT

Signal: Consistent NAP (Name Address Phone) across at least three citations
Share of appearances where signal present: 59%
Notes: Gemini draws mainly from Google profile and two corroborating sources

Signal: .nz TLD combined with contact address in footer
Share of appearances where signal present: 47%
Notes: Helps both Gemini and ChatGPT infer locality

Signal: High-authority review page (Yellow NZ, Google, Facebook)
Share of appearances where signal present: 44%
Notes: Ratings above 4.5 lifted result priority

Schema without corroborating citations scored lower than schema plus citations, but schema alone still beat pure citations without schema. That suggests a staged optimisation plan:

• Add or fix JSON-LD first
  
  
• Audit NAP across directories
  
  
• Secure review volume
  
  
• Maintain a .nz domain or sub-folder for NZ landing pages
  
  

Step-by-step optimisation for Kiwi businesses

1. Publish complete JSON-LD

For service companies use LocalBusiness, add these properties:

• name: legal name
  
  
• address: street, region, postal code, country NZ
  
  
• telephone: E.164 format +64
  
  
• areaServed: "New Zealand" or specific regions
  
  
• openingHoursSpecification: ISO days
  
  

If you already have a Product block on the same page, keep them separate to avoid conflicts.

2. Align NAP in top directories

• Google Business Profile
  
  
• Yellow.co.nz
  
  
• NZBN registry
  
  
• Facebook page “About” section
  
  

Ensure spacing and abbreviations match exactly. Perplexity and Gemini count minor variations as different entities.

3. Collect reviews and display rating markup

Use AggregateRating in your schema once you have at least five genuine reviews. Ratings appear verbatim in Perplexity answers.

4. Use location keywords in headings

ChatGPT relies on context from H1/H2 tags. Include the suburb or town once: “Emergency Dentist Christchurch”.

5. Create landing pages for each region you serve

Gemini rewards explicit geography. A single page at /auckland-electrician with clear contact details outperforms a generic “We cover all of New Zealand” statement.

6. Link social profiles in sameAs

Perplexity matches Instagram bios surprisingly well. Add your handle URL to the schema.

7. Keep site speed high

Large images delay Perplexity’s crawler which only fetches the first 100 KB. Compress hero banners below 150 KB.

8. Request a re-crawl

Ping each model via their feedback tools once changes are live. Both Perplexity and ChatGPT accept URL submissions.

‍

Limitations of our test

Our sample size is small and covers consumer queries only. Business-to-business or technical queries may behave differently. Each model updates weekly, so results can drift. Still, the relative ordering of signal strength matches what EnvokeAI Visibility Tracker sees across hundreds of clients.

‍

What to optimise first

• Structured data: errors here block every engine.
  
  
• Citation consistency: fix NAP mismatches.
  
  
• Ratings: build genuine review volume.
  
  
• Local landing pages: one per service area.
  
  

Do those four and your visibility across Perplexity, Gemini and ChatGPT will rise. Skip them and you risk leaving the door open for competitors five suburbs away.

Perplexity currently rewards schema plus citations more aggressively than Gemini or ChatGPT, making it the easiest win for Kiwi SMEs. ChatGPT adds weight to unstructured mentions, so local PR and blog coverage still matter. Gemini sits somewhere in the middle but leans on Google Business Profiles. The lines will keep shifting yet the underlying principle stays steady: structured, consistent local data gives language models the confidence to talk about you.

Ready to check how often your brand appears in AI answers? Run a free snapshot inside EnvokeAI Visibility Tracker and see where you stand. Then follow the optimisation steps above and watch your business move from invisible to indispensable in the next conversation a customer has with their favourite chat engine.

‍