Can Ads Restore Trust in Elections? A Giant Real-World Test Says Barely

Imagine you wanted to convince skeptical voters that elections are safe.

You hire professional political ad-makers.

You feature local election officials and law enforcement. You send dozens of videos, ads, texts and postcards over a month.

Surely that would work?

A new study in Science Advances tested exactly this, on nearly 150,000 conservative voters in the Houston area during the 2024 US presidential election.

The result was humbling. The campaign barely moved anything.

Why election trust matters

Confidence in US elections has fallen, and it now splits sharply along party lines.

That's not just a feeling. Researchers have linked low trust to weaker support for democracy, lower turnout, and more support for political violence.

The paper cites one survey where 81% of people who trusted the election system voted. Among those who distrusted it, 66% did.

After the 2020 election, officials in Georgia, Arizona, Texas, Colorado, California and elsewhere ran public campaigns explaining how votes are protected.

But nobody knew whether those campaigns actually worked.

The experiment: a real campaign, randomly assigned

Who was targeted

Researchers from UC San Diego, Stanford and Rice University worked with a real campaign in the Greater Houston area.

They focused on registered conservative voters who a statistical model predicted would be moderately distrustful of how elections are run. That gave them 147,488 people.

Coin-flip households

Households were randomly assigned to get the campaign or not.

Random assignment is what makes this study powerful. Any difference between the groups can be put down to the campaign itself.

A serious campaign

This wasn't a single leaflet. From September 30 to November 1, 2024, each treated household received on average:

• About 20 online videos
• About 18 videos on cable TV
• About 24 online display ads
• 3 text messages
• 3 postcards

The messages were made by professionals who work with conservative voters in Texas. They featured election officials and law enforcement figures.

They explained election laws, ballot security, audits, how ballots are tracked, and safeguards for voting by mail.

How it was measured

Just before election day, from October 28 to 31, the researchers ran a survey of people in both groups. Thousands responded.

They also checked official Texas voting records to see who actually turned out.

Big enough to spot small effects

A null result only means something if the study could have seen an effect. This one was designed to.

The survey could detect a shift of roughly 2 percentage points in the trust questions. With nearly 150,000 voters, the turnout check could spot a change of less than half a percentage point.

So when the study says "no change", it means any real effect was probably smaller than that.

The results: almost nothing moved

The team had registered five predictions in advance. Four came back empty.

The one success was trust in voting by mail, a safeguard the campaign explained directly.

But even that effect was small. It was roughly half the size of the smallest effects seen in earlier survey experiments, and a third of others.

Why the gap with earlier studies?

This is the most interesting part. Earlier survey experiments found that similar messages boosted trust quite a lot.

The authors suggest several reasons the real world was different.

In a survey, people see a message and answer questions straight away. In real life, they may scroll past an ad, forget it, or never see it at all.

And in 2024, voters were also hearing lots of competing claims about fraud from politicians and media. One campaign's message had to fight through all that noise.

The study measured the effect of offering the campaign to a household. Among people who actually watched and paid attention, the effect could have been bigger.

Surveys flatter messages

There's another reason survey experiments look better. People taking part know they're in a study, and may give the answers they think researchers want.

Researchers call this a "demand effect". Survey answers also cost nothing, so they can be "cheap talk" rather than a real change of mind.

A field experiment avoids both problems. People don't know they're being studied, and turnout is checked in official records.

What earlier research found

Survey experiments looked promising

A 2025 study in Science Advances tested messages to counter false fraud claims in the US and Brazil.

"Prebunking", or warning people in advance about false claims and explaining election security, increased confidence and reduced fraud beliefs. Corrections from credible sources worked too, at least right away.

Those were online survey experiments, though. People answered just after seeing the messages.

Official messages seemed to break through

A 2025 survey experiment in the British Journal of Political Science, by some of the same researchers, tested official messages about election trust.

According to the new paper, people who saw trust-building messages were more likely to say they planned to vote in 2024. That hinted trust campaigns might even raise turnout.

The new field experiment checked that in official records, and found no turnout effect.

Campaigns rarely persuade

A landmark 2018 study in the American Political Science Review pooled 49 field experiments on campaign contact in US general elections.

The average persuasive effect of mail, calls and door-knocking on how people voted was essentially zero.

That matches what this new study found. Changing minds in the heat of a big election is extremely hard.

An earlier trust campaign didn't raise turnout

The only earlier field experiment on voter trust sent a single election integrity mailer during the 2020 presidential election.

According to the new paper, it had no effect on turnout either.

How it fits together

Put together, the pattern is consistent. Messages that look powerful in a survey tend to shrink, or vanish, in a noisy real election.

That doesn't mean facts don't matter. It means the setting matters enormously.

How much should you trust this?

Strong, for what it tested. It's a large, preregistered, randomised experiment in the real world.

What makes it convincing

• Households were randomly assigned, so the comparison is fair.
• The predictions and analysis plan were registered before the results came in.
• Turnout came from official voting records, not self-reports.
• The data and code are publicly available, and the authors declare no competing interests.

What makes me cautious

• It only covered conservative voters in one US metro area.
• It happened during an intensely polarised presidential race. A quieter election might be different.
• The survey only reached a few thousand people, so small effects could be missed.
• We don't know how many people actually saw or paid attention to the messages.
• The campaign featured law enforcement and election officials. Other messengers might work differently.

What this means for you

If you work in communications, public health or politics, this is an important lesson. What works in a survey may not work in the real world.

For everyone else, a few takeaways:

• Be wary of "one message fixes it" claims. Changing deep beliefs usually takes time and trust.
• Specific facts may land better than general reassurance. The one thing that moved was trust in a specific process, mail voting.
• Check how votes are protected yourself. Your state or county election office usually explains audits, ballot tracking and mail voting safeguards.
• Notice the noise. If you hear constant fraud claims, look for evidence rather than volume.

In the US, USA.gov helps you find your state or local election office.

For election officials

I don't read this study as saying information campaigns are pointless. The lesson I take is about expectations.

Test campaigns in the real world before relying on them, and expect modest effects. Specific, practical facts may do more than broad reassurance.

And build trust between elections, not only in the final weeks, when voters are flooded with competing messages.

This 60 Minutes report looks at "prebunking", the approach earlier survey studies found effective:

What we still don't know

1. Would it work outside an election season? Messages might land better when politics is quieter.
2. Would different messengers work better? Friends, local leaders or party figures could be more persuasive.
3. Would a longer campaign help? Trust might build slowly over years, not weeks.
4. Does it work with other voters? This study only tested conservatives in the Houston area.
5. What happens among people who actually watched? Engagement data could reveal a hidden effect.
6. Can trust be rebuilt at all once it's lost? That's the big question for democracies everywhere.
7. Does the small mail-voting effect change behaviour? More trust in mail ballots might mean more people vote by mail next time.

My take: a valuable dose of reality

What I like about this study is its honesty. The authors tested a serious real-world campaign properly, and published the mostly disappointing results.

That's exactly the kind of research we need more of.

I'm cautious about generalising too far, because it's one region and one very heated election.

But the lesson is powerful.

If you want to change what people believe, a great survey result is only the start. The real test is out in the world, where your message competes with everything else.