Most people believe propaganda works on someone else. Survey researchers have a name for it.
The third-person effect, first described by sociologist W. Phillips Davison in 1983, is the consistent finding that people rate persuasive messaging as more influential on others than on themselves. The assumption underneath is that intelligence is the defence, and that anyone who reads widely and questions what they see will recognise manipulation when it appears.
Whether that assumption holds is one of the more contested questions in misinformation research, and the answer isn’t a straightforward no.
It also isn’t an academic question. The conditions that determine how much the answer matters have shifted sharply, and the shift has been in the direction of making things harder to tell apart.
Why the question got harder
Until recently, producing convincing fake material at scale took money, skill and time. Generative AI has removed most of that cost.
Synthetic voice, video and text can be produced quickly and cheaply enough that volume is no longer a meaningful constraint, and the visual cues people were taught to look for, malformed hands, garbled text, unnatural movement, have largely stopped being reliable.
The second-order effect may matter more than the fakes themselves. Law professors Bobby Chesney and Danielle Citron coined the term liar’s dividend in 2019 for what happens once audiences know convincing forgeries are possible.
Genuine recordings and documents can be dismissed as fabricated, and the person caught on tape gains a defence that didn’t previously exist. The problem is no longer only that false material circulates. It’s that authentic material loses some of its power to settle anything.
Distribution has changed too. Meta announced on 7 January 2025 that it would end its third-party fact-checking programme in the United States in favour of user-written Community Notes, following a similar move at X, and began testing the system that March.
Supporters argued centralised fact-checking had become a credibility problem in itself. Critics argued crowd-sourced correction is slower and thinner precisely where coordinated campaigns are most active, and the numbers give that argument some weight.
The European Fact-Checking Standards Network put roughly 900 visible Community Notes during the US test period against about 35 million fact-checking labels applied across the European Union over a comparable stretch.
In March 2026, Meta’s own Oversight Board advised against rolling Community Notes out globally as a straight replacement, recommending the company hold off in highly polarised countries, in places in crisis or prolonged conflict, and in countries with a history of organised disinformation networks. European Union regulation has kept professional fact-checking contracts in place there for 2026, though at reduced funding.
New Zealand’s own checking capacity has thinned considerably over the same period. Newshub closed in 2024 with 294 jobs lost, TVNZ cut about 68 roles and axed Sunday, and Whakaata Māori cut 27 roles at the end of that year.
RNZ called for voluntary redundancies after its funding was cut in 2025, NZME has since closed 14 community newspaper titles and reduced the NZ Herald newsroom, and Stuff has proposed further editorial cuts. Ministry for Culture and Heritage figures show media and broadcasting employment fell 5.3% in the year to March 2025, the sharpest drop of any cultural sector.
Whether that matters for propaganda specifically is arguable, but it does mean fewer organisations are positioned to check a claim before it circulates widely.
Set against all of that is a serious body of evidence that the alarm is overstated. 2024 was the largest election year on record, with voters going to the polls in more than 70 countries, and it was widely forecast to be the year AI-generated disinformation broke an election.
Several post-election analyses concluded that it largely didn’t. Researchers found AI-generated material was present but its measurable effect on outcomes was limited, and that most of it circulated among people who already held the position it supported.
Demonstrably false content makes up a small fraction of most people’s media consumption, the heaviest consumers are a narrow and already committed group, and measured effects on voting and behaviour are modest.
The counter-argument is that impact isn’t only about persuasion. Propaganda can succeed by producing confusion, sharpening division or eroding confidence in any shared source of evidence.
If an audience concludes every outlet is biased and every expert compromised, they may stop weighing evidence at all, which advantages whichever claim is most memorable rather than best supported. On that reading, the 2024 elections aren’t reassuring. They measured the wrong thing.
Does being smart help?
The case that intelligence offers no protection rests on how persuasion operates. Propaganda rarely tries to win an argument on evidence.
It works by producing fear, anger, pride or threat quickly enough that the audience reacts before assessing the claim, and emotionally charged content has been shown to spread faster and further on social platforms than measured explanation.
The mechanism most often cited is motivated reasoning. On this account, greater cognitive ability doesn’t make people more accurate. It makes them better equipped to defend conclusions they’ve already reached for other reasons.
Yale’s Dan Kahan found that on politically identity-laden topics such as climate risk, the most numerate and scientifically literate participants were the most polarised, not the least, because they used those skills to shore up the position their group already held.
Identity does much of the work. When a claim becomes attached to belonging, correcting it stops being a factual exchange, which is one reason corrections often produce defensiveness rather than a rethink, and why dismissing an audience as gullible tends to strengthen their attachment to whoever misled them.
A substantial body of research pushes back on that. Gordon Pennycook and David Rand have repeatedly found that people who score higher on tests of reflective reasoning are better at
telling accurate headlines from false ones, and that it holds regardless of whether the headline flatters their politics.
Their summary is that the problem is laziness rather than bias. People share false material because they’re moving quickly and not paying attention to accuracy, rather than because their identity demands it. The two explanations point to different remedies, and the disagreement between them isn’t settled.
Where it leaves the question
The disagreement is narrower than it looks. Few researchers argue intelligence is irrelevant, and few argue it’s enough on its own.
Both camps accept that selective presentation works, that a genuine photograph, statistic or quotation stripped of context can mislead without containing a single false statement, and that certainty is persuasive during unstable periods because it resolves the discomfort of not knowing.
Both also broadly accept the practical implications, even where they disagree about why they work. Pausing before sharing material that produces a strong reaction, checking whether a claim traces to an original source or has simply been repeated until it feels established, and asking what has been left out are all supported by the evidence, though effect sizes are often small and tend to fade within weeks.
Media and information literacy programmes sit in the same category. They’re widely endorsed, including by New Zealand agencies, and there’s reasonable evidence that teaching people to recognise manipulation techniques improves their ability to spot it. There’s weaker evidence that the improvement lasts, or that it changes sharing behaviour outside a research setting.
New Zealanders are not unaware of any of this. The Classification Office’s 2021 report The Edge of the Infodemic, based on a survey of more than 2,300 people aged 16 and over, found 82% were somewhat or very concerned about the spread of misinformation. The Office framed that as evidence of public awareness rather than public credulity.
Awareness and immunity are different things, though. If the claim is that intelligence provides protection from being fooled, the evidence doesn’t support it, and the third-person effect suggests the people most confident of their own immunity may be the least accurate judges of it. If the claim is that intelligence makes no difference, that isn’t supported either.
The more defensible reading is that reasoning ability helps when it’s actually applied, and that the speed and emotional intensity propaganda relies on are precisely the conditions under which it’s least likely to be applied.
What has changed isn’t the psychology. It’s the volume of material, the cost of producing it and the number of newsrooms checking it, and none of those are currently moving in the direction of making discernment easier.
SPONSORED
