News

Republicans bought over 90% of AI ads. In Michigan, Democrats won anyway

AI-generated ads, like the one depicting Gov. Gretchen Whitmer about to be run down by a tractor, drew criticism, while Michigan voters elected candidates who promised Big Tech oversight. Experts warn the next fakes will be harder to spot.


Misleading deepfake political ads were plentiful during the 2026 midterms, but Michigan voters took a stance against the technology, voting in Democratic candidates promising to rein in Big Tech and data center expansions. 

Senator-elect Abdul El-Sayed campaigned strongly against data centers, and US Rep.-elect Will Lawrence (D-Lansing) won in the state’s 7th district after speaking out against Flock cameras and the effect of data centers on energy bills, and underlining the need for laws governing safe AI use. 

These wins came during what Erika Franklin Fowler, co-director of the nonpartisan research initiative Wesleyan Media Project, called the country’s first AI election. 

The project, which has analyzed political advertising since 2010, identified 289 AI-generated political ads that ran through October of this year’s election cycle, costing $187 million in spending, and determined that more than 90% of those ads were purchased by Republican candidates and groups. 

In short, voters didn’t fall for fake AI ads. Not the one for gubernatorial candidate Aric Nesbitt, where Gov. Gretchen Whitmer was shown getting nearly hit by a tractor. Nor the ad depicting El-Sayed in a Trojan horse, threatening to usher in communism under the guise of democratic ideals. Yet experts say the technology will only improve, potentially influencing future voters and spreading misinformation. 

“Whenever there was an AI ad out there that was particularly egregious, it became a headline, and it was cartoonish in a way,” Deputy Secretary of State Aghogho Edevbie said.  “And because this technology is so new, it attracts so much attention that you can’t get away with deceiving the public. But that may change, and I think there’s a real danger.” 

The harm of misinformation from AI ads

Nesbitt’s ads showed a tractor heading toward Whitmer, Michigan Secretary of State Jocelyn Benson, and Minnesota Governor Tim Walz with the words “Justice is coming,” and were among the ads that made national headlines this year. Nesbitt later dropped out of the race after the ads drew criticism for portraying violence, even with the required AI disclosures. 

On Facebook, some viewers noted that while the ads were obviously fake, they fully believed the messaging, saying “Maybe (it’s) AI-generated but it’s spot on.” Others called out the hypocrisy, writing, “So he makes fun of data centers then uses AI to make an ad.” 

But even if voters know an ad is using fake visuals and satire to poke fun at candidates, their messaging can still stick. That’s something laws surrounding AI disclosures in political ads can’t address. 

Gang Wang, an associate professor of computer science at the University of Illinois Urbana-Champaign, said the damage from those ads could already be done. In 2023, Wang helped author a study on fake photos, which determined, from a small sample size, that the negative impacts of media that viewers recognize as AI-generated linger even if they know the messaging was inaccurate.

“Even if a user says, ‘Yeah, I realize that’s a deepfake,’ they still feel somehow negatively about the candidate being pictured in this deepfake,” Wang said.

And the technology improves daily. Garrett Brubaker is the owner and creative director at Studio Brubaker, an Indiana-based advertising company that put out about 300 political ads during this election cycle, all for Democratic candidates and causes. Only two used AI, and those ads were not deepfakes, which are typically defined as using AI to make someone appear to do or say something they never did or said. 

“I don’t think people would immediately be able to clock an AI ad if they saw it,” he said. “That’s the scary part because, like I said, it’s only going to get worse.”

Further, not all viewers lock in to watch a political ad, so they might not notice a written disclosure on their screen or watch the visuals and notice they’re too smooth or perfect to be real. 

“They’re often playing in the background, and people are not paying 100% attention,” Wang adds. 

How Michigan handled AI misinformation during the midterms 

Edevbie said the state has been proactive in educating voters on the use of AI in political ads. 

“During the last Legislature, we passed two bills dealing with AI and elections,” Edevbie said. “If you’re going to use AI, you have to disclose it and make it clear to the voter, and that was largely respected during this election cycle.” 

In addition, Edevbie said the state did not see widespread problems with AI disrupting the voting process itself, adding that the office has protocols in place to combat bad information about how to cast ballots and where, should a problem arise. 

Edevbie said he believes voters are becoming more and more savvy about how to spot and interpret AI-generated political ads, but more protections are needed from the federal level. 


“I hope the federal government steps up,” he said. “A good first step is to pass the Protect Elections from Deceptive AI Act, but it’s going to be on local election officials so voters know how to cast their ballots and feel confident doing so.”  

The Michigan laws in place did face at least one challenge during the midterms. A voter filed a complaint against US Senate candidate Mike Roger’s campaign, which used two AI-generated ads without including disclosures. The Secretary of State Office dismissed the complaint on Nov. 4, saying that there was no breach of the law, which does not require disclosure statements for ads that are considered satire or parody.  

Ilana Beller, director of state AI governance and tech policy for TKTKTKT Public Citizen, said that while these laws can’t address satirical ads, they’re still helpful in alerting voters to the type of content.

“These laws are intended to prevent the nightmare-scenario deepfake—a deepfake that deceives the voters into believing a candidate actually did or said something they did not otherwise do and swings an election,” she said.“Because of protections around political speech, they cannot address satirical content. So while there was a lot of satirical content this election cycle, the laws were successful in having their intended effect of preventing the nightmare-scenario content.” 

Michigan leaders face pressure to act

Now, with new Democratic leaders on deck, voters are watching to make sure the promises they made to reign in AI and Big Tech come to fruition. El-Sayed has proposed making AI technology publicly owned while Lawrence has called for a federal moratorium on data centers. 

In the meantime, experts say more regulation is needed from the federal level to protect voters and the democratic process. Katie Paul, director of the nonpartisan watchdog initiative Tech Transparency Project, said the key is to rein in Big Tech companies such as Meta, which profit from running these deepfake ads and have little incentive to identify and remove them quickly, as well as AI companies that can produce the ads quickly and cheaply.

One potential approach is to put a cost on producing AI-generated political ads. “Having an appropriate cost associated with it makes it so [AI ads are] not the more cost-efficient option or nearly free,” she said. “That also lowers the bar of who can run political campaign ads.”

And while the next election cycle is two years away, the imminent threat is generative AI’s ability to revive old rumors with fake video footage, Paul said. 

“The bigger threat is creating visuals to support misinformation, to support stereotypes that include fearmongering of different minority groups and things like that,” she said. “We saw some really vile racism targeted at Muslim candidates like El-Sayed during this election, for instance, and being able to create visuals that support that narrative, when he hasn’t actually engaged in any of the things that were lobbed at him, is the bigger risk.”

Of course, enforcing new oversight of these technologies won’t be easy. 

“We’re talking about technologies that are also constantly evolving, which makes it even more difficult to put together hardline regulation because what you’re regulating may evolve and change within the next six months,” Paul added. 


Categories: ,

Authors

  • Kristine Gill is a freelance writer and editor based in Ohio with a background in newspaper journalism and public-sector communications.