Oct. 30, 2026, 5:04 a.m. ET
Texas Senate candidate James Talarico depicted in a video as marking up a Bible with a crayon and talking about religion and gender.
New York Gov. Kathy Hochul shown gardening with New York City Mayor Zohran Mamdani in a pharmaceutical-style ad promoting a “political delusion pill.”
Michigan Senate candidate Abdul El-Sayed grinning as he hides in a Trojan horse draped with a banner reading, “Definitely NOT Communism.”
These political attack ads showing scenes that never occurred in real life were all generated with the help of artificial intelligence, or AI, whose use has soared in the 2026 elections compared to two years ago, according to researchers.
The Wesleyan Media Project, which tracks political advertising, has identified more than $187 million in spending on at least 289 unique AI-generated or AI-assisted ads this year as of Oct. 29. That’s up from barely $50,000 in 2024, when the technology was less capable and accessible.
“It’s kind of gone from zero to 100 in just that two year term,” said Michael Franz, a professor of government at Bowdoin College who co-directs the research institution at Wesleyan University.
Even the startling stat from Wesleyan is likely an undercount because it doesn’t capture more targeted political AI ads on social media and because AI can be difficult to detect and isn’t always disclosed, Franz said.
Experts said it reflects a 2026 campaign cycle that marks the widest-yet use of AI, from ads to outreach where bots can reply to voters’ text responses.
At the same time, AI chatbots and search summaries are also reshaping how voters find information, said Isabel Linzer, an elections and democracy analyst at the the Center for Democracy and Technology.
While the impact of AI’s growing presence is expected to come into better focus after the election, concern about AI use in elections has long centered around convincing “deepfakes” that taken as real could upend a race.
But so far, Franz said most of the ads his group identified either used AI to enhance an ad, such as by replacing stock footage, or were obvious deepfakes framed as satire, a format that has been dubbed “slopaganda.” The attack ads spoofing Democrats Talarico, Hochul and El-Sayed from the campaigns of their Republican opponents − Texas state Attorney General Ken Paxton, Nassau County Executive Bruce Blakeman and former Rep. Mike Rogers, respectively − are all examples of so-called “slopaganda” genre.
Some of the ads disclose the use of AI, such as an opposition ad targeting Jessica Bichler, a North Carolina state Senate Democratic candidate, who owns a yoga studio. She’s pictured stretching to video clips that the spot says were generated with AI, while a voiceover says, “Some liberals lean to the left. Some stretch all the way to woke. But yoga instructor Jessica Bichler, she’s totally bonkers.”
Of AI-generated ads that Wesleyan researchers counted through mid-October, only 27% percent of them contain a disclaimer.
AI is also being used in ads as voice cloning, which can turn a candidate’s past statements or posts into audio resembling their voice.
In August, Rep. Nancy Pelosi, D-California, and several other female elected officials decried an ad created by Democratic state Sen. Scott Wiener, who is running to replacing the retiring former House speaker. Wiener’s campaign created an AI chatbot to lampoon his opponent San Francisco Supervisor Connie Chan. Billboards directed voters to ConnieChan.ai, which said, “I’m Connie, the world’s first AI trained to say ‘no’ to anything you ask. Bring me an idea and I’ll explain why it can’t, shouldn’t, and won’t happen.”
Chan and some supporters alleged it responded to certain prompts in a racist or sexist manner.
“Women have spent generations fighting to have our own voices heard in the halls of power,” a letter from Pelosi and colleagues stated. “We should not have to fight against artificial intelligence putting words in our mouths.”
Republicans account for most AI ads
Talarico, a repeat target of AI ads, told CNN in September that “Texans are sick of these AI ads trying to deceive voters.”
Political Slopaganda, a report from New York University’s Center on Technology Policy, cited an ad from a Republican-aligned PAC showing Talarico singing a version of the song “My Favorite Things” where the lyrics include, “Boys in white dresses with blue satin sashes.”
“The point isn’t likely to make people believe that Talarico sang the song, but instead might be to make viewers feel a certain way about Democratic priorities and about how Democrats articulate masculinity,” the report said.
The Wesleyan report found that AI-generated ads were dominated by Republicans, with Republican candidates, a Republican Party committee or a Republican-leaning group responsible for just over 90% of spending on ads generated or enhanced by AI through late October.
Democrats are slightly more likely to be wary of AI, according to some surveys, such as a June poll from the Pew Center that found 56% of Democrats are “more concerned than excited” about the use of AI in daily life, versus 49% of Republicans.
President Donald Trump regularly uses attention-getting AI images on social media, such as depicting himself as what critics said resembled a healing Jesus. He later said he thought it depicted him as a physician. He is also an enthusiastic proponent of the economic benefits of developing AI tools and building the data centers needed to power them.
With no federal law governing deepfake content, the midterm election is testing a patchwork of state laws. In 2024, 31 states adopted resolutions or enacted legislation, most requiring disclosure, according to the National Conference of State Legislatures.
When facing complaints about a lack of compliance with state laws mandating disclosure of the use of AI, some candidates, including Blakeman, have argued the spots are exempt because they are satire.
AI search summaries another new frontier
While AI plays a greater role in campaigns, it’s also growing as a way voters find information – bringing its own set of concerns about accuracy and efforts to influence answers.
“Voters are now operating in a situation where the information that they are accessing is really being mediated by AI,” Linzer said.
While just 4% of likely voters said in September they had consulted AI for election information in the previous month, 38% reported getting such info from AI-search summaries, half without realizing it, according to an American Association of Political Consultants Foundation survey.
Companies including OpenAI and Anthropic are working with the nonprofit Democracy Works and the Associated Press to ensure accurate election information, including where and when to vote.
A States United Democracy Center analysis found that answers grew more accurate between fall 2025 and winter 2026, with the rate of verifiable factual errors falling to 0% across both platforms. But it also found incomplete answers, inconsistently directed voters to official state websites and over-reliance on Wikipedia, which is openly editable.
Despite its powerful capabilities, AI is vulnerable to efforts to game its responses, said Deen Freelon, a University of Pennsylvania professor who studies digital politics.
NewsGuard, a group that tracks misinformation, identified 22 partisan sites that have launched this year masquerading as independent local news, raising the total number to 1,181. Those and other sites may often be aimed at shaping AI answers.
The group reported in October that AI picked up on a deepfake video spread by a Russian influence campaign that falsely depicted a student accusing a candidate of sexual assault.
Experts say AI carries a capacity for good, including debunking conspiracy theories or boiling down information that a voter might not otherwise seek out, but it can also be used for nefarious purposes.
For example, a Brennan Center for Justice analysis in October found many major AI models produced fake election content when asked, such as to produce a memo on U.S. government letterhead claiming to have identified millions of ineligible voters.
How much of that swings voters well acquainted with online misinformation isn’t clear. Linzer said it cannot be assumed AI ads and “slopganda” are having an outsized impact on voter candidate choices.
“It’s really hard to persuade voters of anything,” she said.
And there’s plenty of skepticism about AI. The AAPC Foundation survey found that only 1% of voters named AI tools as their most credible source of election information, with 51% of voters younger than 35 saying AI hurts people’s understanding of elections.
The longer term worry, she and others said, is the increased prevalence of AI fakes could lead the public to not even trust accurate information.
A 2026 review of 25 studies on deepfakes found that they can create confusion and reduce trust in online news and government.
“Voters might start to be like, ‘Oh, boy. Now I don’t really know when I see anything if it actually happened or if it’s some spin on something else,’” Franz said. “And that would be a big problem.”