The 2026 midterm cycle has quietly crossed a threshold that researchers and regulators have long dreaded: AI-generated political advertising is now a routine feature of American campaign media, appearing on television screens and social media feeds with a regularity that would have seemed extraordinary just two years ago. And according to new research from the Wesleyan Media Project, one party is driving the trend with striking dominance.
Researchers Travis Ridout, Erika Franklin Fowler and Michael Franz have tracked roughly $80 million in ad spending across nearly 170 unique AI-generated political ads since the current cycle began. Their findings, drawn from media reports, trained student coders and data from ad-tracking firm AdImpact, reveal a sharp partisan divide, a disclosure regime that barely functions, and a regulatory landscape so fragmented it offers voters almost no meaningful protection.
Republican candidates and pro-Republican groups were responsible for 80% of the AI ads tracked and 83% of the associated spending. That asymmetry is not a rounding error. It reflects something real about how each party currently relates to emerging technology, to regulation, and to the question of what is owed to voters in terms of honesty.
What These Ads Actually Look Like
The range of AI use spans a wide spectrum. At one end sit hyperrealistic deepfakes — synthetic videos depicting real people doing things they never did. The researchers identified multiple ads featuring fabricated likenesses of Donald Trump, Nancy Pelosi, Barack Obama, Kamala Harris and Alexandria Ocasio-Cortez. Ocasio-Cortez appears in at least five such ads, making her a particular favorite target of Republican advertisers.
The scenarios are sometimes absurd, sometimes alarming. One ad depicts a Republican Senate candidate from Louisiana driving a school bus full of undocumented immigrants. Another shows a Republican gubernatorial candidate from South Carolina walking arm-in-arm with drag queens — a transparently homophobic smear dressed up in synthetic video. A third places Liz Cheney, Mitt Romney and Mike Pence on the White House lawn carrying pitchforks, a piece of intra-Republican political theater that at least targets public figures rather than manufacturing false threats.
A fake Dr. Anthony Fauci runs through a state fair brandishing a giant syringe. AI-generated crowds populate candidate ads to manufacture the appearance of grassroots enthusiasm. In other cases, the technology is used more subtly — enhancing visuals, adjusting lighting, cleaning up backgrounds — in ways that leave no visible trace and give viewers no signal that what they’re watching has been algorithmically altered.
That last category may be the most consequential. A deepfake of a politician doing something outlandish can, in principle, be identified and debunked. An ad that uses AI to subtly manipulate imagery without any disclosure is harder to challenge precisely because it’s harder to detect.
A Disclosure System That Isn’t Working
Across the 35 states where the researchers tracked AI ad spending, only 31% of ads disclosed the use of AI tools, representing just 22% of total spending. The wording of those disclosures, where they exist, varies wildly. A Georgia ad acknowledged that “this video has been manipulated or generated with artificial intelligence.” An Oklahoma ad called itself “political satire” and noted that “AI-generated images do not depict actual events.” A North Carolina state Senate ad took a breezy approach: “You guessed it! AI was definitely used to generate these silly video clips.”
That last example illustrates the problem neatly. A disclosure regime that allows campaigns to self-describe their manipulations as “silly” and “satire” is not a regime designed to protect voters. It is a regime designed to give campaigns legal cover while minimizing the reputational cost of being caught using deceptive technology.
More troubling still, the researchers found that the presence of a state law requiring AI disclaimers has almost no relationship with whether disclaimers actually appear. In states without AI disclosure laws, 32% of tracked ads included a disclaimer. In states with such laws, that figure actually dropped to 29%. Enforcement, in other words, is effectively nonexistent.
Some of that gap has technical explanations — certain state laws apply only to candidate deepfakes, excluding other AI uses, and some laws were enacted after specific ads had already aired. Minnesota bans deepfakes in political ads outright, yet a synthetic video targeting Democratic Senate candidate Peggy Flanagan aired in May 2026 regardless. Whether it violated state law remains legally uncertain, hinging on whether a “reasonable person” would have believed the fabricated footage was real.
The partisan dimension of the disclosure gap deserves direct attention. Republicans, as a governing philosophy, have consistently opposed campaign finance disclosure requirements — Senate Republicans blocked the DISCLOSE Act in 2022 — and have favored deregulatory approaches to emerging technology. That same philosophy appears to be shaping how Republican campaigns and super PACs approach AI disclosure: as an optional courtesy rather than an obligation to the electorate. Democrats are not blameless actors in the broader campaign finance system, but the empirical record here is unambiguous.
Voters, for their part, are not indifferent. Polling shows that 78% of registered voters support banning AI content that makes deceptive claims about candidates — a level of consensus that spans partisan lines and suggests the political will for stronger federal action exists, if lawmakers choose to act on it. The real challenge ahead involves not just passing laws but designing disclosure requirements that are specific enough to be meaningful, enforceable enough to have teeth, and clear enough that ordinary viewers actually understand what they’re being told.