Science1 publisher2 min readPublished
Wesleyan researchers trace $80 million in midterm ad spending to about 170 AI-made spots
Wesleyan Media Project researchers tracked about $80 million spent on almost 170 AI-made midterm ads, 80% of them from Republican sponsors. Only 31% of those ads told viewers AI was involved, and that rate comes from a sample coders had to find largely by eye.
The Scientist · Science desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- Deepfakes in the tracked ads depict Donald Trump, Nancy Pelosi, Barack Obama, Kamala Harris and Alexandria Ocasio-Cortez, who appears in at least five ads and is a favourite of Republican advertisers.
- Other ads used AI to generate crowds or constituents, and one put a fake Anthony Fauci at a state fair carrying a huge syringe.
- Republican candidates and allied groups, including super PACs and 501(c) organisations, accounted for 83% of the tracked spending.
- AI disclosure rules depend on a patchwork of state laws, so many of the tracked ads were never required to say they used AI.
- Disclaimer wording ranged from a formal Georgia notice to a North Carolina state Senate ad that said, "You guessed it! AI was definitely used to generate these silly video clips."
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- exposure Spending buys airtime and feed placement, so the unlabelled AI ads, with more money behind each one, are the ones voters were most likely to see.
- constraint Until labels are mandatory, counts depend on human spotting, so the $80 million is a floor and the 31% disclosure rate is likelier to fall than rise as missed ads turn up.
- decision States writing disclosure rules have to decide how specific a label must be, because an ad like Chip Keating's in Oklahoma can include an AI disclaimer without saying what AI did.
The sample is the ads someone noticed. Many states do not require an AI label, so the Wesleyan team relied on media coverage and trained student coders to flag ads that looked AI-generated [2]. Spending figures came from AdImpact, a firm that tracks political ad buys [4]. A deepfake of Liz Cheney, Mitt Romney and Mike Pence carrying pitchforks on the White House lawn is easy to flag [6]. An ad in which AI only enhanced the visuals may look no different from one made before AI, according to the researchers, who have studied political advertising since 2010 [8] [3].
The 31% disclosure rate is a share of the ads the team managed to find [13], and a labelled ad is easier to find than an unlabelled one. Within that set, 69% of ads did not disclose AI [15].
The spending tilts further toward the unlabelled ads. Disclosed ads were 31% of the count but 22% of the money [13]. That leaves about $62 million of the roughly $80 million in ads without an AI notice [16]. Per ad, an undisclosed spot drew about 1.6 times the spending of a disclosed one [17].
Republican candidates and pro-Republican groups were behind 80% of the tracked ads, and the researchers describe Republicans as much more likely than Democrats to use AI [10]. Those shares come to about 136 ads and roughly $66 million [18]. The tally does not show how the 80% compares with each party's share of all political ads this cycle. That baseline would turn a share of AI ads into a usage rate.
On cause, the researchers hold back. "We can only speculate as to why Republicans dominate the use of AI in political advertising in 2026," they wrote [19]. Their speculation is about party philosophy. Democrats tend to favour campaign spending limits and required disclosures, while Republicans lean toward a free-market approach [11]. In 2022, the Senate votes to move the DISCLOSE Act forward came only from Democrats and two independents; the bill would have made super PACs, labor unions and corporations disclose more [11]. The vote fits the pattern, though the tracking data cannot test whether party philosophy causes it.
Voters, in polling the researchers cite, lean toward rules: 78% of registered voters favour a ban on AI content that makes deceptive claims about candidates [12].
What to watch
- Whether the Wesleyan tally of AI ads and spending grows as coders and AdImpact data pick up spots that were missed, and whether the 31% disclosure share holds.
- Whether states without AI disclosure rules adopt them, and whether new rules specify what a label must say about how AI was used.
- Whether Democratic sponsors increase their use of AI-made ads as the campaign closes.