$137 million in AI political ads, and most of them never tell you

Kathy Hochul and Zohran Mamdani never filmed an ad together for a "political delusion pill." There is a video of them doing it anyway. It was made by the campaign of Bruce Blakeman, the Republican running for governor of New York, and it carried no AI disclaimer.

That ad is one of more than 220 AI-generated political videos the Wesleyan Media Project has counted this midterm season. On October 16, The New York Times reported the total spending behind them: more than $137 million, according to Wesleyan's tally. In the 2024 cycle, Wesleyan co-director Erika Franklin Fowler said in a news release quoted by Gizmodo, "we identified less than $50,000 in spending on AI-generated ads." That is roughly a 2,700-fold jump in two years.

Most of them don't tell you

The disclosure numbers are bad however you slice them, and outlets slice them slightly differently. Gizmodo, working from Wesleyan's release, puts the share of ads with any AI disclaimer at 27 percent. The Times version has it at 31 percent. Either way, roughly seven in ten AI political ads this year say nothing about it.

"Much of the AI use is not disclosed through any type of disclaimer," Fowler said in the release. And the researchers think they are undercounting. Only about 30 states require AI disclosures on political ads, and in practice a disclosure is often the only way to know for sure. As Fowler told the Times: "Some of the ways in which A.I. is used, honestly, if there wasn't a disclaimer on the ad, it would be hard to know that A.I. was there."

The figures have moved fast. In early October, Wesleyan's researchers were describing about $80 million on almost 170 ads through early September, in a piece for The Conversation republished by Fortune. Now the count is past 220 and the money has nearly doubled. Some of that is new ads. Some is broader criteria for what counts as AI-made, which is worth keeping in mind before anyone quotes the $137 million figure as a precise measurement.

A very one-sided arms race

Republican candidates and groups made 81 percent of the identified ads and paid for 88 percent of the spending, according to Gizmodo's account of the Wesleyan data. Democrats sponsored 12 percent.

Travis Ridout, Wesleyan's other co-director, offered the Times a theory: "Democratic candidates and groups face more reputational risk when using a tool that they are arguing should be regulated." Democratic voters are more skeptical of the technology and more eager for rules on it, so a Democrat caught running a deepfake pays a bigger price with their own side.

Some Democrats have turned that into a selling point. James Talarico, the party's Senate candidate in Texas, ran an ad that ends with the words "Made by humans. Not AI."

The ads Wesleyan flagged are mostly not subtle. Nearly half put fake versions of real politicians on screen, Gizmodo reports, often tying a local candidate to a national Democrat like Hillary Clinton, Nancy Pelosi or Alexandria Ocasio-Cortez. In Oklahoma, a group called Make Oklahoma Great Again ran an ad showing the Democratic candidate for governor embracing Clinton, under the text "Clinton Lover. Trump Hater." It had no AI disclosure, and AdImpact data cited by the Times shows more than $200,000 spent putting it on Oklahoma television. In Michigan, an AI version of Senate candidate Abdul El-Sayed laughed at cards reading "Death to America" and "Celebrate 9/11."

Blakeman alone has posted at least 22 AI-generated attack ads on social media since April. The New York State Democratic Party filed a complaint with the state Board of Elections in August. His campaign's position is that satire doesn't need a label.

The cheap version of persuasion

Not every AI ad is a smear. The Times found campaigns using the tools for crowd scenes, recreated childhood settings and animated photos, the kind of thing stock footage used to cover. Republican Chip Keaton's campaign for Oklahoma governor animated an old picture of him as a state trooper so that his head tilts up into a smile.

The money argument is hard to resist. Ad executive Fred Davis told the Times he cut one planned ad from nearly $2 million to about $200,000 by using AI. "It's the year of the A.I. election," he said.

Cheap is the whole point, and cheap means more of it. A label only helps if it's there, and most of the time it isn't. Security technologist Bruce Schneier boiled it down further: "The problem is deception."

There is no federal rule on AI in political ads, and the midterms are on November 3. Between now and then, the cheapest way to put words in an opponent's mouth is to rent a model, generate the mouth, and let a state election board sort it out in the spring.

Sources: The New York Times (via DNYUZ), Gizmodo, Fortune / The Conversation.