Political advertising on Facebook and Instagram is becoming harder for voters to see and understand as a whole. Parties can begin with broad audiences, but Meta's automated advertising systems can optimise delivery towards the people most likely to respond — potentially producing very different election campaigns for different voters.
eLocal Report: Based on reporting published by The Conversation, Algorithms are choosing who sees what political ads. NZ urgently needs transparency
A social media user views content on a smartphone. Photo: Catherine Falls Commercial/Getty Images.
A recent examination of New Zealand political advertising has raised a deceptively simple question ahead of the election:
When a political advertisement appears on your phone, who actually decided that you should see it?
The obvious answer would seem to be the political party paying for the advertisement.
Increasingly, however, that is only part of the story.
Political parties can decide what advertisements to create and define the audiences they initially want to reach. But automated advertising systems can then optimise where those advertisements are delivered according to predictions about which users are most likely to respond.
That creates an important distinction between political targeting and algorithmic delivery.
And for voters, it makes political advertising considerably less transparent.
The Numbers That Started the Question
A recent RNZ analysis examined political advertising running on Facebook and Instagram during the 2026 election year.
Of more than 500 advertisements National had run, 92 were seen exclusively by men. Those advertisements attracted as many as 3.9 million views and involved spending of up to $21,000.
Only one advertisement was recorded as being seen solely by women. It ran for one day, cost less than $100 and National told RNZ it had not placed it.
ACT displayed a similar gender imbalance. Fourteen of its advertisements were seen solely by men, while none were seen solely by women.
The Greens showed the reverse pattern, with 13 advertisements seen only by women and none exclusively by men.
Those figures appear, at first glance, to show political parties deliberately separating voters according to gender.
But that conclusion may be too simple.
Targeting Isn't Necessarily the Same as Delivery
Political advertising on Meta's platforms is not new in New Zealand elections.
What has changed is the technology deciding which advertisements appear in front of particular users.
Traditionally, digital advertising involved an advertiser defining an audience according to characteristics such as age, gender, location or interests.
The platform then attempted to deliver the advertisement to that group.
Artificial intelligence and automated optimisation have increasingly changed that relationship.
Meta's Andromeda advertising system is designed to narrow enormous numbers of potential advertisements and variations down to those considered most relevant to an individual user.
Meta's Advantage+ advertising tools can also automate elements of audience selection, budget allocation and placement.
Rather than advertisers necessarily specifying increasingly narrow groups, modern advertising systems can begin with broader audiences and then analyse which people respond to particular advertisements.
The consequence is significant.
A political party might initially make an advertisement available to men and women across a wide age range.
But if the platform determines that men aged 45 to 64 are more likely to click, watch, share or otherwise engage with it, delivery can increasingly favour those users.
The campaign may not have explicitly instructed the platform to exclude anyone.
The result can nevertheless be an audience that looks highly targeted.
The Algorithm Has Its Own Objective
This matters because an advertising platform and a democratic election campaign are trying to achieve different things.
A political party wants votes.
A voter ideally wants enough information to compare parties and policies.
An advertising platform is attempting to deliver advertisements efficiently according to the objectives selected by advertisers and the behaviour its systems predict from users.
Those goals do not necessarily produce the same outcome.
If showing a particular political message to one group costs less or generates greater engagement than showing it to another, an optimisation system has an incentive to concentrate delivery where it performs best.
Researchers have been examining this phenomenon for several years.
A 2019 study of Facebook advertising delivery found that the platform's delivery system could produce skewed audiences by gender and race even when advertisers used inclusive targeting.
Other research has suggested voters statistically less likely to support a political party can sometimes cost considerably more for campaigns to reach.
An analysis of ten Indian elections found Facebook charged the governing Bharatiya Janata Party lower advertising rates than rival parties in nine of them, potentially allowing it to reach more voters for the same expenditure.
More recently, a study of the 2024 European Parliament elections examined more than 110,000 advertisements across 25 countries.
Researchers found men were significantly more likely to receive advertisements from populist and far-right parties even after attempting to account for advertisers' targeting choices.
None of those studies establishes exactly how Meta's current algorithms are distributing New Zealand political advertising in 2026.
That is precisely the problem.
Researchers and voters can often see who received an advertisement.
Determining why they received it is considerably harder.
Two Questions Voters Should Be Able to Answer
Political advertising transparency therefore requires answering two separate questions.
Who did the political party ask to reach?
And:
Who did the platform actually reach?
At present, those can be very different things.
If a party explicitly tells Meta to advertise only to men aged 35 to 60, that is a political campaign decision voters may reasonably want to know about.
But if a party selects all adults and Meta's optimisation system subsequently delivers most impressions to men because its algorithm predicts greater engagement, that is a different phenomenon.
Both outcomes matter.
But accountability is difficult without knowing which occurred.
New Zealand's Rules Were Built for a Different Era
New Zealand election advertising law already provides some transparency.
The Electoral Commission's guidance on social media campaigning explains the requirements surrounding election advertising and promoter statements, while spending limits and reporting requirements apply during regulated election periods.
But those rules say relatively little about how the eventual digital audience for an advertisement was selected.
A promoter statement can tell voters who authorised an advertisement.
It does not tell them why they personally received it while another voter never saw it.
Nor does it reveal whether an advertiser specifically targeted their demographic or whether an automated system subsequently concentrated delivery towards people like them.
That distinction becomes increasingly important as artificial intelligence assumes greater control over advertising distribution.
Political Advertising Is Becoming Personalised
There is another democratic consequence.
Two neighbours can now experience substantially different election campaigns.
One may repeatedly encounter advertisements about taxation and government spending.
Another may receive messages about housing, climate policy or healthcare.
A younger voter might encounter an entirely different collection of political messages from someone aged over 65.
RNZ's analysis found differences not only according to gender but also age. Advertisements run by Te Pāti Māori and the Greens attracted younger audiences, while ACT appeared to reach proportionately more people aged 65 and over.
This does not necessarily indicate manipulation.
Political campaigns have always tailored messages to different audiences.
Farmers receive agricultural messages. Businesses hear about taxation. Families hear about education.
The difference is that traditional campaign advertising was largely visible to everyone.
A billboard could be photographed.
A television advertisement could be recorded.
A newspaper advertisement could be archived.
Algorithmically delivered political advertising can be almost invisible outside the audience selected to receive it.
Europe Tried a Different Approach
Europe has attempted to address the issue through stronger disclosure requirements.
Since October 2025, European Union political advertising rules have required greater disclosure around political advertisements, including information about sponsors, costs and targeting techniques.
The response from major technology companies demonstrated the difficulty regulators face.
Meta announced it would stop allowing political, electoral and social-issue advertising in the European Union rather than operate political advertising under the new regulatory environment. Google also withdrew political advertising from the EU.
That produces its own consequences.
Large political parties may still have television, radio, outdoor advertising, mailing lists and substantial campaign infrastructure.
Smaller parties and independent campaigners may depend much more heavily on inexpensive digital advertising.
Regulation intended to increase transparency can therefore unintentionally make political communication more expensive and strengthen established political organisations.
Microtargeting May Not Be the Biggest Problem
There is also reason not to exaggerate the persuasive power of political microtargeting.
A randomised experiment involving 8,587 participants found political messages tailored to individual characteristics were not significantly more persuasive than messages shown uniformly.
That suggests the democratic concern may not simply be that sophisticated algorithms can secretly change people's political opinions.
The bigger problem may be visibility.
If voters cannot see the full range of messages political parties are distributing — and cannot determine whether those parties or the platform chose the audience — public scrutiny becomes much harder.
Transparency Without Banning Digital Campaigning
New Zealand does not necessarily need to prohibit targeted political advertising to address the issue.
A simpler starting point would be disclosure.
Political parties and other promoters could publish the audience settings they selected when purchasing digital political advertisements.
Platforms could separately publish aggregated demographic information showing who actually received those advertisements.
The two datasets could then be compared.
If a party deliberately targeted a narrow demographic, voters would know.
If the party selected a broad audience but the algorithm dramatically concentrated delivery among one demographic, voters would know that too.
That would preserve digital political campaigning while making the machinery behind it considerably more visible.
Look Beyond Your Own Feed
Voters do have tools available to examine political advertising beyond what appears on their own screens.
Meta's Ad Library allows users to search advertisements being run by political parties and other advertisers, including advertisements they may never personally have received.
Political advertisers on Meta are also subject to authorisation requirements, with political and electoral advertising carrying information identifying who paid for it. Meta has outlined its election-related advertising policies in its global elections guidance.
These tools provide some visibility.
But transparency should not depend entirely upon individual voters conducting their own investigation.
Does This Affect New Zealand?
Yes — particularly as the country moves deeper into an election campaign increasingly conducted through individually curated digital feeds.
The central issue is not whether National, ACT, the Greens or any other party should be permitted to advertise to particular groups. Political campaigns have always sought audiences they believe might support them.
The issue is whether voters can see how that targeting occurred.
A social media feed is not a neutral summary of an election campaign. It is a personalised selection influenced partly by the advertiser and partly by systems predicting what will attract an individual's attention.
People of different ages, genders and browsing histories may therefore experience remarkably different versions of the same election.
That raises a wider democratic question.
Election advertising rules have traditionally focused on who created the message, who authorised it and who paid for it.
Algorithmic advertising adds another decision-maker between the political campaign and the voter.
As those systems assume greater responsibility for deciding who receives political messages, New Zealand's transparency rules may eventually need to recognise a basic reality:
Knowing who paid for a political advertisement is no longer enough. We also need to know who decided who would see it.
Source: The Conversation — Algorithms are choosing who sees what political ads. NZ urgently needs transparency, published 26 August 2026. Supporting New Zealand political advertising figures from RNZ's analysis of Facebook and Instagram political advertising.
Independent reporting. Original context. Credited sources.