AI Midterm Ads Passed $137 Million, and Most Never Said So

Just over two weeks before the US midterms, the clearest measure yet of AI's role in campaign advertising has arrived, and the headline number is large. According to a review by the Wesleyan Media Project and the ad-tracking firm AdImpact, reported by The New York Times on October 16, campaigns and political action committees have spent more than $137 million on ads that were generated or enhanced with AI in the 2026 cycle.

The more important number is smaller. In the broader review, about 70% of those ads told viewers nothing about it.

Three ways to count

The $137 million figure uses the broadest definition: any ad where AI shaped the imagery or audio. Researchers offered narrower counts too, and the spread is instructive.

A tighter Wesleyan tally covering at least 164 unique ads came to just under $80 million. AdImpact's most conservative count, limited to ads carrying an identifiable AI disclosure, found slightly more than $13 million. In the narrower set, only 31% of ads disclosed AI use.

Wesleyan says all of its counts probably understate the total. Ads that run in states with no disclosure rules are harder to detect, because nothing in them flags the technique. The ads most likely to be missed are the ones that never admit what they are.

Most of it is mundane

It helps to separate two very different uses.

Much of the AI in political advertising is production work that would have been done by hand a few years ago: animating a still photograph, generating a crowd scene, recreating a candidate's childhood neighbourhood, replacing stock footage. Fred N. Davis III, a Hollywood ad executive who works on campaigns, told the Times that a commercial that once cost close to $2 million can now be made for about $200,000 with AI. That is a genuine shift in who can afford broadcast-quality spots.

The other use is the one the disclosure fight is about: synthetic depictions of real politicians saying or doing things that never happened.

The ads that made the news

The Times and other outlets catalogued several examples.

A video from Bruce Blakeman's campaign showed synthetic versions of New York City Mayor Zohran Mamdani and Governor Kathy Hochul promoting a fictional "political delusion pill," with no AI disclaimer. A group called Make Oklahoma Great Again ran an ad showing a synthetic Democratic gubernatorial candidate embracing Hillary Clinton; AdImpact tracked more than $200,000 in Oklahoma TV buys for that ad alone. Other ads showed a fabricated Abdul El-Sayed laughing at cards bearing extremist messages and a fabricated James Fishback appearance alongside Representative Ilhan Omar and Iranian leaders. Per the Times, neither of those events occurred.

In the broader dataset, more than 80% of the reviewed ads came from Republican campaigns or Republican-aligned PACs. The narrower Wesleyan accounting put roughly 83% of tracked spending with Republican-aligned groups. The data shows who is using the technique most heavily this cycle; it does not show that the other side has sworn it off.

A patchwork of rules

There is no federal disclosure requirement for AI in political ads. According to CNN, roughly half of US states require disclosure of AI-generated or manipulated material in political advertising, and the rules vary widely.

Michigan's is among the most specific: a three-second voiceover acknowledging AI use plus a four-second written disclosure. New York has a law requiring "materially deceptive" political communications to carry an AI label. The New York State Democratic Party filed a complaint with the state Board of Elections over the Blakeman ads in August; the campaign called the ad satire and said it needed no disclaimer. Some state deepfake laws have been blocked in court on free-speech grounds.

That leaves an odd map. The same synthetic clip can require a label in one state and none across the border, while streaming and social platforms deliver it to both.

The American Association of Political Consultants, the industry's own trade group, now supports federal rules that would preempt the state patchwork, require labels for deceptive AI content and preserve broad protections for political speech. When the people making the ads ask for a single rulebook, the current one is not working.

Why labels matter more than the tools

Wesleyan's co-directors framed the problem in practical terms. Erika Franklin Fowler noted that some AI uses are hard for viewers to spot without a label. Travis Ridout warned, in comments to CNN, that voters can be shown politicians appearing to say or do things they never did.

The technology itself is not the scandal. A cheaper crowd shot is not a threat to democracy. The problem is the combination of photorealistic fabrication with no obligation to say so, delivered at a scale and price that was impossible in the last midterm cycle.

The takeaway

The $137 million figure will be quoted as proof that AI has taken over political advertising. It is better read as proof that disclosure has not kept up. Most of these ads are legal. Many are harmless. But roughly seven in ten arrived without a label, and the share that did carry one depended on geography.

With barely two weeks left in the cycle, the useful question for voters is not whether an ad used AI. It is whether anyone was required to tell them.