We've tracked $80M of spending on almost 170 AI-generated political ads this year - it's weird

The 2026 U.S. midterm campaign is the first election cycle in which AI-generated political ads have regularly appeared on television and in social media feeds. This judgment comes from the Wesleyan Media Project, which has long studied political advertising and has continuously tracked political ad placements since 2010.

Using media reports, student coders and AdImpact data, which tracks political ad spending, the team tracked roughly $80 million spent on nearly 170 ads that used AI. The research also found that AI use ranged from hyper-realistic deepfakes to subtle image enhancements, not only the most extreme category.

Among the tracked ads were hyper-realistic deepfakes of well-known political figures including Trump, Pelosi, Obama, Harris and Ocasio-Cortez, with Ocasio-Cortez appearing in at least 5 ads.

The deepfake scenes included: a Louisiana Republican Senate candidate driving a school bus packed with undocumented immigrants; a South Carolina Republican gubernatorial candidate walking arm in arm with a drag queen; and Cheney, Romney and Pence appearing on the White House lawn holding pitchforks.

Non-political figures were not spared either: the ads included a fake Fauci running through a state fair with a giant syringe. The research also noted that several ads used AI to generate crowds or voter images.

Not all AI use was intended to deceive. One ad from Oklahoma Republican gubernatorial candidate Chip Keating carried an AI disclaimer but did not explain specifically how AI was used. The research noted that ads using AI to enhance images do not necessarily look different from ads from before the AI era, making it harder for viewers to tell whether the images presented false content.

The research showed that Republican candidates and pro-Republican groups accounted for 80% of the tracked ads and 83% of the related spending, clearly higher than the Democratic side in the use of AI political ads.

The researchers could only speculate about why Republicans dominated 2026 AI political ads. One explanation they offered is that Democrats tend to favor a regulatory approach of setting limits on campaign funds and requiring disclosure, while Republicans tend to favor a free-market approach, and these general philosophies may be reflected in the use of generative AI. This is the researchers' speculation, not a confirmed conclusion.

Voters' attitudes toward regulation are relatively clear. Polling shows that 78% of registered voters support banning AI content that makes deceptive claims about candidates. This figure comes from polling and is a source claim.

Across 35 states, only 31% of the tracked ads disclosed the use of AI tools, and those ads accounted for 22% of the related spending. Because AI regulation depends on a patchwork of state legislation, many ads are not required to disclose AI use, and the absence of a disclaimer itself makes tracking AI use difficult.

Disclaimer wording varied widely. One Georgia ad said "this video was manipulated or generated by artificial intelligence"; one Oklahoma ad said "political satire. AI-generated images do not depict real events"; and a North Carolina state Senate ad said "you guessed it! These funny video clips really did use AI generation."

The research found that state laws requiring disclaimers on AI ads had almost no relationship to whether disclaimers actually appeared: in states without a relevant law, 32% of ads carried a disclaimer, versus 29% in states with one.

This comparison is not perfect. Some ads in the database were not covered by the laws of their states; for example, Colorado's law applies only to candidate deepfakes, and Louisiana's AI law was enacted after the ads aired. Therefore, the research did not determine why there was almost no relationship between state law and actual disclosure rates.

The Minnesota case illustrates the ambiguity at the enforcement level. The state's law generally prohibits deepfakes in political advertising, but an ad using a synthesized video of Democratic Senate candidate Peggy Flanagan still aired in May 2026. Whether it was illegal is uncertain, because the law requires media to be "realistic enough that a reasonable person would believe it depicted speech or conduct that a person did not actually make or engage in."

The policy questions left by the research center on three points: how enforceable existing laws are; which types of disclosure actually help voters; and whether the tracking data covers all AI political ads, which cannot be confirmed.

What can be confirmed is that in the 2026 cycle, AI political ads have reached scale, while disclosure has not kept up: roughly $80 million, nearly 170 ads, and only 31% of ads and 22% of spending carrying an AI note.