Thomas Massie, a seven-term Republican congressman from Kentucky, lost his primary on May 19. The race cost $32.6 million, making it the most expensive congressional primary in U.S. history. Among the weapons deployed against him: an AI-generated video showing him checking into a hotel room with Representatives Alexandria Ocasio-Cortez and Ilhan Omar, with their hands clasped. The narrator called it "worse than adultery." A disclaimer at the bottom of the screen, in small text, read: "This satirical ad was created with artificial intelligence."
I watched the ad. It looks ridiculous. The movements are slightly too smooth, the lighting is off, and the scenario is so cartoonish that I genuinely could not tell at first whether it was meant to be taken seriously. But I'm 17, and I've spent the last four years watching AI-generated content get better in real time. I know what the seams look like. Massie, in his post-loss interview on Meet the Press, was blunt about who the ad actually reached: "It was actually very effective on the boomers."
He mentioned something else that stuck with me. His daughter, who lives outside his district, overheard someone in the voting booth next to her trying to find his opponent's name on the ballot. The voter didn't even realize Massie wasn't their congressman.
One voter was fooled by AI into believing something that never happened. Another didn't know who their own representative was. Different failures. Same election.
This is not a future problem. NBC News has tracked at least 15 campaign ads featuring AI-generated content since November 2025. They cross party lines. In Massachusetts, Republican gubernatorial candidate Brian Shortsleeve's campaign posted a radio ad that sounded exactly like Democratic Governor Maura Healey listing her own policy failures. It was entirely AI-generated. Her voice was cloned. The things she appeared to be saying, she never said. The campaign's defense was that the caption described it as what her ads would sound like "if she was honest," as though the joke somehow canceled out the deception. No explicit AI disclaimer appeared in the ad itself. In New York, Andrew Cuomo's campaign ran AI-generated attack ads against his mayoral opponent, one of which was pulled and blamed on a "junior staffer" after being widely condemned as racist. In Massie's own race, both sides ran deepfakes. The pro-Massie PAC produced its own AI video of Massie's opponent fleeing Donald Trump on a fictional battlefield. Kentucky passed a synthetic-media election law in 2025. The MAGA KY spot cited a satire exception. The pro-Massie ad didn't bother with a disclaimer at all. Roughly 30 states now have some version of an AI disclosure law. None of them prevented any of these ads from running.
The production costs have collapsed. David Martin of the advertising firm Adwave told the Washington Times that what used to require $10,000 to $15,000 and a full production team can now be done for about $50, overnight.
Most of the conversation about this focuses on regulation and labeling. I think that misses the deeper problem. A deepfake doesn't need to be technically undetectable to work. It just needs to reach someone who has no baseline for questioning it.
Most Americans can't name their own House member. A Haven Insights poll put the number at 37%. Think about what that means for a deepfake ad. If you've never heard of the person being depicted, you have no reference point. The fake version and the real version carry the same weight, because you never knew the real version to begin with. Six states have completed mid-decade redistricting since 2025, which resets whatever name recognition existed. New lines, new representatives, new confusion. A synthetic video drops into that vacuum, and there's nothing to push back against it.
For many older voters, the problem isn't even falling for a specific trick. It's that the category doesn't exist for them. Nearly 40% of adults aged 65 and older have never heard of the term "deepfake," according to a 2025 iProov survey. These are also the voters who show up the most reliably and get the bulk of their political information from broadcast television. The FCC does not currently require any disclosure when an AI-generated ad airs on TV.
I'm a high school senior at Thomas Jefferson High School for Science and Technology, where I've taken two semesters of AI coursework. But the reason I can usually spot a synthetic video has almost nothing to do with what I learned in class. It's because I've been immersed in this stuff since middle school. My friends and I constantly share AI-generated memes. We've all seen enough face swaps, voice clones, and generated images to develop a kind of instinct for when something is off. It's like how you can tell a counterfeit bill feels wrong before you even think to check the watermark.
I should be careful about how far I push that, though. The same iProov study found that only 0.1% of people across all age groups correctly identified every deepfake they were shown. Young adults were actually the most overconfident about their detection skills. So my generation is not deepfake-proof. We can't reliably tell real from fake any better than anyone else in a controlled test. What's different is that we assume political content might be synthetic in the first place. We carry that suspicion by default, and older voters mostly don't. That default suspicion is probably more protective than any detection skill.
There are millions of Americans who have voted in every election for decades and have never heard the word "deepfake." I learned what one was in eighth grade.
I keep coming back to the Cuomo mayoral race and the Massie primary because they seem to show this playing out in opposite directions, though I'm not sure the comparison is as clean as I'd like. In New York, younger voters identified Cuomo's AI ads immediately and mocked them as "boomer slop." The synthetic content became a liability for his campaign. CIRCLE at Tufts found 75% of NYC youth voted for Mamdani versus 19% for Cuomo, though Cuomo was already unpopular with young voters for a dozen reasons that had nothing to do with AI. In Kentucky, the electorate in Massie's district skews well over 60, and Massie himself predicted what would happen: "They're hoping the older generation won't realize it's an AI-generated lie." Two races don't prove a pattern. But the underlying logic is hard to dismiss: deepfakes land harder when the audience has no familiarity with the form.
The generation that has that familiarity votes so little. In the 2022 midterms, 27.6% of 18-to-24-year-olds turned out, compared to about 68% of voters aged 65 to 74. There are real structural reasons for this gap. Young people move more often, which creates registration friction. They have less flexible schedules. They don't have the decades of built-up civic habit or the direct financial stake in programs like Social Security that keeps older voters engaged.
Massie probably didn't lose because of a deepfake. Trump's endorsement and tens of millions in outside spending likely mattered more. Whether the AI video moved enough people to change the result is honestly unknowable.
But it didn't need to determine the outcome. It just needed to work on the margin, in a primary where the electorate skewed toward people who had never encountered the form and were the most likely to actually cast a ballot. The next synthetic ad targeting a congressional district will cost $50 to produce. The voters most likely to see it on broadcast TV will have no frame of reference to question it.
Recognizing synthetic media is becoming as basic a civic skill as reading a ballot. Right now, the people who have that skill are mostly sitting out the elections where it matters most.
Maneesh Vaddi is a 17-year-old rising senior at Thomas Jefferson High School for Science and Technology. His research on gerrymandering and voter turnout has been published in the GMU Journal of Student-Scientist Research. He delivered a pledge at the United Nations ECOSOC Chamber for the Transforming Global Education Summit and spoke at the Right Here, Right Now Global Youth Climate Summit, hosted by Oxford Saïd Business School, the Smith School of Enterprise and Environment, and UN Human Rights.
