Campaigns and outside groups have spent more than $137 million on AI political ads in the 2026 midterms, the Wesleyan Media Project reported on October 15. It counted 220 ads made or enhanced with AI, in 34 states. Only 27% of them carry a disclaimer saying so.
That gap is the problem for voters a week before Election Day. Some of these ads show real candidates saying or doing things they never did. Where labels do exist, they can be tiny or brief, as in the Texas ads below. Starting today, Meta also stops accepting new political ads on Facebook and Instagram until the polls close.
What the $137 million actually counts
The Wesleyan Media Project tracks political advertising with spending estimates from the ad-tracking firm AdImpact. It counts an ad as AI-made if the ad discloses it, if news outlets or AdImpact reported it, or if its coders flagged it and the team confirmed it. When coders can’t tell, the ad stays out. The project calls its total “almost certainly an undercount.”
Even so, the jump is steep. The project found less than $50,000 in AI ad spending in the whole 2024 cycle. Its September 4 report counted 164 ads and just under $80 million. Six weeks later, the count had passed 220 ads and $137 million.
“AI is making its way into the 2026 midterms in a big way this year,” said Erika Franklin Fowler, the project’s co-director.
Republicans account for most of it. Republican candidates and groups sponsored 81% of the AI ads and 88% of the spending, by the project’s count.
Democrats sponsored 12% of the ads, and independent candidates and other groups the rest. The $137 million is the airtime bought for ads that used AI somewhere. That ranges from a generated background to a fake candidate, so it doesn’t measure deepfakes alone.
Deepfakes of real candidates, from Texas to New York
Most of the widely reported examples came from Republicans, sometimes aimed at fellow Republicans.
In Texas, Republican Senate nominee Ken Paxton’s campaign released a 30-second ad on September 8. It shows an AI-generated James Talarico, his Democratic opponent, reading from a fictional “King James (Talarico) Bible.” FactCheck.org found that the ad strings together real remarks from different years, strips their context and changes some wording. It ran on cable, streaming and digital platforms, per AdImpact data the site cited. A small-print AI notice appears about 20 seconds in.
A Talarico spokesperson said Paxton and his backers “literally put words in James’ mouth with AI deepfake videos.” The Paxton campaign did not answer FactCheck.org’s questions.
It wasn’t the first.
In March, the National Republican Senatorial Committee posted an 85-second AI “Talarico” reading his old tweets, with a faint “AI GENERATED” label in one corner, CNN reported. During the Republican primary, Paxton’s campaign posted a fake John Cornyn dancing with Democratic Rep. Jasmine Crockett. Cornyn’s campaign posted fake clips of rival Wesley Hunt holding a dog, with no label at all.
In Georgia, the Senate Leadership Fund, a Republican super PAC, released an AI video showing Democratic Sen. Jon Ossoff in a salon chair, WRDW reported. Ossoff called deepfake ads “a slippery slope toward the destruction of truth in political debate.”
In New York, Republican governor nominee Bruce Blakeman has run AI ads using a voice and likeness that appear to be Gov. Kathy Hochul’s, ABC7 reported. Blakeman calls them satire. Hochul’s campaign calls them “straight up lying.”
On the Democratic side, the reporting we reviewed found no comparable deepfake of a Republican opponent in a paid ad. California Gov. Gavin Newsom has posted openly satirical AI videos of President Trump, but he isn’t on this year’s ballot.
A patchwork of state laws, and Georgia has none
There is no federal requirement to label AI in political ads, the Wesleyan project notes. So the rules depend on the state.
Public Citizen’s state deepfake legislation tracker, updated October 21, lists enacted laws in 33 states. Two of those no longer apply: a court struck down Hawaii’s, and Massachusetts’ expired in 2025. Most of these laws require a disclosure rather than banning the content. Georgia’s bill failed in January, so the Ossoff ad needed no label under state law.
Texas is one of the stricter states on paper. Its Election Code Section 255.004 makes it a Class A misdemeanor to create a “deep fake video” meant to injure a candidate and publish it within 30 days of an election. For November 3, that window opened on October 4, weeks after the Paxton ad first ran.
New York requires labels on AI depictions that misrepresent a candidate. State Democratic Party chair Jay Jacobs filed a complaint with the state Board of Elections in August over Blakeman’s videos, Spectrum News reported. Blakeman’s campaign says the law exempts satire.
What the FEC and FCC have done
Federal regulators have mostly applied old rules to the new tools. In September 2024, the Federal Election Commission declined to write AI-specific rules. It instead issued an interpretive rule saying its ban on fraudulent misrepresentation is “technology neutral,” so it covers AI. That ban reaches candidates and their agents, not most outside groups.
The Federal Communications Commission ruled in February 2024 that AI-generated voices in robocalls count as “artificial” under federal robocall law. In July 2024 it proposed on-air AI disclosures for TV and radio political ads. No such requirement covers this year’s ads.
How to spot an AI political ad
The platforms have their own rules, and they are where most labels show up. Here is where to look.
- Google and YouTube: Election advertisers must tick an “altered or synthetic content” box for ads that make a real person appear to say or do something they didn’t. In YouTube feeds, Shorts and in-stream ads, Google then adds the disclosure itself. In other formats the advertiser must add a clear one, such as “This video content was synthetically generated.”
- Facebook and Instagram: Advertisers must disclose photorealistic AI images, video or audio in political ads “in certain cases,” and Meta adds that information to the ad and to its Ad Library. Since June, Meta’s “About this ad” tool also flags ads it detects came from third-party AI tools.
- TV and streaming: Look for small print in a corner or near the end. In the Paxton ad it appeared about 20 seconds into a 30-second spot.
- Anywhere: Find the “Paid for by” line, then look the sponsor up in Meta’s Ad Library or Google’s Ads Transparency Center. If a clip shows a candidate saying something startling, search for the original speech or post before you share it.
Many voters say AI ads have fooled them. In a Scripps News/Talker Research poll of 1,000 registered voters, more than four in ten said they had seen something in a political ad they thought was real, then learned it was AI. The online survey ran September 30 to October 1. More than three-quarters of Democrats and Republicans alike said AI in politics needs rules.
Meta’s ad freeze starts today
Meta won’t accept new US ads about social issues, elections or politics from 12:01 a.m. Pacific time on October 27 through November 3, according to its developer notice as reported by PPC Land. Ads that ran before the cutoff can keep running. Meta says the reason is that late claims leave too little time to contest them.
Our guide to the 2026 midterms covers what’s at stake, and our rundown of Senate races to watch includes Texas and Georgia.
The freeze applies only to Meta’s own apps. It doesn’t reach the broadcast, cable and streaming slots where campaigns spent much of the $137 million, and those ads can keep running until polls close on November 3.