ChatGPT was publicly available for less than a year the last time Kiwis headed to the general election polls. Now, AI tools are being used by three quarters of New Zealanders.
Kiwis across the country are currently asking AI for information about the political parties and the issues most important to them.
This will be the first New Zealand election influenced by the answers AI is providing.
That's a concerning thought, but it doesn't mean AI tools are trying to put a particular party in power. AI's influence is much more subtle than this.
For every political prompt or question being asked, AI is scanning the internet and providing the information it thinks is most appropriate. It steps into a new role as the editor of what is on the internet.
Can we trust the immensely complicated AI algorithms will do this fairly with an even balance across the political spectrum?
Well… Kind of. That's not a good answer, but ultimately the AI is incredibly smart and works hard to present us with balanced, unbiased views. However, that doesn't mean the AI answers don't have faults.
AI shapes the choice before you make it
I've been working with the team at Onsomble, a Kiwi start-up that has built a world class AI search tool. We tested 174 election-related questions across ChatGPT, Gemini, Google AI Mode and Google AI Overviews. The study captured 815 answers across issues including housing, healthcare, jobs, tax and transport.
While it's clear AI isn't telling us who to vote for, it is deciding which parties to compare and which policies deserve attention.
Even which AI platform you're using can make a meaningful difference to the answers you're given. For example, in one matched test, when the same question was delivered to four different tools the number of political parties given meaningful consideration ranged from three to six.
In a way this reminds me of the smaller political parties fighting to be included in the official media debates. For some prompts the AI is making decisions on which parties are heard.
The invisible sources behind the answer
Where the information comes from is just as interesting.
The AI algorithms typically prioritise high authority sources. This is especially true for high-stakes topics like politics, finance and health.
This means answers for political questions often stem from party manifestos, government agencies, established media or election authorities. However, our research revealed this isn't always the case.
Kiwi Coaches, an Auckland bus operator, appeared in source records associated with 54 AI answers across 20 questions. Its election guide compared transport policies across seven parties.
A Grey Power policy comparison was associated with 38 answers across 17 questions. But that comparison contained responses from only five of the seven highest polling parties because one declined and another did not respond.
Both these websites had already made choices before AI ever encountered the information. AI makes another layer of choices when it turns those sources into an answer.
The missing media link
Why does AI sometimes pull information from websites like Kiwi Coaches instead of our most established news outlets, such as RNZ, Stuff and NZ Herald?
Part of the answer is that some of our major media outlets restrict AI access to their content.
They understandably don't want the big AI companies benefiting from their journalism without paying for it. Fair enough.
But these restrictions aren't going to stop New Zealanders from turning to AI for information. And that raises a bigger question about the role of independent journalism in society as more people rely on AI-generated answers.
If credible New Zealand journalism is harder for AI to access, what information is taking its place?
It's unclear how this will pan out. We're already seeing partnerships between AI companies and media organisations around the world. But New Zealand is a small market, and I suspect we'll be waiting some time before similar arrangements become widespread here.
In the meantime, I'd encourage everyone to make an effort to read the excellent journalism being produced in this country. Let's support it with paid subscriptions.
AI can summarise and explain the news, but it can't replace the original reporting and investigative work that makes quality journalism possible.
Small errors can have big consequences
Another risk with AI is accuracy.
The most concerning AI error is not always an outrageous hallucination. It can be one small mistake within an otherwise useful answer.
The Onsomble research found examples including an incorrect deadline for returning some overseas voting papers, outdated FamilyBoost settings, and a Dunedin hospital opening being described in a way that blurred the difference between a building opening and patient services actually beginning.
These may be small errors, but there are undoubtedly more out there. As New Zealanders become more reliant on AI tools giving them the answers they need to make decisions, there will be more consequences that stem from trusting the wrong information.
Limiting AI's influence
None of this means AI is going to decide the election, but I do believe it's influencing how many New Zealanders are processing political information.
The most forward-thinking businesses are already paying close attention to how they're represented by AI tools, but are our political parties? Are their party websites optimised so AI can easily find, understand and accurately represent their policies? Are the political communications plans factoring in which third-party websites are sharing what they want voters to find?
While the AI algorithms are complex, they can be influenced. This can impact what products a consumer purchases, or what party they ultimately vote for.
As New Zealand's AI use continues to skyrocket, it becomes more important to analyse the answers we are given and question where the information comes from.
We also can't rely on AI summaries as our only source of truth, we must continue to engage with traditional media and directly with the sources.