Three of four AI political ads won't say what they are. Midterms are in two weeks.

If you're in a competitive district and you've been watching TV or scrolling Facebook over the last few weeks, you've probably seen political ads. Some of them were made, at least in part, by AI. Most of those won't say so.

The Wesleyan Media Project, an academic group that has tracked political advertising for more than two decades, has identified at least 164 AI-generated or AI-enhanced ads in the 2026 midterm cycle, representing more than $137 million in spending. Erika Franklin Fowler, one of the project's co-directors, told reporters that in the entire 2024 election cycle the project found less than $50,000 in spending on AI-generated ads.

This cycle: $137 million. That is not a trend. That is a step-change.

What "AI-generated political ad" actually means

AI-generated ad doesn't mean the whole ad was made by a robot. It means the ad used AI to create or substantially change a voice, a face, or a scene. Could be a synthetic voice reading attack lines that sound like a real candidate. Could be digitally altered footage placing someone at an event they didn't attend. Could be a fully fictional crowd scene generated from scratch.

The tells used to be obvious. Teeth looked wrong. Background loops stuttered. Lighting in AI-generated scenes was too uniform — no shadows, no grain, no real light source. These artifacts are getting rarer. The gap between real and fake is closing faster than the laws designed to address it, which is the whole problem.

Source spread

• Wesleyan Media Project — [academic]. Two decades of political ad tracking. The primary source behind these figures; co-directors Erika Franklin Fowler and Travis Ridout are cited throughout.
• Gizmodo — $137M, 27% disclosure — [skeptic]. Reports the most recent Wesleyan figures; includes the counterintuitive state-law finding.
• CNN — AI imagery in campaign ads — [skeptic]. Separate methodology, different cutoff date. Identified $16M with AI disclosure labels and at least $38M appearing AI-generated without one.
• Al Jazeera — Deepfakes in midterms — [skeptic]. Global framing; names specific examples across multiple states.

What's real

The scale jump is documented. Less than $50,000 in 2024. More than $137 million in 2026. Whatever you think about AI in politics, those numbers belong in the same sentence exactly once.

State laws are real. Thirty-one states have passed legislation requiring campaigns to disclose AI use in political advertising. That is more than half the country, and it happened in under two years.

Some campaigns are complying. Twenty-seven percent is not zero. Roughly one in four AI-generated ads carried a disclosure label. That means at least some campaigns made a deliberate choice to disclose, which suggests the norm can function.

What deserves a side-eye

Here is the counterintuitive finding from the Wesleyan data, and it deserves more attention than it's getting.

Travis Ridout, the project's co-director, found that in states that have no disclosure law, 32% of AI-generated ads voluntarily included a disclaimer anyway. In states that do have a disclosure law, the figure was 29%. Regulated states are complying less than unregulated ones.

Ridout was careful to say the finding has caveats. One possible explanation: campaigns in regulated states have legal counsel who knows exactly where the line is and are staying on the right side of a narrow definition. Another: enforcement is close to zero, so the law produces no real pressure. Either way, the policy isn't achieving its stated goal.

There is also a partisan breakdown worth noting: 80% of the identified AI ads were sponsored by Republican candidates or groups, accounting for 88% of the total spending. A New York court complaint over undisclosed AI in Bruce Blakeman's gubernatorial ads is one of the first formal enforcement attempts anywhere. That's a limited track record for a $137 million phenomenon.

What to do about it

• Look for the label. In states with disclosure laws, AI-generated political ads are supposed to carry text saying something like "Made with AI" or "AI-generated imagery." It usually appears briefly, in small font, at the bottom of the screen. Make a habit of looking in the last two seconds of any political ad.
• Trust the tells. Watch the mouth and the background edges. AI video generation still struggles with teeth, consistent lip sync, and background lighting. When something looks slightly off in a political ad, that instinct is worth examining.
• Look up who paid for it before you share it. The "Paid for by" line is legally required on licensed political ads. If you're sharing an ad that makes you angry and you can't identify who paid for it, slow down.
• Cross-check anything extraordinary. AI voices can say anything. An ad where a candidate appears to say something surprising or completely out of character is worth ten seconds of searching before you react.
• The Wesleyan tracker at mediaproject.wesleyan.edu publishes data by race and state. If you want to know what's running in your specific district, that's the most credible public resource available.

Further reading

• Wesleyan Media Project — the academic tracker behind the numbers
• Gizmodo — $137M, 27% disclosure, full Wesleyan findings
• CNN — AI imagery in campaign ads analysis, Jan–Sept 2026
• Al Jazeera — AI deepfakes in midterm campaigns
• Fast Company — midterms AI ad spending analysis