What Sources Does ChatGPT Trust? A Guide for NZ Business Owners

ChatGPT and Perplexity do not draw from the same pool of sources, and understanding which pool is which is the most useful thing you can do for your AI search visibility in 2026. A study analysing 680 million AI citations found only about 11% of domains appear in both platforms. ChatGPT leans toward established encyclopaedias, major directories, and editorial publications; Perplexity leans toward recent community discussions, Reddit threads, and real-time web content. If your Tauranga plumbing business or Auckland dental practice is only visible in one platform’s preferred source set, it is effectively invisible to the other.

Why does it matter which sources ChatGPT and Perplexity trust?

The sources an AI tool cites determine which businesses it names, recommends, or links to when someone asks a question about your industry. If your business appears in those sources, the AI may mention you. If it does not, you simply do not feature in that answer.

More New Zealanders are starting their research for local services inside AI tools rather than typing into a Google search bar. A homeowner in Hamilton asking “who does good heat pump installs near me?” is increasingly getting their first answer from ChatGPT or Perplexity, not from scrolling through a page of search results. The business that appears in that AI answer has a significant advantage over the one that does not.

This matters for three concrete reasons.

First, ChatGPT and Perplexity have almost entirely separate source preferences. Research covering more than 150 client engagements found that ChatGPT cites sources in 41% of its answers. Perplexity cites sources in 87% of its answers, the highest citation rate among the major AI platforms. Despite both being widely used, only about 11% of domains appear in both platforms’ citation sets. Building visibility on one does not automatically build it on the other.

Second, your Google ranking is not a reliable proxy for AI visibility. AI engines use their own retrieval logic. A Wellington physiotherapy clinic sitting at the top of Google’s results may receive zero mentions in AI answers if it lacks the signals those platforms look for. The two systems share some inputs but operate independently.

Third, for New Zealand businesses the consequences of weak local signals are steeper than in most markets. Research by EnvokeAI found that when a test query was switched from an Auckland location to a Melbourne location, 83% of Perplexity’s New Zealand business mentions flipped to Australian results. Weak local signals in AI search do not produce lower NZ rankings; they produce Australian competitors appearing in place of local businesses entirely.

Understanding which sources ChatGPT and Perplexity trust is therefore the starting point for making sure your business appears when potential customers are deciding who to call.

What sources does ChatGPT actually rely on?

ChatGPT primarily draws on Wikipedia, established editorial publications, major directories, and the Bing search index for live web content. Wikipedia appears in roughly 47% of factual AI answers. For commercial queries, ChatGPT retrieves content through Bing, which means your website needs to be indexed by Bing (not just Google) for ChatGPT to draw on it directly.

Research analysing more than 21,000 individual AI citations found that ChatGPT tends to cite fewer sources per query than Perplexity, but each cited source carries substantially higher influence on the generated answer. Being mentioned once in a source ChatGPT trusts carries more weight than scattered mentions across many lower-authority sites.

Separate research identified that 73% of all AI citations come from just seven source types: Wikipedia and Wikidata, community platforms such as Reddit and Quora, editorial publications, industry-specific sites, government and academic domains, aggregate review platforms, and a long tail of other sources spread across thousands of smaller domains. Understanding which of these you appear in is the most direct audit you can run on your current AI visibility.

The source categories ChatGPT favours, roughly in order of influence:

Wikipedia and Wikidata. These appear in approximately 47% of factual query answers across both platforms. ChatGPT absorbed enormous amounts of Wikipedia content during its development, and Wikipedia’s structure (neutral tone, dense cross-linking, consistent entity naming) aligns closely with how it encodes and retrieves information about people, places, and organisations.

Government and institutional domains. Sites with .gov and .edu suffixes account for around 9% of AI citations. For New Zealand businesses, a mention in a Ministry of Business, Innovation and Employment document, a district council publication, or a New Zealand university resource carries real weight. Professional registration with a recognised body matters here too: when the Plumbers, Gasfitters and Drainlayers Board lists your name, that entry exists on a .govt.nz domain that AI engines treat as authoritative.

Established editorial publications. International outlets such as Reuters and the Financial Times appear frequently in ChatGPT’s source set. In the New Zealand context, the New Zealand Herald, Stuff, and RNZ carry similar signals for NZ-specific queries. Media coverage is hard to earn but highly effective when you get it; a genuine news angle (a notable project, a local problem you have solved, an industry-first service) is what tends to make it happen.

Recognised directories and review platforms. Yelp and TripAdvisor are commonly cited in the international market. For New Zealand, Yellow.co.nz, Google Business Profile, and trade association directories serve the same function. Being absent from these is a clear gap in your AI visibility.

Your own website, via Bing. ChatGPT uses Bing to retrieve live web content for current and commercial queries. Many New Zealand businesses have worked hard on their Google rankings but have never checked their Bing indexing status. Logging into Bing Webmaster Tools and verifying your site is indexed there is a straightforward step that many businesses have skipped entirely.

The volume of third-party mentions also matters significantly. Brands cited across 25 or more trusted third-party sources appear in AI answers at four times the rate of brands with fewer than ten third-party mentions. This is not about chasing every possible link; it is about being consistently present in the authoritative, structured sources within your industry and region.

What sources does Perplexity trust most?

Perplexity uses a fundamentally different retrieval approach from ChatGPT, with Reddit accounting for 46.7% of its cited content. Perplexity was built as a live answer engine that searches the web fresh for every query, and it strongly favours recent, community-validated content over static reference material.

Perplexity cites sources in 87% of its answers and pulls from three to four sources per query on average. Its retrieval pipeline evaluates each potential source across five criteria: relevance, recency, entity clarity, extractability (how easily the answer can be drawn from the text), and authority. A mid-tier website that clearly and directly answers the specific question can appear alongside high-authority sites if the other criteria are strong.

The source types Perplexity favours most:

Reddit. At 46.7% of cited content, Reddit is Perplexity’s single most dominant source category, nearly five times ChatGPT’s rate of Reddit usage. This reflects both licensing arrangements and Perplexity’s architectural preference for real-time discussion. For a Nelson landscaper or a Southland agricultural contractor, genuine conversations happening in relevant New Zealand Reddit communities feed directly into Perplexity’s answer pool.

Fresh, recently updated content. Perplexity has a clear freshness preference, processing new content within roughly 7 to 10 days. A blog post published recently about current waterproofing costs in Christchurch is a stronger signal for Perplexity than an older, more comprehensive guide on the same topic, because recency is an explicit criterion in its selection pipeline.

Academic journals and research publications. Perplexity attracts a research-oriented audience and cites academic sources frequently for health, technical, and professional queries. For an optometrist in Dunedin or a structural engineer in Wellington, content grounded in professional standards or citing published research will perform better in Perplexity than generic promotional copy.

Mid-tier sources with specific, factual answers. Perplexity is more tolerant than ChatGPT of lower-authority websites, provided the content is specific, factual, and directly responsive to the query. A well-structured FAQ page from a Whanganui business that explains a practical topic with genuine clarity can appear in a Perplexity answer even if the domain has no traditional SEO authority, as long as the content directly and completely answers the question.

Combined, Reddit and Quora account for 24% of citations across both AI platforms. But Perplexity’s 46.7% Reddit figure alone shows how far the two platforms diverge on this source type, and why a single approach cannot serve both platforms well.

Why do ChatGPT and Perplexity use such different sources?

The two platforms were built with different architectures and different goals. ChatGPT is a reasoning tool built on a large body of pre-learned content, supplemented by Bing for live retrieval. Perplexity was designed from the start as a live web answer engine, making source recency and breadth central to how it operates.

This architectural difference produces very different source preferences in practice.

ChatGPT processed vast amounts of content during its development, building a strong internal model of what constitutes authoritative information. For live queries, it supplements this with targeted Bing searches, generally selecting sources that align with established patterns: high-authority, well-structured, consistently reliable. This favours Wikipedia, government domains, and established editorial brands.

Perplexity retrieves content fresh for every query using its five-criteria selection pipeline. It casts a wider net and evaluates sources more dynamically, which is why recency and community corroboration carry so much weight. A recent Reddit thread answering someone’s exact question can outperform an authoritative but older webpage on the same topic.

The result is that only about 11% of domains appear in both platforms’ citation sets, based on a study of 680 million AI citations. This is the most important number in AI search visibility for 2026. It means that improving your ChatGPT visibility and improving your Perplexity visibility are largely separate projects with limited overlap between them.

For New Zealand business owners, the practical implication is that these two platforms need to be thought of as separate audiences with separate content preferences, not as a single “AI search” target. The foundational work benefits both: a clean website, consistent directory listings, and genuine reviews all help with ChatGPT and Perplexity alike. It is the platform-specific work above that foundation where the two diverge.

For more on how ChatGPT’s search functionality differs from a standard Google query, see How is ChatGPT search different from Google?.

What does this mean for a New Zealand business specifically?

New Zealand businesses face a specific risk that most US-based AI search advice overlooks: when local signals are weak, AI engines default to Australian results. Research by EnvokeAI found that switching a test query from Auckland to Melbourne caused 83% of Perplexity’s New Zealand business mentions to flip to Australian competitors. Strong NZ-specific signals are not optional extras; they are the baseline requirement for appearing in AI answers at all.

The underlying reason is data density. In the United States, there is a thick ecosystem of authoritative local sources: hundreds of city-specific publications, deep directory data, layers of government citations. AI engines can find dozens of corroborating sources for almost any local business query. In New Zealand, that ecosystem is thinner. For many local service queries, an AI engine may find only three to five relevant sources. In this environment, a single missing or inconsistent listing is not a minor gap; it can be the reason your business is completely absent from AI answers.

This creates genuine opportunity alongside the risk. Because the field of well-structured New Zealand local content is not yet crowded, a focused and consistent effort can achieve AI visibility that would require significantly larger investment in a more competitive market.

The NZ-specific signals that matter most:

LocalBusiness schema with NZ-specific detail. Pages with JSON-LD LocalBusiness schema that includes a full New Zealand address, a +64 phone number in the correct international format, an areaServed field specifying New Zealand regions, and opening hours in ISO format appeared in 68% of business mentions across queries tested in New Zealand research. This markup tells AI engines unambiguously that your business is based in New Zealand and serves New Zealand customers.

Consistent NAP across NZ directories. AI engines cross-reference your business name, address, and phone number (NAP) across multiple sources. Consistent details across at least three of the major NZ directories, specifically Google Business Profile, Yellow.co.nz, and the NZBN registry, correlated with 59% of AI appearances in New Zealand-focused testing. A discrepancy as minor as “Street” versus “St” in your listed address can fragment the signal an AI engine receives and cause it to treat your listings as different entities.

A .nz domain. A .nz domain with a physical New Zealand address in the footer helped AI engines correctly infer locality in 47% of observed mentions in NZ-specific research. This signal carries considerably more weight than a .com domain, because .com is geographically neutral while .nz is unambiguous. If your business currently operates on a .com address, securing the .nz equivalent and using it as your primary domain is worth serious consideration.

NZ-specific review platforms. High-quality reviews on Google Reviews, Yellow NZ, and Facebook, with ratings above 4.5 stars, appeared in 44% of AI mentions in New Zealand research. Reviews on industry association sites (the websites of registered professional bodies for your trade or profession) also count for NZ-specific queries more than generic social profiles do.

Original New Zealand data. Pages that publish genuinely NZ-specific information, such as a 2026 survey of commercial cleaning rates in Wellington or a breakdown of standard callout charges for Bay of Plenty plumbers, are uniquely citable because no other source has that data. If your Gisborne electrical business publishes clear, honest pricing guidance for the local market, an AI answering that query has a specific reason to cite you rather than a generic national source.

For businesses in smaller centres, from Invercargill to Whanganui, the opportunity is particularly clear. AI tools need local sources, and local businesses willing to publish clear, structured, specific information will become those sources.

What are the biggest myths about AI source trust?

The most widespread myth is that ChatGPT and Perplexity draw from essentially the same sources, so one set of actions will build visibility on both platforms simultaneously. They do not, and it will not. A second widely held belief is that ranking at the top of Google guarantees AI visibility. It does not. Clearing up these misconceptions will save you from spending time and money in the wrong places.

Myth: Both AI tools use the same sources. Research analysing 680 million citations found only 11% of domains appear in both platforms’ source sets. What earns a ChatGPT citation (Wikipedia, established directories, editorial mentions, Bing-indexed commercial content) is almost entirely different from what earns a Perplexity citation (Reddit threads, fresh content, community-corroborated discussions, real-time web pages). There is no single “AI search” optimisation approach that serves both platforms equally well.

Myth: ChatGPT only trusts Wikipedia and big news brands. Wikipedia is ChatGPT’s most heavily cited source type, appearing in about 47% of factual answers. But Reddit accounts for roughly 11% of ChatGPT citations too. ChatGPT also draws from directories, government sites, well-structured commercial pages indexed in Bing, and industry-specific sources. The source set is broader than most people realise, and there are practical entry points beyond earning media coverage.

Myth: Perplexity is too small to matter. Perplexity exceeded 500 million monthly queries in 2026. For audiences researching professional services, health questions, or technical matters, it is often a primary research tool rather than a secondary one. An Auckland engineering firm or a Christchurch specialist clinic may find that Perplexity visibility is as commercially significant for their specific audience as traditional Google rankings.

Myth: High Reddit citation rates mean Reddit is editorially verified. Reddit’s dominance in AI citations reflects partly licensing agreements between AI platforms and Reddit, rather than AI platforms editorially judging Reddit as reliable. This means AI tools can cite Reddit content that is unverified, anecdotal, or occasionally incorrect. The implication for your business is to encourage genuine, accurate customer discussions in relevant communities, not to seed promotional content. Promotional threads that attract hostile community responses can produce accurate but damaging citations.

Myth: Schema markup is a technical detail that does not change outcomes. Businesses with comprehensive structured data covering Organisation, Article, FAQ, HowTo, and Person schema where applicable are cited at 2.3 times the rate of businesses with thin or homepage-only schema. This figure comes from 2026 research covering more than 150 client engagements. Schema is among the most reliably evidenced actions available to NZ businesses for improving AI citation rates, and it is something any competent web developer can implement.

Myth: Ranking at position one on Google means AI will find you. ChatGPT shows the weakest alignment with Google’s top search results of the major AI platforms, drawing from sources that standard search often overlooks entirely. Perplexity aligns more closely with Google but still has its own independent source preferences. Your Google SEO investment builds a useful foundation but is not sufficient on its own to drive AI visibility.

A broader point: ChatGPT, Perplexity, Google AI Overviews, and Claude each operate by their own distinct source logic. Treating AI search as a single, unified discipline produces mediocre results across all platforms. Approaching it as a set of related but distinct challenges produces meaningfully better outcomes.

How do I get my NZ business into the sources these AI tools cite?

Seven practical actions account for the majority of AI citation outcomes in 2026: establishing a presence on Wikipedia or Wikidata, pursuing inclusion in industry roundup lists, building genuine community presence on Reddit, adding comprehensive schema markup to your website, structuring content as question-and-answer pairs, building review volume across multiple platforms, and maintaining consistent entity information across all directories.

1. Create a Wikipedia or Wikidata entry. Wikipedia content appears in approximately 47% of factual AI answers. If your business is notable enough by Wikipedia’s standards (meaning it has been independently covered in reliable sources without you initiating that coverage), a well-written, neutral, sourced Wikipedia article is the highest single-impact move available. If a full Wikipedia article is not appropriate for your business size, a Wikidata entry with verified links to your website, LinkedIn profile, and any news coverage still signals a canonical identity record to AI engines. A Palmerston North construction company covered in regional news and listed in multiple industry directories has a reasonable case for a Wikidata entry.

2. Pursue inclusion in industry roundup lists. Lists published as “Top X services in [region]” or “Best Y providers in New Zealand” generate derivative mentions as other sites link to or reference them. Identifying 10 to 15 relevant lists in your industry and region, then pitching for inclusion with a genuine angle (a specific project, a local specialisation, or original data you have published), can produce a sustained increase in third-party mentions over the following 60 to 90 days.

3. Build genuine presence on Reddit and community platforms. For Perplexity specifically, Reddit accounts for 46.7% of citations. You do not need to run a complex social operation to benefit from this. Identifying three to five subreddits or Quora topics where potential customers in your industry ask genuine questions, then providing accurate and helpful answers without promotional intent, creates exactly the kind of content Perplexity’s pipeline favours. An Otago builder who consistently answers questions about earthquake-strengthening requirements in older homes is building Perplexity visibility while demonstrating expertise to real people at the same time.

4. Add comprehensive schema markup to your website. Every commercial page should carry relevant structured data. For a New Zealand service business, the priority list is: LocalBusiness schema with a full NZ address and +64 phone number, FAQ schema on your frequently asked questions pages, Article schema on blog posts, and HowTo schema on process-explanation pages. Include sameAs links pointing to your verified profiles on Google Business Profile, LinkedIn, Facebook, and any relevant industry association directory. Businesses with comprehensive schema are cited at 2.3 times the rate of those without it.

5. Structure your content as question-and-answer pairs. Pages where three to five H2 headings are phrased as genuine customer questions, each followed immediately by a clear 40 to 80 word answer in the first paragraph, are cited at 1.8 times the rate of pages using statement-style headings. A page with a heading like “How much does a kitchen renovation cost in Auckland?” outperforms one titled “Kitchen renovation pricing.” The answer should arrive before any additional context or caveats, not after them.

6. Build review volume across multiple platforms. Research shows that 100 reviews at 4.5 stars is cited 1.6 times more often than 20 reviews at 5.0 stars. Volume and spread matter more than a perfect score. Aim for consistent new reviews each month across at least three platforms: Google Reviews, Yellow NZ, and one industry-specific platform such as a trade association directory or sector review site. Responding professionally to all reviews, including critical ones, signals active business management to both AI tools and potential customers.

7. Maintain consistent entity information across all directories. AI engines filter ambiguous business records. If your business name, address, and phone number differ across directories, even in minor formatting details, the AI may not be able to confirm which listing is authoritative and may exclude your business from answers as a result. Audit every directory where your business appears and standardise the information completely. For New Zealand businesses, the priority directories are Google Business Profile, Yellow.co.nz, the NZBN registry, Facebook, and your relevant industry body’s member listing.

For a comprehensive guide to appearing in AI answers across the major platforms, see How to get your NZ business cited in ChatGPT, Gemini and Perplexity.

How do I check whether I am already showing up in AI answers?

The fastest way to assess your current AI visibility is to run ten to fifteen test queries directly in ChatGPT and Perplexity, using the same phrases your potential customers would type, and record whether your business appears, what is said, and which sources are cited. This takes roughly 20 minutes and produces a concrete baseline you can measure against as you implement changes.

Useful query formats to test:

“Best [your service] in [your city or region] 2026”

“[Your service] cost in [your region]”

“How to choose a [your service type] in New Zealand”

“Top [your industry] providers in [your region]”

“[Your service] versus [alternative approach]”

Run each query in both ChatGPT and Perplexity and note who appears in the answer, whether your business is named, which URLs are cited, and whether any information about your business is accurate. If a competitor appears but you do not, examine the cited sources for that competitor. Understanding exactly which directory listing, Reddit thread, or piece of content drove the citation tells you directly where to focus your own effort.

Pay particular attention to accuracy. AI tools can present outdated or incorrect information, particularly for businesses whose details have changed recently. Perplexity tends to update within 7 to 10 days of a source changing; ChatGPT Search may take 14 to 30 days. If you find incorrect information in an AI answer, updating the original source page directly is the most reliable way to correct it, rather than trying to influence the AI output indirectly.

Running the same test queries every three months gives you a practical way to track whether the actions you take are moving your AI visibility in the right direction.

For a step-by-step guide to this testing process, including how to interpret what you find and what to do next, see How to Check if Your Business Appears in AI Search.

Does my Google ranking still matter, or is AI search a separate thing?

Google ranking still matters, and it provides a genuine foundation for AI visibility, but it does not guarantee it. The two systems overlap partially, with Perplexity aligning more closely with Google’s top results than ChatGPT does. ChatGPT shows the weakest correlation with traditional Google rankings of the major AI platforms, drawing frequently from sources that standard search overlooks entirely.

The most accurate way to think about this is that Google SEO and AI visibility are related but distinct disciplines with overlapping foundations and diverging upper-level tactics.

The foundation is shared. A clear, accurate, well-structured website with consistent business information, a strong Google Business Profile, and genuine customer reviews helps both Google rankings and AI visibility. None of this is either-or investment.

Above the foundation, the tactics diverge. Optimising for Google primarily means earning authoritative inbound links, publishing content that matches search intent, and maintaining strong technical site performance. Optimising for AI visibility means adding comprehensive schema markup, structuring content as question-and-answer pairs, building Bing indexing alongside Google indexing, pursuing third-party mentions in directories and editorial sources, and cultivating review volume across multiple platforms.

For New Zealand businesses, there is a practical reason to treat AI visibility as a distinct priority rather than assuming it follows automatically from Google SEO effort. The New Zealand search landscape is competitive at the Google level in most professional service categories. Dental clinics in Auckland, building contractors in Christchurch, and physiotherapy practices in Wellington have been investing in Google SEO for years. The AI search landscape in New Zealand is less developed, which means the opportunity to establish visible presence with a focused effort is genuine and available now.

A Rotorua tourism operator, a Lower Hutt accountancy practice, or a Waikato veterinary clinic that builds clear NZ-specific signals today is positioning itself ahead of businesses that are waiting to see how AI search develops. The businesses appearing in AI answers today are mostly those with structured data, consistent directory listings, and question-focused content. The technical bar is meaningful but well within reach for any professional service business in New Zealand willing to put in consistent, targeted effort.