Why AI ads and foreign meddling did not seem to sway the US midterm elections

World

The real harm may have been done elsewhere, in the deepening of divisions and erosion of trust, experts told CNA.

New: You can now listen to articles.

This audio is generated by an AI tool.

12 Nov 2026 04:23PM

SINGAPORE: A hyperrealistic video showed then-United States election candidate Abdul El-Sayed, from the Democratic party, laughing at cards that read “Celebrate 9/11”. In another video, James Talarico, also a Democratic hopeful, was depicted defacing a Bible with a crayon.

These were advertisements generated with AI. In the last weekend before the Nov 3 US midterm elections, the American research outfit Wesleyan Media Project - which tracks political advertising - identified close to 300 such ads that were mostly sponsored by the Republican party, and with over US$187 million spent promoting them.

Besides these ads, US intelligence agencies had flagged a separate but related threat: Covert online campaigns linked to foreign governments like Russia, and increasingly run with the help of AI.

Yet after months of such warnings, the elections passed without major disruption or surprises. The Democrats won a landslide victory, as many polls predicted. Among the victors were Mr El-Sayed, who beat Republican Mike Rogers in Michigan; while in Texas Mr Talarico defeated Republican Ken Paxton, whose campaign had released the AI-generated Bible video.

Did the AI ads and foreign meddling fail to sway voters? Or did they fail to make enough of a dent in the results, which the US media labelled as a “blue tsunami” in favour of the Democrats?

Experts pointed to the foreign operations struggling to find an audience, and Republican ads coming up against voter discontent. They cautioned, however, that some interference may not have come to light yet, and that online campaigns could deepen divisions and distrust long after an election.

AI TOO “ABSURD”

The midterms were the first US election in which AI was “built into every major social media platform”, as well as across chatbots, media generation tools and content moderation systems, according to the Institute for Strategic Dialogue think-tank.

Spending on AI-generated ads also soared from about US$50,000 in 2024 to over US$187 million in this election, said Professor Erika Fowler, co-director of the Wesleyan Media Project.

But researchers and campaign strategists found that many of the AI-generated ads, mostly sponsored by Republicans, had used AI to alter real photographs and videos, or to show scenes too absurd to believe. The combined effect could have been voters seeing the ads as manipulative, the New York Times reported.

Meanwhile US intelligence agencies found that Russia, Iran and others had tried to sway voters this year, though these efforts appeared minimal overall.

Foreign meddling has been a problem for US elections since 2016, but such activity did not surge around election day as expected, and there was no sign that voting systems had been hacked or ballot counts altered, according to reporting by the New York Times and Politico.

Experts cautioned that the activity spotted so far might not be the full picture.

Associate Professor Chong Ja Ian at the National University of Singapore noted that not every attempt to sway the midterms may have come to light. Some might have been quietly stopped, while others might have gone unnoticed only to be caught much later. Individuals who have been targeted “have an incentive not to publicise such information”, he added.

NOT ONE-OFF

Online, foreign campaigns seemed to struggle to reach real voters. Social network analysis firm Graphika found that three China-linked groups of accounts that were active on X ahead of the midterms drew most attention from within their own circles rather than real users.

Achieving reach is the first hurdle, said Associate Professor Saifuddin Ahmed of Nanyang Technological University of Singapore’s Wee Kim Wee School of Communication and Information. “Most small-scale operations are likely to fail to cross this barrier,” he said, though they can have greater impact if social media influencers, trusted public figures or politicians share their content.

Where misleading content did achieve some spread, those behind it appeared to have goals beyond the US and beyond the midterms.

Threat intelligence firm Alethea reported on a long-running, Russian state-backed operation known as Doppelgänger, which in early October targeted the US, Israeli and upcoming French elections all at once. It used more than 50 fake accounts and more than 150,000 posts on X, boosted by over 1,000 automated accounts known as bots.

Another covert campaign, linked to Iran, published about 100 articles under fake writers’ names since July 2025, according to an early October report by OpenAI. Most were about US-Iran relations rather than the midterms, and they appeared more often after war broke out between the two countries in February.

For countries waging such campaigns against the US, any apparent win or loss in a particular contest should be seen as part of a “back and forth”, said Assoc Prof Chong. “Information operations often look to changing and managing perceptions over time, rather than one-off events.”

THE WIDER THREAT

The midterms suggest that producing more content – whether it comes from abroad or within the US – does not automatically mean reaching or winning over voters.

Yet experts noted that the impact of online campaigns is notoriously hard to measure. A relatively quiet election also leaves open questions about safeguards in place and whether they will last.

After the US government scaled back its election protection teams in 2025, firms like Anthropic, OpenAI and Meta set up dedicated election units to monitor harmful content. But Meta and other social media companies have in recent years also downsized the teams tasked with keeping their platforms safe.

Assoc Prof Saifuddin said it was “still too early to say whether the (election) safeguards worked”. Without hard data, it was difficult to draw firm conclusions, he said, adding that researchers’ reliance on social media companies to share that data was “problematic”.

Disinformation spreading through private messaging apps was another concern, as this was far harder to track.

Timing matters in catching such content, Assoc Prof Saifuddin said. Falsehoods that spread just before voting leave little time for correction, especially when they are about how to vote.

Widespread AI-generated disinformation also affects society in ways that go beyond election outcomes, he warned, citing his research on deepfakes in several countries, including the US. Fakes allow real evidence to be dismissed as made-up, fuelling wider doubt about what people see online, he said.

“The bigger threat … is not necessarily changing how people vote but slowly eroding their trust in institutions,” he said. “Election results cannot capture this gradual loss of political and institutional trust.”

Source: CNA/jo