A 1-Month Snapshot: How LLMs are Describing NZ Political Parties to Voters

There has been a lot of conversation already about the impact of AI on political campaigning in the lead-up to New Zealand’s General Election.

For voters, there are worries about fake news pages that pedal misinformation, fuelled by AI slop videos and images. One such page on Facebook, reported on by RNZ, has posted content that includes an AI-generated video of Prime Minister Christopher Luxon at Waitangi on Waitangi Day. When RNZ investigated the page, it had 5000 followers. In seven months, that number has jumped to 13000.

There are also concerns about AI usage within the Government itself, particularly with the public sector’s headcount on the chopping block following the 2026 Budget, which included a plan to cut around 8,700 public service roles by mid-2029.

However, the discussion so far has had one obvious gap. What are legitimate LLMs like ChatGPT, which 40% of Kiwis use daily to seek out quick answers and facts, saying about the parties on the ballot? Only recently has a pollster warned that “a significant chunk of Kiwis [will] use AI to tell them how they should vote.” On top of where things seem to be heading, New Zealanders are also less likely to have undertaken AI-related education (24%) compared to the global average (39%).

For government relations teams, this has become an obvious reputational risk with real-world consequences when pens are put to the ballot.

Table of Contents 

1. Who Does this Affect 
2. Methodology 
3. Where Does Each Party Rank for Overall AI Visibility?
4. Party Policy Strengths & Weaknesses, According to AI 
5. What This Means for All PR & Comms Teams 
6. Where To Start with AI Visibility 
   

Who Does this Affect

According to a recent report from One NZ, there is an increasing number of Kiwis (29%) who are “confident, frequent early adopters, who are optimistic around AI’s societal impact.”

Although trust has lagged, usage has certainly not. Recent numbers from OpenAI claim that ChatGPT alone has earned over 2 million Kiwi users. Of people over the age of 18, some 62.1% of that usage is going towards getting information or answering questions.

Our Methodology 

To discover how LLMs are answering queries on New Zealand’s political parties, we created several targeted AI Visibility Dashboards built on manual and Streem AI-suggested prompts such as “what NZ political party is better for the economy” and “what new zealand political party is best for education?” Streem AI then identified New Zealand’s political parties within the data, which served as our entities in the platform that we looked at to surface insights on AI mentions, sources, and sentiment.

Where Does Each Party Rank for Overall AI Visibility?

In the table of results below, the Green Party of Aotearoa New Zealand finished first as the most-mentioned party across ChatGPT, Claude, Gemini, and Grok.

However, the National Party owned a higher average position within LLM responses, coming in at 1.2 compared to the Green Party’s average of 2.7. This means that although the Green Party is showing up slightly more consistently across the board, the National Party is more likely to be mentioned first.

Further down, party rankings diverged sharply by platform. Each produced a materially different picture of New Zealand’s political landscape even when they were asked the same questions. Significantly, ChatGPT mentioned just two parties, Labour and the Green Party, in any of its responses about general party information in the month of September. Each of the four platforms also laid out a different top party by mentions.

Most-Mentioned Political Parties by AI Platform 

But what sources are these AI platforms referencing? And do those sources differ between platforms? Is a ChatGPT user getting different political information than a Claude user would? By looking at top-cited sources by platform, we were able to find out, in a 1-month snapshot, which LLMs were letting parties describe themselves, and which prioritised more independent sources.

By industry, government domains were cited most, making up 38.6% of all sources, followed by uncategorised domains (23.5%), nonprofit domains (12.1%), and news and publishing domains (11.4%), further proof of the clear link we’re seeing between earned media, AEO and GEO.

Drilling down into the specifics of these sources, the dominant insight from this analysis is a reliance on party-owned content to answer subjective questions about policy success. The top three most-referenced domains across four platforms were those owned by the National, Green, and Labour parties. Of the major parties, only ACT New Zealand and Te Pāti Māori did not have their domains feature in the top ten. In fourth place, Wikipedia won 8.7% of the domain share. Even as an information resource that makes “no guarantee of validity”, it is prioritised because it is structured, verified by humans, and constantly updated. RNZ and the NZ Herald were the top mastheads, with citations driven by explainer-type content. Research database ebsco.com also made an appearance, alongside third-party websites partymap.co.nz and nzvotecheck.com, both created as guides for voters. While domain authority, that being a 1-to-100 score which rates a website based on how likely it is to rank on a search engine results pages, is often key to earning AI citations, the latter both have authority scores of 2. This suggests that text content and domain names contributed to query relevance.

Of the four platforms, ChatGPT pulled the most content from the parties themselves. It is the only platform to source from party-hosted files as well as online pages, meaning the answers it gives largely reflect what parties are saying about themselves. Its two most-referenced URLs were The Green Party’s 2026 Manifesto and Labour’s Policies page.

To summarise the trends for each platform independently: 

• ChatGPT was a clear outlier, describing each party through their owned content rather than pulling from independent secondary sources.

• Google AI Overview leaned on video/social platforms, even for a factual political query, using a mix of news media and influencer content.

• Grok and Perplexity drew heavily from institutional and civic sources, plus mainstream press. This suggests that the platforms were using more of a "how government works" framing rather than a "who are the parties" framing.  
  

• Claude weighted reference and encyclopedic sources most heavily. Wikipedia and Britannica alone made up 30% of citations, and it's the only platform with no party-run site in its top ten. However, it also had the shakiest outlier domains of any platform.

Top-Cited Sources by AI Platform

Party Policy Strengths and Weaknesses, According to AI

When these same LLMs are asked about each political party’s policy strengths and weaknesses, mention volumes and sentiment trends change. We looked into how different parties are held up to the light for eight policy areas including Economy, Housing, Health & Welfare, Education, Environment & Energy, Crime, Infrastructure & Transport, and Security & Foreign Affairs.

Most-Mentioned Parties by LLM Sentiment 

Overall, LLM sentiment largely mirrored historical party-brand associations, with The National Party ranking higher for Economy, Housing, and Infrastructure & Transport and Security & Foreign Affairs, and The Green Party doing better with Health & Welfare, Education, and Environment & Energy. ACT New Zealand was the only other party that earned a first rank for positive sentiment, specifically for Crime.

The below is a response from Grok to the query “what new zealand political party is best for addressing crime” that contributed to this result.

“National and ACT emphasize tougher sentencing, more police, and gang crackdowns, with the coalition reporting drops in violent crime and ram raids. Labour and Greens prioritize prevention and rehabilitation. "Best" depends on priorities.”

Just two sources were used to land on this conclusion, the National Party’s ‘Law and Order’ page and a February 2026 article from RNZ following the release of the New Zealand Crime and Victims Survey.

What This Means for All PR & Comms Teams

The dynamics that we’ve observed here go beyond New Zealand’s political landscape. The same trends apply to every organisation that has an audience using AI Search. If over 2 million New Zealanders are using ChatGPT alone, the combined number of users across all of today’s leading LLMs no doubt necessitates urgent strategic consideration.

Here are some takeaways from our analysis that apply to any PR or communications teams, not just the government relations professionals.

1. Owned content is key, but gaps in earned and independent sources are risky.

The top three most-cited domains in this snapshot belonged to the National, Green and Labour parties, with eight of ChatGPT’s top ten domains owned by government entities. However, what this means is that outdated policy pages, FAQs, “about” pages, and downloadable reports are now a live reputational risk.

What we’ve also seen is that relevance can beat authority. Voter-guide sites partymap.co.nz and nzvotecheck.com earned strong citation shares despite having little to no domain authority due to the density of relevant keywords and domain naming. Major mastheads and other independent sources also feature consistently, with no LLM citing just one media type.

Where independent coverage is thin, AI will fill the gap with whatever is available. Communications strategies that work with this, and prioritise an effectively dispersed PESO strategy with key messaging will be able to earn citations that can combat misinformation and negative sentiment.

2. Visibility and brand narratives will differ between LLMs.

Discrepancies between how different AI platforms talk about your brand mean that your audience is being segmented by the LLMs they use.

In this research, ChatGPT, Claude, Gemini, and Grok each surfaced a different top party, and ChatGPT mentioned just two parties across its responses. As a result, being visible on one LLM doesn't mean you're visible on another, and how your brand is appearing on each LLM will differ. It’s safer to assume that your position on one doesn’t reflect the rest.

3. Expect to be judged by your legacy brand associations.

With LLM sentiment of political parties and their policy strengths and weaknesses largely mirroring long-standing perceptions, combating these perceptions calls for well-sourced messaging over time to change the pattern.

With LLM sentiment analysis, start with where negative sentiment is being sourced from. Can you do outreach to those media outlets? Or, looking at your positive sentiment, can you do outreach to the outlets you’ve been successful with, and defend against harmful messaging in the content you’re having issues with?

4. LLM answers often draw from a small sample of sources.

Just as with the sample response from Grok, which rested on just two sources to form a conclusion on which party had the best track record with handling crime, a small sample of sources can have a big influence on answers related to your organisation. When only a handful of sources drive a response, a single page you publish (or fail to publish) can define the narrative on an LLM. It’s key to find out which pages are doing that work for you, beneficially or not.

Where to start with AI Visibility 

With so many Kiwis turning to AI for quick answers, first impressions of your organisation are increasingly being engineered by LLMs. The teams that understand where those answers come from, and shape those sources responsibly, will be best placed to make sure the picture is accurate.

If your organisation is looking into AI Visibility as a priority area, start by booking in a meeting with our team to explore how Streem's AI Visibility Dashboards combines LLM data with Media Monitoring, Social Listening, and Outreach in one simple workflow.