AI-generated political advertising has moved from novelty to major line item in the 2026 U.S. midterms. The Wesleyan Media Project has identified 220 ads made at least partly with AI tools, backed by more than $137 million in spending—yet only 27% carried any disclaimer that the technology was used. The findings, released on October 15, underscore how quickly synthetic media has become a routine campaign instrument while transparency lags far behind.
- $137 million and counting: 220 AI-assisted ads have run in 34 states, up from under $50,000 in the entire 2024 cycle.
- Disclosure is the exception: Only 27% of ads included an AI warning, down from 31% earlier in October.
- Laws are not closing the gap: States with AI ad rules showed 29% disclosure versus 32% in states without them.
A Partisan Spending Gap and a Blurred Line
The spending is heavily concentrated on one side. Republican candidates and groups sponsored 81% of the identified ads and 88% of the spending, while Democrats accounted for 12%; independent groups and single-issue causes made up the rest. The ads have run in 34 states. Nearly half feature deepfake depictions of real politicians, often placing local candidates alongside national Democratic figures such as Hillary Clinton, Nancy Pelosi and Alexandria Ocasio-Cortez.
Some Democrats are making disclosure part of their message: a Senate ad from Texas Democrat James Talarico closes with “Made by humans. Not AI.”
“AI is making its way into the 2026 midterms in a big way this year,” said Erika Franklin Fowler, co-director of the Wesleyan Media Project. “Much of the AI use is not disclosed through any type of disclaimer.”
The line between deception and ordinary production is not always clear. AI use ranges from hyperrealistic deepfakes to subtle edits and stock-footage generation. Not every undisclosed ad is meant to mislead, but voters have no reliable way to tell a synthetic performance from a conventional spot.
The researchers caution that the total is likely an undercount, based only on ads where AI use could be confirmed through a disclaimer, internal review, press reports or data from ad-tracking firm AdImpact. The 27% figure refers to ad count, not spending. Earlier in October, roughly $80 million across 170 ads came with a 31% disclosure rate.
Why State Laws Have Not Delivered Transparency
The report’s most striking finding concerns regulation. In states without a law governing AI in election ads, 32% of AI-assisted ads included a disclaimer; in states with such a law, 29% did. The difference is small enough that the researchers conclude legislation has little relationship with actual disclosure.
The patchwork helps explain why. Colorado’s statute covers only candidate deepfakes; Minnesota bans political deepfakes more broadly; Louisiana passed its law after an ad had already aired. Roughly 30 states have some election-deepfake rule, but not all require a disclaimer. Federally, no dedicated framework exists.
“Congress hasn’t done anything yet, and neither have the regulators,” said Travis Ridout, the project’s other co-director.
Platform policies are stricter on paper. Google requires electoral advertisers to flag realistic synthetic content and adds automatic notices in some formats, though minor edits such as cropping or color correction are excluded. Meta has required since 2024 that political ads disclose AI when they show a real person saying or doing something they did not, or depict realistic events that did not occur.
Those rules do not cover traditional television, and the Wesleyan report does not break down AI ads by platform, leaving unclear how many undisclosed spots ran on TV versus social media. Enforcement is uneven: Meta’s Oversight Board recently called its AI rules “fundamentally inadequate” while ordering two deepfakes removed, and Google only presented its first high-profile political deepfake detected with SynthID in July.
Labels Alone Are a Weak Corrective
Even well-enforced labels may not achieve much. An October 2024 NYU study found that labels can hurt the candidate sponsoring an ad and provoke backlash against attack-ad sponsors, while up to 37% of participants did not remember seeing the label. Scott Babwah Brennen, director of NYU’s Center on Technology Policy, has argued that labels suit narrowly defined deceptive content—a last resort, not a full solution.
Europe is testing a different approach. The European Union activated its obligation to label AI-generated content that appears authentic on August 2 under the AI Act, replacing a state-by-state mosaic with a common rule.
In the U.S., the midterms take place on November 3, with all 435 House seats and 35 Senate seats on the ballot. With roughly two and a half weeks of campaigning left, the disclosure gap is unlikely to close on its own.
The deeper problem is incentives, not technology. AI makes political advertising cheaper and faster, and skipping a disclaimer usually carries no consequence—even where a law exists. As long as penalties remain theoretical and disclosure is a political liability, nearly three in four campaigns will keep quiet. A rule that does not change behavior is little more than a statement of intent.