Algorithms are choosing who sees what political ads. NZ urgently needs transparency

A recent RNZ analysis of political advertising on Facebook and Instagram revealed an intriguing pattern in who was seeing the ads.

Of more than 500 ads National

ran this year, 92 were seen only by men, attracting up to 3.9 million views and costing as much as $21,000. Only one was seen solely by women – for a single day, at a cost of less than $100 – and the party says it didn’t place it.

ACT showed a similar tilt, with 14 ads seen only by men and none solely by women. For the Greens, the pattern was reversed: 13 ads were seen only by women and none only by men.

Paid campaigning on Meta’s platforms is nothing new in New Zealand elections. What has changed is how decisions about who sees those ads are made.

At first glance, RNZ’s findings might suggest parties are deliberately choosing to speak to men and women separately. But that assumes parties themselves determine who ultimately sees each ad.

Increasingly, that isn’t necessarily the case.

Who decides who sees an ad?

For much of the past decade, digital advertising was relatively straightforward: advertisers selected audiences based on factors such as age, gender or interests, and platforms delivered ads accordingly.

That model has shifted. Meta now leans heavily on AI to judge which ads are most relevant to users. Its Andromeda system narrows tens of millions of potential ads and variations to a few thousand likely matches before ranking them. Meta’s Advantage+ tools can also automate decisions about audiences, budgets and ad placement.

This makes the distinction between targeting and optimisation important: advertisers choose who they initially want to reach, while the platform learns who responds and increasingly delivers ads to them.

The result is that a political party can begin with a deliberately broad audience and still end up reaching disproportionately more men, younger voters – or people already inclined to agree with its message.

Researchers have been documenting this problem for several years. In 2019, a study found Facebook’s ad delivery could skew audiences by gender and race even when advertisers used inclusive targeting.

Other research found voters statistically less likely to support a party could cost two or three times as much to reach, while campaigns had little visibility over how those decisions were made.

Similar pricing effects have appeared in elections. An analysis of ten Indian elections found Facebook charged the governing Bharatiya Janata Party (BJP) lower advertising rates than rival parties in nine of them, allowing it to reach more voters for the same spend.

More recently, a study of the 2024 European Parliament elections analysed more than 110,000 ads across 25 countries. Men were significantly more likely to receive ads from populist and far-right parties, even after researchers accounted for advertisers’ targeting choices.

These studies cannot tell us exactly what Meta’s current systems are doing. Its technology continues to evolve, while outside researchers have limited visibility into how ads are selected and delivered. We can see who received an ad, but establishing why is much harder.

Europe has attempted to improve transparency. Since October 2025, EU rules have required political ads to disclose their sponsor, cost and use of targeting techniques. Meta and Google responded by stopping political advertising in the EU rather than complying.

That illustrates a challenge for regulators: stronger disclosure can improve transparency, but global platforms can simply withdraw from smaller advertising markets. Smaller parties and campaigners may feel that loss most, given they have fewer cheap alternatives for reaching voters.

Does New Zealand need new rules?

Concerns about political microtargeting are not always backed by strong evidence.

In a randomised experiment involving 8,587 participants, tailoring political messages to individual characteristics did not make them significantly more persuasive than showing everyone the same message.

The bigger issue is transparency around how political advertising is distributed.

New Zealand law requires election advertisements to carry a promoter statement, while spending limits and reporting requirements apply during the regulated period. But these rules say little about how an ad’s audience was selected.

For digital advertising, that leaves two important questions: who did the party ask the platform to reach and who did the platform actually show the ad to? A party may select a broad audience, but the platform can still concentrate delivery among people it predicts are most likely to respond.

Greater disclosure could help close that gap. Parties and promoters could publish the targeting settings used for political ads, while platforms could provide aggregate data showing who actually received them.

This would make it easier to distinguish deliberate targeting by a political party from optimisation by the platform.

Voters can also get a broader view of the campaign than their feeds provide. Meta’s Ad Library allows anyone to search the political ads a party is running, including those they might otherwise never see.

Election advertisements should also carry a promoter statement, while political advertisers on Meta must be authorised and labelled “paid for by”.

Ultimately, a social media feed is a sample rather than a summary of an election campaign. It is influenced partly by whoever paid for the advertising and partly by a system predicting what will hold someone’s attention.

People of different ages, genders and browsing histories may therefore see very different campaigns. Party websites, news coverage and ad libraries can help fill in what is missing.

Katherine Jensen does not work for, consult, own shares in or receive funding from any company or organisation that would benefit from this article, and has disclosed no relevant affiliations beyond their academic appointment.