From under $50,000 to over $137 million
The Wesleyan Media Project, which tracks US political advertising with AdImpact data, now counts 220 ads in the 2026 midterm cycle that were generated at least partly with AI tools. More than $137 million has been spent promoting those ads to voters. Across the full 2024 cycle, the same tracking operation identified less than $50,000 in spending on such ads.
Read the pair of numbers for what they are: identified spending under one tracking methodology, in two very different cycles. Detection and classification have changed between them, so the contrast cannot by itself establish a precise growth rate for AI use in campaigns. What it does document is a dramatic change in what researchers could find. Two cycles ago, an AI-made ad was an experiment a campaign had to defend. This cycle, it is a line item.
The category is broad, and the breadth matters. New York Times reporting on the same dataset, as carried by secondary coverage — we have not read the Times piece directly — found that nearly half of the tracked ads include fabricated depictions of real politicians. Others use AI for smaller jobs: a generated crowd, a cleaned-up image, a synthetic voice on a routine spot. Not every ad in the count is a deepfake, and reading the total as if it were one overstates the deception while understating the shift. The controversial examples draw the coverage; the routine production uses are the ones that may prove harder to reverse.
The label that usually is not there
Of the 220 ads Wesleyan identified, 27% carry a disclaimer that AI was involved. The rest carry none. Campaigns & Elections, reporting on the same release, notes that 30 states have some form of AI-disclosure requirement for political advertising. That fact needs handling with care. The requirements vary in scope from state to state, and some cover only materially deceptive synthetic content rather than any AI assistance at all. A state having a law on the books does not mean every AI-assisted ad in that state was required to carry a label, and the 27% figure measures how often labels appeared in Wesleyan's identified sample — not how many ads violated a requirement. Federal rules do not fill the gaps between the state regimes.
The researchers are equally careful about the limits of their own count. Ads enter the database when trained coders flag them, when a disclaimer admits the use, when reporting surfaces them, or when AdImpact tags them. When the team cannot confidently tell whether an ad was AI-generated or AI-enhanced, the ad stays out. The published figures are therefore a documented minimum under Wesleyan's methodology: a count of what could be proved, not a comprehensive estimate of AI advertising across the election.
One side is doing most of it
The identified use is lopsided. Republican-affiliated sponsors account for most of the AI-assisted advertising Wesleyan could identify: 81% of the identified ads and 88% of the spending behind them. Democratic sponsors account for 12% of the ads, with the remainder from independent or issue sponsors. That is a larger documented footprint in the category. What the spending figures do not establish is whether these ads are more persuasive, or more effective at moving voters, than conventional ones.
Wesleyan's co-directors have suggested, in reporting around the release, that campaigns arguing for tighter AI rules may carry more reputational risk when they use the tools themselves. Whatever the mix of reasons, the pattern fits a wider sequence: production tools can become commercially useful well before disclosure standards catch up with them. The midterms are simply a particularly visible example — $137 million in identified spending, most of the tracked ads unlabelled, and a research count that everyone involved agrees runs low.
Sources
- [1] Wesleyan Media Project, release of 15 October 2026, updated 16 October (primary)Read source
- [2] Campaigns & Elections, "Political Advertisers Have Spent At Least $137 Million on AI-Generated Ads", 16 October 2026Read source
- [3] New York Times, "$137 Million in A.I.-Generated Ads Flood the Midterms", 16 October 2026 (as carried by secondary coverage; not read directly)Read source
- [4] Gizmodo, "$137 Million Has Been Spent on AI-Generated Ads Ahead of Midterms", 16 October 2026Read source